comparative performance of four single extreme outlier ... · researcharticle comparative...
TRANSCRIPT
Research ArticleComparative Performance of Four Single Extreme OutlierDiscordancy Tests from Monte Carlo Simulations
Surendra P Verma1 Lorena Diacuteaz-Gonzaacutelez2 Mauricio Rosales-Rivera3
and Alfredo Quiroz-Ruiz4
1 Departamento de Sistemas Energeticos Instituto de Energıas Renovables Universidad Nacional Autonoma de Mexico62580 Temixco MOR Mexico
2 Facultad de Ciencias Universidad Autonoma del Estado de Morelos 62209 Cuernavaca MOR Mexico3 Posgrado en Ciencias Facultad de Ciencias Universidad Autonoma del Estado de Morelos 62209 Cuernavaca MOR Mexico4Departamento de Computacion Instituto de Energıas Renovables Universidad Nacional Autonoma de Mexico62580 Temixco MOR Mexico
Correspondence should be addressed to Surendra P Verma spvierunammx
Received 18 October 2013 Accepted 5 December 2013 Published 11 March 2014
Academic Editors J Pacheco and Z Wu
Copyright copy 2014 Surendra P Verma et al This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited
Using highly precise and accurateMonte Carlo simulations of 20000000 replications and 102 independent simulation experimentswith extremely low simulation errors and total uncertainties we evaluated the performance of four single outlier discordancytests (Grubbs test N2 Dixon test N8 skewness test N14 and kurtosis test N15) for normal samples of sizes 5 to 20 Statisticalcontaminations of a single observation resulting from parameters called 120575 from plusmn01 up to plusmn20 for modeling the slippage of centraltendency or 120576 from plusmn11 up to plusmn200 for slippage of dispersion as well as no contamination (120575 = 0 and 120576 = plusmn1) were simulatedBecause of the use of precise and accurate random and normally distributed simulated data very large replications and a largenumber of independent experiments this paper presents a novel approach for precise and accurate estimations of power functionsof four popular discordancy tests and therefore should not be considered as a simple simulation exercise unrelated to probabilityand statistics From both criteria of the Power of Test proposed by Hayes and Kinsella and the Test Performance Criterion ofBarnett and Lewis Dixon test N8 performs less well than the other three tests The overall performance of these four tests could besummarized as N2 cong N15 gt N14 gt N8
1 Introduction
As summarized by Barnett and Lewis [1] a large number ofdiscordancy tests are available for determining an outlier asan extreme (ie legitimate) or a discordant (ie contami-nant) observation in normal samples at a given confidenceor significance level These discordancy tests are likely to becharacterized by different power or performance Numerousresearchers [1ndash6] have commented on the properties of thesetests under the slippage of location or central tendency andslippage of scale or dispersion by one or more observationsbut very few studies have been reported on the use ofMonteCarlo simulation for precise and accurate performance
measures of these tests Relatively more recently usingMonteCarlo simulation of 119872 = 100 000 replications or runsHayes and Kinsella [7] evaluated the performance criteriaof two discordancy tests (Grubbs single outlier test N2 andGrubbs multiple outlier test N4k2 the nomenclature is afterBarnett and Lewis [1]) and discussed their spurious andnonspurious components of type II error and power functionHowever four single extreme outlier type discordancy testsalso called two-sided discordancy tests by Barnett and Lewis[1] are available which are Grubbs type N2 Dixon type N8skewness N14 and kurtosis N15 Their relative performancemeasures should be useful for choosing among the differenttests for specific applications
Hindawi Publishing Corporatione Scientific World JournalVolume 2014 Article ID 746451 27 pageshttpdxdoiorg1011552014746451
2 The Scientific World Journal
Monte Carlo simulation methods have been exten-sively used in numerous simulation studies [8ndash18] Some ofthe relatively recent papers are by Efstathiou [12] Gottardo etal [13] Khedhiri andMontasser [14] P A Patel and J S Patel[15] Noughabi and Arghami [16] Krishnamoorthy and Lian[17] and Verma [18] For example Noughabi and Arghami[16] compared seven normality tests (Kolmogorov-SmirnovAnderson-Darling Kuiper Jarque-Bera Cramer von MisesShapiro-Wilk and Vasicek) for sample sizes of 10 20 30and 50 and under different circumstances recommended theuse of Jarque-Bera Anderson-Darling Shapiro-Wilk andVasicek tests
We used Monte Carlo simulations to evaluate compara-tive efficiency of four extreme outlier type discordancy tests(N2 N8 N14 and N15 the nomenclature after Barnett andLewis [1]) for sample sizes of 5 to 20 Our approach tothe statistical problem of test performance is novel becauseinstead of using commercial or freely available software weprogrammed and generated extremely precise and accuraterandom numbers and normally distributed data used verylarge replications of 20000000 performed 102 independentexperiments and reduced the simulation errors to such anextent that the differences in test performance are far greaterthan the total uncertainties expressed as 99 confidenceintervals of the mean This is an approach hitherto practicedby none (see eg [8ndash18]) except by our group [19ndash23]This work therefore supersedes the approximate simulationresults of test performance reported by the statisticians Hayesand Kinsella [7]
2 Discordancy Tests
For a data array 1199091 1199092 1199093 119909
119899minus2 119909119899minus1
119909119899or an ordered
array 119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
of 119899 observationswith mean 119909 and standard deviation 119904 four test statisticswere objectively evaluated in this work For a statisticallycontaminated sample of size of 5 to 20 119899 minus 1 observationsof this data array were obtained from a normal distribution119873(0 1) and the remaining observation was taken from acentral tendency shifted distribution119873(0+120575 1) or dispersionshifted distribution 119873(0 1 times 120576) where the contaminantparameters 120575 for modeling slippage of central tendency and120576 for slippage of dispersion can be either positive or negativeFor an uncontaminated sample the simulations were donefor 120575 = 0 and 120576 = plusmn1 In order to achieve an unbiasedcomparison the application of the tests was always forced tothe upper outlier 119909
(119899)for positive values of 120575 or 120576 and to the
lower outlier 119909(1)
for negative values of 120575 or 120576Thus the first test was the Grubbs test N2 [24] for an
extreme outlier 119909(119899)
or 119909(1) for which the test statistic is as
follows
TN2 =
119909(119899)
minus 119909
119904 119909(119899)
tested if 120575 gt 0 or 120576 gt 1
119909 minus 119909(1)
119904 119909(1)
tested if 120575 lt 0 or 120576 lt 1
Max(119909(119899)
minus 119909
119904119909 minus 119909(1)
119904)
119909(119899)
or 119909(1)
tested if 120575=0 or 120576=1
(1)
The second test was the Dixon test N8 [2] as follows
TN8 =
119909(119899)
minus 119909(119899minus1)
119909(119899)
minus 119909(1)
119909(119899)
tested if 120575 gt 0 or 120576 gt 1
119909(2)
minus 119909(1)
119909(119899)
minus 119909(1)
119909(1)
tested if 120575 lt 0 or 120576 lt 1
Max(119909(119899)
minus 119909(119899minus1)
119909(119899)
minus 119909(1)
119909(2)
minus 119909(1)
119909(119899)
minus 119909(1)
)
119909(119899)
or119909(1)
tested if 120575=0 or 120576=1
(2)
The third test was sample skewness N14 as (note that theabsolute value is used for evaluation)
TN14 =
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(119899)
tested if 120575 gt 0 or 120576 gt 1
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(1)
tested if 120575 lt 0 or 120576 lt 1
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(119899)
tested if11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
gt 0
120575 = 0 or 120576 = 1
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(1)
tested if11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
lt 0
120575 = 0 or 120576 = minus1
(3)
The Scientific World Journal 3
Finally the fourth test was the sample kurtosis N15 asfollows
TN15 =
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(119899)
tested if 120575 gt 0 or 120576 gt 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(1)
tested if 120575 lt 0 or 120576 lt 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(119899)
tested if (119909(119899)
minus 119909) gt (119909 minus 119909(1)
)
120575 = 0 or 120576 = 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(1)
tested if (119909(119899)
minus 119909) lt (119909 minus 119909(1)
)
120575 gt 0 or 120576 = minus1
(4)
All testswere applied at a strict 99confidence level usingthe new precise and accurate critical values (CV
99) simulated
using Monte Carlo procedure by Verma et al [19] for N2N8 and N15 and Verma andQuiroz-Ruiz [20] for N14 whichpermitted an objective comparison of their performance
3 Monte Carlo Simulations
Random numbers 119880(0 1) uniformly distributed in the inter-val (0 1) and normal random variates119873(0 1)were generatedfrom the method summarized by Verma and Quiroz-Ruiz[21] However instead of only 10 series or streams of 119873(0 1)
as done byVerma andQuiroz-Ruiz [21] a total of 102 differentstreams of 119873(0 1) were simulated Similarly the replicationswere much more than those used by Verma and Quiroz-Ruiz[21] for generating precise and accurate critical values
For a data array of size 119899 (119899 minus 1) observations1199091 1199092 1199093 119909
119899minus2 119909119899minus1
were drawn from one stream of119873(0 1) and the contaminant observation (119909
119888) was added
from a different central tendency shifted stream of119873(0+120575 1)
where 120575was varied from+01minus+20 andminus01minusminus20 or a disper-sion shifted distribution119873(0 1 times 120576) where 120576 was varied from+11minus+200 andminus11minusminus200The simulation experimentswerealso carried out for uncontaminated distributions in which(119899 minus 1) observations were taken from one stream of normalrandom variates 119873(0 1) and an additional observation wasincorporated from a different stream with no contaminationthat is 120575 = 0 and 120576 = plusmn1
Now if we were to arrange the complete array from thelowest to the highest observations the ordered array could becalled 119909
(1) 119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
after Barnett andLewis [1] All four tests under evaluation could then beapplied to the resulting data array
If 120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1 (the contaminant 119909119888
present) two possibilities would arise for the ordered array
119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
as follows (Table 1) (i) thecontaminant 119909
119888occupies an inner position in the ordered
array that is 119909119888lt 119909(119899)
if 120575 gt 0 or 120576 gt 1 or 119909119888gt 119909(1)
if 120575 lt 0 or120576 lt minus1 this array is called a119862 type event and the contaminant119909119888was not used in the test statistic and (ii) the contaminant
119909119888occupies the extreme position that is 119909
119888= 119909(119899)
if 120575 gt 0 or120576 gt 1 or 119909
119888= 119909(1)
if 120575 lt 0 or 120576 lt minus1 this array was calleda 119862 type event and the contaminant 119909
119888was used in the test
statisticTo an event of 119862 type when any of these four tests (N2
N8 N14 or N15) was applied the outcome was called either aspurious type II error probability (120587
119863119862) if the test was not
significant or a spurious power (120587119863119862
) if it was significant(Table 1) For this decision the calculated test statistic TN(TN2 TN8 TN14 or TN15) for a sample was compared withthe respective CV
99[19 20] If TN leCV
99 the outcome of the
test was considered as not significant else when TN gt CV99
the outcome of the test was considered as significant (Table 1)Similarly to an event of 119862 type when a discordancy test
was applied the outcome was either a nonspurious type IIerror probability (120587
119863119862) if the test was not significant or a
nonspurious power (120587119863119862
) if the test was significant (Table 1)If 120575 = 0 or 120576 = plusmn1 (the contaminant 119909
119888absent)
and a discordancy test was applied to the ordered array119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
to evaluate the extremeobservation 119909
(119899)or 119909(1) the outcome would either be a
true negative (the respective probability 120587119863) if the test was
not significant that is if it failed to detect 119909(119899)
or 119909(1)
asdiscordant or a type I error (probability 120587
119863) if the test was
significant that is it succeeded in detecting 119909(119899)
or 119909(1)
asdiscordant (Table 1)
4 Test Performance Criteria
Hayes and Kinsella [7] documented that a good discordancytest would be characterized by a high nonspurious powerprobability (high120587
119863119862) a low spurious power probability (low
120587119863119862
) and a low nonspurious type II error probability (low120587119863119862
)Hayes and Kinsella [7] defined the Power of Test (Ω) as
Ω = 120587119863119862
+ 120587119863119862
(5)
Similarly they also defined the Test Performance Crite-rion 120587
119863|119862(which is equivalent to the probability P5 of Barnett
and Lewis [1]) or the Conditional Power as
120587119863|119862
= P5 =120587119863119862
(120587119863119862
+ 120587119863119862
) (6)
5 Optimum Replications
The optimum replications (119872) required for minimizingthe errors of Monte Carlo simulations were decided fromrepresentative results summarized in Figures 1 and 2 inwhichthe vertical error bar represents total uncertainty at 99 con-fidence level (119906
99 equivalent to 99confidence interval of the
mean) for 102 simulation experiments For example for 119899 = 5
and 120575 = 10 Power of TestΩ is plotted in Figures 1(a)ndash1(d) as a
4 The Scientific World Journal
04858
04860
04862
04864
04866
0 5E6 1E7 15E7 2E7
M
N2 (E5 minus 55E6)
N2 (6E6 minus 2E7)
u99
Ω
(a) 119899 = 5 and 120575 = 10
04806
04808
04810
04812
04814
0 5E6 1E7 15E7 2E7
M
N8 (E5 minus 55E6)
N8 (6E6 minus 2E7)
u99
Ω
(b) 119899 = 5 and 120575 = 10
0M
04856
04858
04860
04862
04864
N14 (E5 minus 55E6)
N14 (6E6 minus 2E7)
5E6 1E7 15E7 2E7
u 99
Ω
(c) 119899 = 5 and 120575 = 10
04856
04858
04860
04862
0M
5E6 1E7 15E7 2E7
Ω
N15 (E5 minus 55E6)
N15 (6E6 minus 2E7)
u99
(d) 119899 = 5 and 120575 = 10
Figure 1 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for sample size 119899 = 5 andcontaminant parameter 120575 = 10 symbols are explained in each figure the vertical error bars represent uncertainty (119906
99) at 99 confidence
level from 102 simulations (a) test N2 (b) test N8 (c) test N14 and (d) test N15
The Scientific World Journal 5
Table 1 Sample simulation and test outcome (modified after Hayes and Kinsella [7])
Discordancy testresult
Contaminant 119909119888(119872 replications)
Present (120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1)Absent
(120575 = 0 or 120576 = 1)Contaminant 119909119888not used in a
discordancy test(119862 type event)
Contaminant 119909119888used in a
discordancy test(119862 type event)
Not significant(TN le CV99)
Spurious type II errorprobability (120587
119863119862)
(Figure 3)
Nonspurious type II errorprobability (120587
119863119862)
(Figures 5 and 7)
True negative probability(120587119863)
(Figures 3 5 and 7)
Significant(TN gt CV99)
Spurious power probability(120587119863119862
)(Figure 4)
Nonspurious power probability(120587119863119862
)(Figures 6 and 8)
Type I error probability(120587119863)
(Figures 4 6 and 8)TN calculated test statistic for a sample (TN2 TN8 TN14 or TN15) CV99 critical value for a given discordancy test at 99 confidence level
function of the replications (119872 = 100 000 to 20000000) forN2 N8 N14 and N15 AlthoughΩmean values remain prac-tically constant (within the confidence limits of the mean)for replications of about 8000000 still higher replicationsof 20000000 (Figures 1 and 2) were used in all simulationexperiments
SimilarlyΩ for all four tests as a function of replications isalso shown in Figures 2(a)ndash2(d) which allows a visual com-parison of this performance parameter for different samplesizes and 120575 values Error bars (119906
99) for the 102 simulation
experiments are not shown for simplicity but for replicationslarger than 10000000 they were certainly within the size ofthe symbols The replications of 20000000 routinely usedfor comparing the performance of discordancy tests clearlyshow that the differences amongΩ values (Figures 2(a)ndash2(d))are statistically significant at a high confidence level thatis these differences are much greater than the simulationerrors
Alternatively following Krishnamoorthy and Lian [17]the simulation error for the replications of 20000000 usedroutinely in our work can be estimated approximately as2 times radic05 times 0520000000 = 000022
Because we carried out 102 independent simulationexperiments each with 20000000 replications our sim-ulation errors were even less than the above value Thusthe Monte Carlo simulations can be considered highly pre-cise They can also be said to be highly accurate becauseour procedure was modified after the highly precise andaccurate method of Verma and Quiroz-Ruiz [21] Theseauthors had shown high precision and accuracy of each119880(0 1) and 119873(0 1) experiments and had also applied allkinds of simulated data quality tests suggested by Lawand Kelton [25] Besides in the present work a largenumber of such experiments (204 streams of 119880(0 1) and102 streams of 119873(0 1)) have been carried out Thereforeas an innovation in Monte Carlo simulations we presentthe mean (119909) values as well as the total uncertainty (119906
99)
of 102 independent experiments in terms of the confi-dence interval of the mean at the strict 99 confidencelevel
Finally in order to evaluate the test performance test N2was used as a reference and differences inmean (Δ119909N119895) valuesof the other three tests were calculated as
Δ119909N119895 = (119909N119895 minus 119909N2
119909N2) times 100 (7)
where the subscript N119895 stands for N8 N14 or N15
6 Results and Discussion
61 119862 Type and Contaminant-Absent Events According toBarnett and Lewis [1] this type of events is of no majorconcern because the contaminant 119909
119888occupies an inner
position in the ordered array and the extremeobservation119909(119899)
or 119909(1)
under evaluation from discordancy tests is a legitimateobservation An inner position of the contaminant wouldaffect much less the sample mean and standard deviation [1]For small values of 120575 or 120576 close to 0 or plusmn1 respectively mostevents generated from the Monte Carlo simulation are of 119862type The 120587
119863119862and 120587
119863119862values for 119899 = 5 to 119899 = 20 as a
function of 120575 are presented in Figures 3(a)ndash3(d) and Figures4(a)ndash4(d) respectively For 120576 these parameters behave verysimilarly and therefore the corresponding diagrams are notpresented
When the contaminant is absent (120575 = 0 or 120576 = plusmn1)the 120587
119863119862and 120587
119863119862values are close to the expected values
of 099 and 001 respectively because the discordancy testswere applied at the 99 confidence level (open circles inFigures 3(a)ndash3(d) and Figures 4(a)ndash4(d)) As 120575 changes from0 to about plusmn25 the 120587
119863119862values slightly increase from 099
to about 0996 for 119899 = 5 (Figure 3(a)) 0996 for 119899 = 10
(Figure 3(b)) 0994-0995 for 119899 = 15 (Figure 3(c)) and 0993-0994 for 119899 = 20 (Figure 3(d)) The 120587
119863119862values show the
complementary behavior (Figures 4(a)ndash4(d)) Because in thistype of events (119862) a legitimate extreme observation is beingtested our best desire is that the 120587
119863119862and 120587
119863119862values remain
close to the theoretical values of 099 and 001 respectivelyfor contaminant-absent events This is actually observed inFigures 3 and 4
6 The Scientific World Journal
Table2Non
Spurious
type
IIerrorp
robability(120587119863119862)p
aram
eter
forfou
rsinglee
xtremeo
utlierd
iscordancytests
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
50
09900
03000
00057
0990026
000
00058
00023
09900
05000
00057
000
020989994
000
00058
minus000
095
plusmn01
0989948
00000139
0989975
00000140
00027
0989950
00000140
00002
0989940
00000144
minus000
095
plusmn2
0977285
00000113
0977358
00000125
00075
0977290
00000111
00005
0977267
00000116
minus00018
5plusmn3
0961486
000
00148
0961632
000
00142
00152
0961497
000
00146
00012
0961459
000
00146
minus00028
5plusmn5
0891864
00000205
0892491
00000194
0070
0891901
00000200
000
430891823
00000208
minus000
465
plusmn6
0834635
000
00222
0835779
000
00231
0137
0834704
000
00226
00084
08346
02000
00225
minus00038
5plusmn7
07646
60000
00240
0766559
000
00257
0248
0764766
000
00250
00139
07646
61000
00237
000
025
plusmn8
0685336
00000253
0688202
00000255
0418
0685496
00000249
00234
0685401
00000247
00095
5plusmn10
0514122
00000280
0519336
00000282
101
05144
0700000292
0056
0514394
00000287
0053
5plusmn12
035110
3000
00258
0358468
000
00275
210
0351500
000
00257
0113
0351629
000
00258
0150
5plusmn15
0165615
000
00194
01740
53000
00215
51
0166070
000
00196
0275
0166367
000
00196
045
5plusmn18
0062959
000
00126
0069566
000
00148
105
0063321
000
00127
058
0063618
000
00134
105
5plusmn20
0029322
000
00082
0033997
000
00089
159
0029577
000
00084
087
0029815
000
00081
168
100
09900
08000
00057
09900
41000
00057
00034
09900
03000
00054
minus000
060990025
000
00059
00017
10plusmn01
0989851
00000197
0989888
00000173
00037
0989844
00000207
minus000
060989869
000
00195
00019
10plusmn1
0978810
00000183
0979321
00000174
0052
0978911
00000179
00103
0978879
00000188
00070
10plusmn2
0949169
00000168
0951501
00000163
0246
0949704
00000165
0056
0949433
00000175
00279
10plusmn3
0878974
00000199
0886717
00000192
088
0880804
00000197
0208
0879762
00000202
0090
10plusmn4
0736953
000
00241
0756911
000
00242
271
0741679
000
00266
064
0738849
000
00253
0257
10plusmn5
0523889
000
00273
0561084
000
00257
710532523
000
00263
165
0527197
000
00266
063
10plusmn6
0300940
000
00248
0349468
000
00258
161
0311799
000
00260
361
0304921
000
00249
132
10plusmn7
0136560
000
00182
0181585
000
00226
330
0146058
000
00174
700139899
000
00185
244
10plusmn8
004
8488
00000122
0079155
00000144
6300544
2200000118
122
0050495
00000130
414
10plusmn9
0013431
000
00074
0029234
000
00092
1180016139
000
00075
202
0014300
000
00074
65
10plusmn10
0002900
000
00031
000
9248
000
00056
219
0003821
000
00036
318
0003180
000
00030
9610
plusmn11
000
0489
000
00013
0002529
000
00029
418
000
0726
000
00016
49000
0556
000
00014
138
150
0989998
0000006
0990015
00000062
00017
098998
00000061
minus00018
0989989
0000006
minus00010
15plusmn01
0989768
00000204
0989811
00000210
00044
0989766
00000209
minus00002
0989759
00000200
minus000
0915
plusmn05
0985307
00000236
0985665
00000202
00364
0985507
00000249
00204
0985345
00000240
00039
15plusmn2
0974943
00000225
0976229
00000186
0132
0975702
00000241
0078
0975113
00000236
00174
15plusmn5
0932551
000
00202
0938851
000
00184
068
09360
96000
00207
0380
0933417
000
00206
0093
15plusmn4
0827672
000
00245
0848433
000
00236
251
0838586
000
00251
132
0830350
000
00236
0324
15plusmn45
0621647
000
00288
0669717
000
00270
77064
4852
000
00332
373
0627283
000
00285
091
15plusmn5
0487591
000
00276
0550018
000
00319
128
0516243
000
00322
59
0494476
000
00290
141
15plusmn5
0351577
000
00269
0423601
000
00266
205
0382834
000
00290
89
0358972
000
00287
210
15plusmn6
0137443
000
00198
0204194
000
00211
490162735
000
00196
184
0143150
000
00201
415
15plusmn7
0035757
000
00093
0074779
00000144
109
004
8090
00000112
345
0038352
00000098
7315
plusmn8
000
6107
000
00043
002118
4000
00094
249
000
9822
000
00054
61000
6819
000
0004
2117
15plusmn9
000
0682
000
00015
000
4748
000
00039
600
0001393
000
00019
104
000
0802
000
00017
176
The Scientific World Journal 7
Table2Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn10
000
0049
000
0000
4000
0858
000
00017
1640
000
0138
000
00007
180
000
0062
000
00005
257
200
0990017
000
00057
09900
04000
0006
minus00014
0990026
000
0006
000
0909900
04000
00059
minus00014
20plusmn01
0989743
00000259
0989771
00000248
00028
0989786
00000243
000
430989734
00000277
minus000
0920
plusmn05
09846
0000000243
0985127
00000258
0054
0985093
00000265
0050
09846
8200000246
00083
20plusmn1
0972754
00000237
0974729
00000256
0203
09744
3900000213
0173
0973084
00000225
00339
20plusmn2
0922859
000
00259
0932611
000
00219
106
0930457
000
00234
082
09244
63000
00266
0174
20plusmn3
0798422
000
00265
0829677
000
00230
391
0820494
000
00248
276
0803167
000
00283
059
20plusmn4
0561409
000
00283
0628839
000
00267
120
060
4043
000
00273
760570474
000
00288
161
20plusmn45
0415514
00000280
0498225
00000261
199
046
4616
00000262
118
0425784
00000285
247
20plusmn5
0276626
000
00244
0365289
000
00265
321
0325675
000
00297
177
0286659
000
00245
363
20plusmn6
0085535
000
00163
0152881
000
00198
790116
833
000
00185
366
0091506
000
00175
7020
plusmn7
0015809
000
00070
004
6048
000
00127
191
0026842
000
00087
700017699
000
00079
120
20plusmn8
0001716
000
00021
0010242
000
00053
500
0003921
000
00036
128
0002041
000
00025
189
20plusmn9
000
0109
000
0000
60001733
000
00024
1500
000
0366
000
00011
237
000
0139
000
00007
285
20plusmn10
000
0004
000
00001
000
0229
000
00010
5600
000
0022
000
00002
450
000
0006
000
00001
413
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
2 The Scientific World Journal
Monte Carlo simulation methods have been exten-sively used in numerous simulation studies [8ndash18] Some ofthe relatively recent papers are by Efstathiou [12] Gottardo etal [13] Khedhiri andMontasser [14] P A Patel and J S Patel[15] Noughabi and Arghami [16] Krishnamoorthy and Lian[17] and Verma [18] For example Noughabi and Arghami[16] compared seven normality tests (Kolmogorov-SmirnovAnderson-Darling Kuiper Jarque-Bera Cramer von MisesShapiro-Wilk and Vasicek) for sample sizes of 10 20 30and 50 and under different circumstances recommended theuse of Jarque-Bera Anderson-Darling Shapiro-Wilk andVasicek tests
We used Monte Carlo simulations to evaluate compara-tive efficiency of four extreme outlier type discordancy tests(N2 N8 N14 and N15 the nomenclature after Barnett andLewis [1]) for sample sizes of 5 to 20 Our approach tothe statistical problem of test performance is novel becauseinstead of using commercial or freely available software weprogrammed and generated extremely precise and accuraterandom numbers and normally distributed data used verylarge replications of 20000000 performed 102 independentexperiments and reduced the simulation errors to such anextent that the differences in test performance are far greaterthan the total uncertainties expressed as 99 confidenceintervals of the mean This is an approach hitherto practicedby none (see eg [8ndash18]) except by our group [19ndash23]This work therefore supersedes the approximate simulationresults of test performance reported by the statisticians Hayesand Kinsella [7]
2 Discordancy Tests
For a data array 1199091 1199092 1199093 119909
119899minus2 119909119899minus1
119909119899or an ordered
array 119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
of 119899 observationswith mean 119909 and standard deviation 119904 four test statisticswere objectively evaluated in this work For a statisticallycontaminated sample of size of 5 to 20 119899 minus 1 observationsof this data array were obtained from a normal distribution119873(0 1) and the remaining observation was taken from acentral tendency shifted distribution119873(0+120575 1) or dispersionshifted distribution 119873(0 1 times 120576) where the contaminantparameters 120575 for modeling slippage of central tendency and120576 for slippage of dispersion can be either positive or negativeFor an uncontaminated sample the simulations were donefor 120575 = 0 and 120576 = plusmn1 In order to achieve an unbiasedcomparison the application of the tests was always forced tothe upper outlier 119909
(119899)for positive values of 120575 or 120576 and to the
lower outlier 119909(1)
for negative values of 120575 or 120576Thus the first test was the Grubbs test N2 [24] for an
extreme outlier 119909(119899)
or 119909(1) for which the test statistic is as
follows
TN2 =
119909(119899)
minus 119909
119904 119909(119899)
tested if 120575 gt 0 or 120576 gt 1
119909 minus 119909(1)
119904 119909(1)
tested if 120575 lt 0 or 120576 lt 1
Max(119909(119899)
minus 119909
119904119909 minus 119909(1)
119904)
119909(119899)
or 119909(1)
tested if 120575=0 or 120576=1
(1)
The second test was the Dixon test N8 [2] as follows
TN8 =
119909(119899)
minus 119909(119899minus1)
119909(119899)
minus 119909(1)
119909(119899)
tested if 120575 gt 0 or 120576 gt 1
119909(2)
minus 119909(1)
119909(119899)
minus 119909(1)
119909(1)
tested if 120575 lt 0 or 120576 lt 1
Max(119909(119899)
minus 119909(119899minus1)
119909(119899)
minus 119909(1)
119909(2)
minus 119909(1)
119909(119899)
minus 119909(1)
)
119909(119899)
or119909(1)
tested if 120575=0 or 120576=1
(2)
The third test was sample skewness N14 as (note that theabsolute value is used for evaluation)
TN14 =
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(119899)
tested if 120575 gt 0 or 120576 gt 1
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(1)
tested if 120575 lt 0 or 120576 lt 1
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(119899)
tested if11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
gt 0
120575 = 0 or 120576 = 1
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
1003816100381610038161003816100381610038161003816100381610038161003816100381610038161003816
119909(1)
tested if11989912
sum119899
119894=1(119909119894minus 119909)3
sum119899
119894=1(119909119894minus 119909)2
32
lt 0
120575 = 0 or 120576 = minus1
(3)
The Scientific World Journal 3
Finally the fourth test was the sample kurtosis N15 asfollows
TN15 =
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(119899)
tested if 120575 gt 0 or 120576 gt 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(1)
tested if 120575 lt 0 or 120576 lt 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(119899)
tested if (119909(119899)
minus 119909) gt (119909 minus 119909(1)
)
120575 = 0 or 120576 = 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(1)
tested if (119909(119899)
minus 119909) lt (119909 minus 119909(1)
)
120575 gt 0 or 120576 = minus1
(4)
All testswere applied at a strict 99confidence level usingthe new precise and accurate critical values (CV
99) simulated
using Monte Carlo procedure by Verma et al [19] for N2N8 and N15 and Verma andQuiroz-Ruiz [20] for N14 whichpermitted an objective comparison of their performance
3 Monte Carlo Simulations
Random numbers 119880(0 1) uniformly distributed in the inter-val (0 1) and normal random variates119873(0 1)were generatedfrom the method summarized by Verma and Quiroz-Ruiz[21] However instead of only 10 series or streams of 119873(0 1)
as done byVerma andQuiroz-Ruiz [21] a total of 102 differentstreams of 119873(0 1) were simulated Similarly the replicationswere much more than those used by Verma and Quiroz-Ruiz[21] for generating precise and accurate critical values
For a data array of size 119899 (119899 minus 1) observations1199091 1199092 1199093 119909
119899minus2 119909119899minus1
were drawn from one stream of119873(0 1) and the contaminant observation (119909
119888) was added
from a different central tendency shifted stream of119873(0+120575 1)
where 120575was varied from+01minus+20 andminus01minusminus20 or a disper-sion shifted distribution119873(0 1 times 120576) where 120576 was varied from+11minus+200 andminus11minusminus200The simulation experimentswerealso carried out for uncontaminated distributions in which(119899 minus 1) observations were taken from one stream of normalrandom variates 119873(0 1) and an additional observation wasincorporated from a different stream with no contaminationthat is 120575 = 0 and 120576 = plusmn1
Now if we were to arrange the complete array from thelowest to the highest observations the ordered array could becalled 119909
(1) 119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
after Barnett andLewis [1] All four tests under evaluation could then beapplied to the resulting data array
If 120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1 (the contaminant 119909119888
present) two possibilities would arise for the ordered array
119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
as follows (Table 1) (i) thecontaminant 119909
119888occupies an inner position in the ordered
array that is 119909119888lt 119909(119899)
if 120575 gt 0 or 120576 gt 1 or 119909119888gt 119909(1)
if 120575 lt 0 or120576 lt minus1 this array is called a119862 type event and the contaminant119909119888was not used in the test statistic and (ii) the contaminant
119909119888occupies the extreme position that is 119909
119888= 119909(119899)
if 120575 gt 0 or120576 gt 1 or 119909
119888= 119909(1)
if 120575 lt 0 or 120576 lt minus1 this array was calleda 119862 type event and the contaminant 119909
119888was used in the test
statisticTo an event of 119862 type when any of these four tests (N2
N8 N14 or N15) was applied the outcome was called either aspurious type II error probability (120587
119863119862) if the test was not
significant or a spurious power (120587119863119862
) if it was significant(Table 1) For this decision the calculated test statistic TN(TN2 TN8 TN14 or TN15) for a sample was compared withthe respective CV
99[19 20] If TN leCV
99 the outcome of the
test was considered as not significant else when TN gt CV99
the outcome of the test was considered as significant (Table 1)Similarly to an event of 119862 type when a discordancy test
was applied the outcome was either a nonspurious type IIerror probability (120587
119863119862) if the test was not significant or a
nonspurious power (120587119863119862
) if the test was significant (Table 1)If 120575 = 0 or 120576 = plusmn1 (the contaminant 119909
119888absent)
and a discordancy test was applied to the ordered array119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
to evaluate the extremeobservation 119909
(119899)or 119909(1) the outcome would either be a
true negative (the respective probability 120587119863) if the test was
not significant that is if it failed to detect 119909(119899)
or 119909(1)
asdiscordant or a type I error (probability 120587
119863) if the test was
significant that is it succeeded in detecting 119909(119899)
or 119909(1)
asdiscordant (Table 1)
4 Test Performance Criteria
Hayes and Kinsella [7] documented that a good discordancytest would be characterized by a high nonspurious powerprobability (high120587
119863119862) a low spurious power probability (low
120587119863119862
) and a low nonspurious type II error probability (low120587119863119862
)Hayes and Kinsella [7] defined the Power of Test (Ω) as
Ω = 120587119863119862
+ 120587119863119862
(5)
Similarly they also defined the Test Performance Crite-rion 120587
119863|119862(which is equivalent to the probability P5 of Barnett
and Lewis [1]) or the Conditional Power as
120587119863|119862
= P5 =120587119863119862
(120587119863119862
+ 120587119863119862
) (6)
5 Optimum Replications
The optimum replications (119872) required for minimizingthe errors of Monte Carlo simulations were decided fromrepresentative results summarized in Figures 1 and 2 inwhichthe vertical error bar represents total uncertainty at 99 con-fidence level (119906
99 equivalent to 99confidence interval of the
mean) for 102 simulation experiments For example for 119899 = 5
and 120575 = 10 Power of TestΩ is plotted in Figures 1(a)ndash1(d) as a
4 The Scientific World Journal
04858
04860
04862
04864
04866
0 5E6 1E7 15E7 2E7
M
N2 (E5 minus 55E6)
N2 (6E6 minus 2E7)
u99
Ω
(a) 119899 = 5 and 120575 = 10
04806
04808
04810
04812
04814
0 5E6 1E7 15E7 2E7
M
N8 (E5 minus 55E6)
N8 (6E6 minus 2E7)
u99
Ω
(b) 119899 = 5 and 120575 = 10
0M
04856
04858
04860
04862
04864
N14 (E5 minus 55E6)
N14 (6E6 minus 2E7)
5E6 1E7 15E7 2E7
u 99
Ω
(c) 119899 = 5 and 120575 = 10
04856
04858
04860
04862
0M
5E6 1E7 15E7 2E7
Ω
N15 (E5 minus 55E6)
N15 (6E6 minus 2E7)
u99
(d) 119899 = 5 and 120575 = 10
Figure 1 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for sample size 119899 = 5 andcontaminant parameter 120575 = 10 symbols are explained in each figure the vertical error bars represent uncertainty (119906
99) at 99 confidence
level from 102 simulations (a) test N2 (b) test N8 (c) test N14 and (d) test N15
The Scientific World Journal 5
Table 1 Sample simulation and test outcome (modified after Hayes and Kinsella [7])
Discordancy testresult
Contaminant 119909119888(119872 replications)
Present (120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1)Absent
(120575 = 0 or 120576 = 1)Contaminant 119909119888not used in a
discordancy test(119862 type event)
Contaminant 119909119888used in a
discordancy test(119862 type event)
Not significant(TN le CV99)
Spurious type II errorprobability (120587
119863119862)
(Figure 3)
Nonspurious type II errorprobability (120587
119863119862)
(Figures 5 and 7)
True negative probability(120587119863)
(Figures 3 5 and 7)
Significant(TN gt CV99)
Spurious power probability(120587119863119862
)(Figure 4)
Nonspurious power probability(120587119863119862
)(Figures 6 and 8)
Type I error probability(120587119863)
(Figures 4 6 and 8)TN calculated test statistic for a sample (TN2 TN8 TN14 or TN15) CV99 critical value for a given discordancy test at 99 confidence level
function of the replications (119872 = 100 000 to 20000000) forN2 N8 N14 and N15 AlthoughΩmean values remain prac-tically constant (within the confidence limits of the mean)for replications of about 8000000 still higher replicationsof 20000000 (Figures 1 and 2) were used in all simulationexperiments
SimilarlyΩ for all four tests as a function of replications isalso shown in Figures 2(a)ndash2(d) which allows a visual com-parison of this performance parameter for different samplesizes and 120575 values Error bars (119906
99) for the 102 simulation
experiments are not shown for simplicity but for replicationslarger than 10000000 they were certainly within the size ofthe symbols The replications of 20000000 routinely usedfor comparing the performance of discordancy tests clearlyshow that the differences amongΩ values (Figures 2(a)ndash2(d))are statistically significant at a high confidence level thatis these differences are much greater than the simulationerrors
Alternatively following Krishnamoorthy and Lian [17]the simulation error for the replications of 20000000 usedroutinely in our work can be estimated approximately as2 times radic05 times 0520000000 = 000022
Because we carried out 102 independent simulationexperiments each with 20000000 replications our sim-ulation errors were even less than the above value Thusthe Monte Carlo simulations can be considered highly pre-cise They can also be said to be highly accurate becauseour procedure was modified after the highly precise andaccurate method of Verma and Quiroz-Ruiz [21] Theseauthors had shown high precision and accuracy of each119880(0 1) and 119873(0 1) experiments and had also applied allkinds of simulated data quality tests suggested by Lawand Kelton [25] Besides in the present work a largenumber of such experiments (204 streams of 119880(0 1) and102 streams of 119873(0 1)) have been carried out Thereforeas an innovation in Monte Carlo simulations we presentthe mean (119909) values as well as the total uncertainty (119906
99)
of 102 independent experiments in terms of the confi-dence interval of the mean at the strict 99 confidencelevel
Finally in order to evaluate the test performance test N2was used as a reference and differences inmean (Δ119909N119895) valuesof the other three tests were calculated as
Δ119909N119895 = (119909N119895 minus 119909N2
119909N2) times 100 (7)
where the subscript N119895 stands for N8 N14 or N15
6 Results and Discussion
61 119862 Type and Contaminant-Absent Events According toBarnett and Lewis [1] this type of events is of no majorconcern because the contaminant 119909
119888occupies an inner
position in the ordered array and the extremeobservation119909(119899)
or 119909(1)
under evaluation from discordancy tests is a legitimateobservation An inner position of the contaminant wouldaffect much less the sample mean and standard deviation [1]For small values of 120575 or 120576 close to 0 or plusmn1 respectively mostevents generated from the Monte Carlo simulation are of 119862type The 120587
119863119862and 120587
119863119862values for 119899 = 5 to 119899 = 20 as a
function of 120575 are presented in Figures 3(a)ndash3(d) and Figures4(a)ndash4(d) respectively For 120576 these parameters behave verysimilarly and therefore the corresponding diagrams are notpresented
When the contaminant is absent (120575 = 0 or 120576 = plusmn1)the 120587
119863119862and 120587
119863119862values are close to the expected values
of 099 and 001 respectively because the discordancy testswere applied at the 99 confidence level (open circles inFigures 3(a)ndash3(d) and Figures 4(a)ndash4(d)) As 120575 changes from0 to about plusmn25 the 120587
119863119862values slightly increase from 099
to about 0996 for 119899 = 5 (Figure 3(a)) 0996 for 119899 = 10
(Figure 3(b)) 0994-0995 for 119899 = 15 (Figure 3(c)) and 0993-0994 for 119899 = 20 (Figure 3(d)) The 120587
119863119862values show the
complementary behavior (Figures 4(a)ndash4(d)) Because in thistype of events (119862) a legitimate extreme observation is beingtested our best desire is that the 120587
119863119862and 120587
119863119862values remain
close to the theoretical values of 099 and 001 respectivelyfor contaminant-absent events This is actually observed inFigures 3 and 4
6 The Scientific World Journal
Table2Non
Spurious
type
IIerrorp
robability(120587119863119862)p
aram
eter
forfou
rsinglee
xtremeo
utlierd
iscordancytests
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
50
09900
03000
00057
0990026
000
00058
00023
09900
05000
00057
000
020989994
000
00058
minus000
095
plusmn01
0989948
00000139
0989975
00000140
00027
0989950
00000140
00002
0989940
00000144
minus000
095
plusmn2
0977285
00000113
0977358
00000125
00075
0977290
00000111
00005
0977267
00000116
minus00018
5plusmn3
0961486
000
00148
0961632
000
00142
00152
0961497
000
00146
00012
0961459
000
00146
minus00028
5plusmn5
0891864
00000205
0892491
00000194
0070
0891901
00000200
000
430891823
00000208
minus000
465
plusmn6
0834635
000
00222
0835779
000
00231
0137
0834704
000
00226
00084
08346
02000
00225
minus00038
5plusmn7
07646
60000
00240
0766559
000
00257
0248
0764766
000
00250
00139
07646
61000
00237
000
025
plusmn8
0685336
00000253
0688202
00000255
0418
0685496
00000249
00234
0685401
00000247
00095
5plusmn10
0514122
00000280
0519336
00000282
101
05144
0700000292
0056
0514394
00000287
0053
5plusmn12
035110
3000
00258
0358468
000
00275
210
0351500
000
00257
0113
0351629
000
00258
0150
5plusmn15
0165615
000
00194
01740
53000
00215
51
0166070
000
00196
0275
0166367
000
00196
045
5plusmn18
0062959
000
00126
0069566
000
00148
105
0063321
000
00127
058
0063618
000
00134
105
5plusmn20
0029322
000
00082
0033997
000
00089
159
0029577
000
00084
087
0029815
000
00081
168
100
09900
08000
00057
09900
41000
00057
00034
09900
03000
00054
minus000
060990025
000
00059
00017
10plusmn01
0989851
00000197
0989888
00000173
00037
0989844
00000207
minus000
060989869
000
00195
00019
10plusmn1
0978810
00000183
0979321
00000174
0052
0978911
00000179
00103
0978879
00000188
00070
10plusmn2
0949169
00000168
0951501
00000163
0246
0949704
00000165
0056
0949433
00000175
00279
10plusmn3
0878974
00000199
0886717
00000192
088
0880804
00000197
0208
0879762
00000202
0090
10plusmn4
0736953
000
00241
0756911
000
00242
271
0741679
000
00266
064
0738849
000
00253
0257
10plusmn5
0523889
000
00273
0561084
000
00257
710532523
000
00263
165
0527197
000
00266
063
10plusmn6
0300940
000
00248
0349468
000
00258
161
0311799
000
00260
361
0304921
000
00249
132
10plusmn7
0136560
000
00182
0181585
000
00226
330
0146058
000
00174
700139899
000
00185
244
10plusmn8
004
8488
00000122
0079155
00000144
6300544
2200000118
122
0050495
00000130
414
10plusmn9
0013431
000
00074
0029234
000
00092
1180016139
000
00075
202
0014300
000
00074
65
10plusmn10
0002900
000
00031
000
9248
000
00056
219
0003821
000
00036
318
0003180
000
00030
9610
plusmn11
000
0489
000
00013
0002529
000
00029
418
000
0726
000
00016
49000
0556
000
00014
138
150
0989998
0000006
0990015
00000062
00017
098998
00000061
minus00018
0989989
0000006
minus00010
15plusmn01
0989768
00000204
0989811
00000210
00044
0989766
00000209
minus00002
0989759
00000200
minus000
0915
plusmn05
0985307
00000236
0985665
00000202
00364
0985507
00000249
00204
0985345
00000240
00039
15plusmn2
0974943
00000225
0976229
00000186
0132
0975702
00000241
0078
0975113
00000236
00174
15plusmn5
0932551
000
00202
0938851
000
00184
068
09360
96000
00207
0380
0933417
000
00206
0093
15plusmn4
0827672
000
00245
0848433
000
00236
251
0838586
000
00251
132
0830350
000
00236
0324
15plusmn45
0621647
000
00288
0669717
000
00270
77064
4852
000
00332
373
0627283
000
00285
091
15plusmn5
0487591
000
00276
0550018
000
00319
128
0516243
000
00322
59
0494476
000
00290
141
15plusmn5
0351577
000
00269
0423601
000
00266
205
0382834
000
00290
89
0358972
000
00287
210
15plusmn6
0137443
000
00198
0204194
000
00211
490162735
000
00196
184
0143150
000
00201
415
15plusmn7
0035757
000
00093
0074779
00000144
109
004
8090
00000112
345
0038352
00000098
7315
plusmn8
000
6107
000
00043
002118
4000
00094
249
000
9822
000
00054
61000
6819
000
0004
2117
15plusmn9
000
0682
000
00015
000
4748
000
00039
600
0001393
000
00019
104
000
0802
000
00017
176
The Scientific World Journal 7
Table2Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn10
000
0049
000
0000
4000
0858
000
00017
1640
000
0138
000
00007
180
000
0062
000
00005
257
200
0990017
000
00057
09900
04000
0006
minus00014
0990026
000
0006
000
0909900
04000
00059
minus00014
20plusmn01
0989743
00000259
0989771
00000248
00028
0989786
00000243
000
430989734
00000277
minus000
0920
plusmn05
09846
0000000243
0985127
00000258
0054
0985093
00000265
0050
09846
8200000246
00083
20plusmn1
0972754
00000237
0974729
00000256
0203
09744
3900000213
0173
0973084
00000225
00339
20plusmn2
0922859
000
00259
0932611
000
00219
106
0930457
000
00234
082
09244
63000
00266
0174
20plusmn3
0798422
000
00265
0829677
000
00230
391
0820494
000
00248
276
0803167
000
00283
059
20plusmn4
0561409
000
00283
0628839
000
00267
120
060
4043
000
00273
760570474
000
00288
161
20plusmn45
0415514
00000280
0498225
00000261
199
046
4616
00000262
118
0425784
00000285
247
20plusmn5
0276626
000
00244
0365289
000
00265
321
0325675
000
00297
177
0286659
000
00245
363
20plusmn6
0085535
000
00163
0152881
000
00198
790116
833
000
00185
366
0091506
000
00175
7020
plusmn7
0015809
000
00070
004
6048
000
00127
191
0026842
000
00087
700017699
000
00079
120
20plusmn8
0001716
000
00021
0010242
000
00053
500
0003921
000
00036
128
0002041
000
00025
189
20plusmn9
000
0109
000
0000
60001733
000
00024
1500
000
0366
000
00011
237
000
0139
000
00007
285
20plusmn10
000
0004
000
00001
000
0229
000
00010
5600
000
0022
000
00002
450
000
0006
000
00001
413
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 3
Finally the fourth test was the sample kurtosis N15 asfollows
TN15 =
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(119899)
tested if 120575 gt 0 or 120576 gt 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(1)
tested if 120575 lt 0 or 120576 lt 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(119899)
tested if (119909(119899)
minus 119909) gt (119909 minus 119909(1)
)
120575 = 0 or 120576 = 1
119899 sum119899
119894=1(119909119894minus 119909)4
sum119899
119894=1(119909119894minus 119909)2
2
119909(1)
tested if (119909(119899)
minus 119909) lt (119909 minus 119909(1)
)
120575 gt 0 or 120576 = minus1
(4)
All testswere applied at a strict 99confidence level usingthe new precise and accurate critical values (CV
99) simulated
using Monte Carlo procedure by Verma et al [19] for N2N8 and N15 and Verma andQuiroz-Ruiz [20] for N14 whichpermitted an objective comparison of their performance
3 Monte Carlo Simulations
Random numbers 119880(0 1) uniformly distributed in the inter-val (0 1) and normal random variates119873(0 1)were generatedfrom the method summarized by Verma and Quiroz-Ruiz[21] However instead of only 10 series or streams of 119873(0 1)
as done byVerma andQuiroz-Ruiz [21] a total of 102 differentstreams of 119873(0 1) were simulated Similarly the replicationswere much more than those used by Verma and Quiroz-Ruiz[21] for generating precise and accurate critical values
For a data array of size 119899 (119899 minus 1) observations1199091 1199092 1199093 119909
119899minus2 119909119899minus1
were drawn from one stream of119873(0 1) and the contaminant observation (119909
119888) was added
from a different central tendency shifted stream of119873(0+120575 1)
where 120575was varied from+01minus+20 andminus01minusminus20 or a disper-sion shifted distribution119873(0 1 times 120576) where 120576 was varied from+11minus+200 andminus11minusminus200The simulation experimentswerealso carried out for uncontaminated distributions in which(119899 minus 1) observations were taken from one stream of normalrandom variates 119873(0 1) and an additional observation wasincorporated from a different stream with no contaminationthat is 120575 = 0 and 120576 = plusmn1
Now if we were to arrange the complete array from thelowest to the highest observations the ordered array could becalled 119909
(1) 119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
after Barnett andLewis [1] All four tests under evaluation could then beapplied to the resulting data array
If 120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1 (the contaminant 119909119888
present) two possibilities would arise for the ordered array
119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
as follows (Table 1) (i) thecontaminant 119909
119888occupies an inner position in the ordered
array that is 119909119888lt 119909(119899)
if 120575 gt 0 or 120576 gt 1 or 119909119888gt 119909(1)
if 120575 lt 0 or120576 lt minus1 this array is called a119862 type event and the contaminant119909119888was not used in the test statistic and (ii) the contaminant
119909119888occupies the extreme position that is 119909
119888= 119909(119899)
if 120575 gt 0 or120576 gt 1 or 119909
119888= 119909(1)
if 120575 lt 0 or 120576 lt minus1 this array was calleda 119862 type event and the contaminant 119909
119888was used in the test
statisticTo an event of 119862 type when any of these four tests (N2
N8 N14 or N15) was applied the outcome was called either aspurious type II error probability (120587
119863119862) if the test was not
significant or a spurious power (120587119863119862
) if it was significant(Table 1) For this decision the calculated test statistic TN(TN2 TN8 TN14 or TN15) for a sample was compared withthe respective CV
99[19 20] If TN leCV
99 the outcome of the
test was considered as not significant else when TN gt CV99
the outcome of the test was considered as significant (Table 1)Similarly to an event of 119862 type when a discordancy test
was applied the outcome was either a nonspurious type IIerror probability (120587
119863119862) if the test was not significant or a
nonspurious power (120587119863119862
) if the test was significant (Table 1)If 120575 = 0 or 120576 = plusmn1 (the contaminant 119909
119888absent)
and a discordancy test was applied to the ordered array119909(1)
119909(2)
119909(3)
119909(119899minus2)
119909(119899minus1)
119909(119899)
to evaluate the extremeobservation 119909
(119899)or 119909(1) the outcome would either be a
true negative (the respective probability 120587119863) if the test was
not significant that is if it failed to detect 119909(119899)
or 119909(1)
asdiscordant or a type I error (probability 120587
119863) if the test was
significant that is it succeeded in detecting 119909(119899)
or 119909(1)
asdiscordant (Table 1)
4 Test Performance Criteria
Hayes and Kinsella [7] documented that a good discordancytest would be characterized by a high nonspurious powerprobability (high120587
119863119862) a low spurious power probability (low
120587119863119862
) and a low nonspurious type II error probability (low120587119863119862
)Hayes and Kinsella [7] defined the Power of Test (Ω) as
Ω = 120587119863119862
+ 120587119863119862
(5)
Similarly they also defined the Test Performance Crite-rion 120587
119863|119862(which is equivalent to the probability P5 of Barnett
and Lewis [1]) or the Conditional Power as
120587119863|119862
= P5 =120587119863119862
(120587119863119862
+ 120587119863119862
) (6)
5 Optimum Replications
The optimum replications (119872) required for minimizingthe errors of Monte Carlo simulations were decided fromrepresentative results summarized in Figures 1 and 2 inwhichthe vertical error bar represents total uncertainty at 99 con-fidence level (119906
99 equivalent to 99confidence interval of the
mean) for 102 simulation experiments For example for 119899 = 5
and 120575 = 10 Power of TestΩ is plotted in Figures 1(a)ndash1(d) as a
4 The Scientific World Journal
04858
04860
04862
04864
04866
0 5E6 1E7 15E7 2E7
M
N2 (E5 minus 55E6)
N2 (6E6 minus 2E7)
u99
Ω
(a) 119899 = 5 and 120575 = 10
04806
04808
04810
04812
04814
0 5E6 1E7 15E7 2E7
M
N8 (E5 minus 55E6)
N8 (6E6 minus 2E7)
u99
Ω
(b) 119899 = 5 and 120575 = 10
0M
04856
04858
04860
04862
04864
N14 (E5 minus 55E6)
N14 (6E6 minus 2E7)
5E6 1E7 15E7 2E7
u 99
Ω
(c) 119899 = 5 and 120575 = 10
04856
04858
04860
04862
0M
5E6 1E7 15E7 2E7
Ω
N15 (E5 minus 55E6)
N15 (6E6 minus 2E7)
u99
(d) 119899 = 5 and 120575 = 10
Figure 1 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for sample size 119899 = 5 andcontaminant parameter 120575 = 10 symbols are explained in each figure the vertical error bars represent uncertainty (119906
99) at 99 confidence
level from 102 simulations (a) test N2 (b) test N8 (c) test N14 and (d) test N15
The Scientific World Journal 5
Table 1 Sample simulation and test outcome (modified after Hayes and Kinsella [7])
Discordancy testresult
Contaminant 119909119888(119872 replications)
Present (120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1)Absent
(120575 = 0 or 120576 = 1)Contaminant 119909119888not used in a
discordancy test(119862 type event)
Contaminant 119909119888used in a
discordancy test(119862 type event)
Not significant(TN le CV99)
Spurious type II errorprobability (120587
119863119862)
(Figure 3)
Nonspurious type II errorprobability (120587
119863119862)
(Figures 5 and 7)
True negative probability(120587119863)
(Figures 3 5 and 7)
Significant(TN gt CV99)
Spurious power probability(120587119863119862
)(Figure 4)
Nonspurious power probability(120587119863119862
)(Figures 6 and 8)
Type I error probability(120587119863)
(Figures 4 6 and 8)TN calculated test statistic for a sample (TN2 TN8 TN14 or TN15) CV99 critical value for a given discordancy test at 99 confidence level
function of the replications (119872 = 100 000 to 20000000) forN2 N8 N14 and N15 AlthoughΩmean values remain prac-tically constant (within the confidence limits of the mean)for replications of about 8000000 still higher replicationsof 20000000 (Figures 1 and 2) were used in all simulationexperiments
SimilarlyΩ for all four tests as a function of replications isalso shown in Figures 2(a)ndash2(d) which allows a visual com-parison of this performance parameter for different samplesizes and 120575 values Error bars (119906
99) for the 102 simulation
experiments are not shown for simplicity but for replicationslarger than 10000000 they were certainly within the size ofthe symbols The replications of 20000000 routinely usedfor comparing the performance of discordancy tests clearlyshow that the differences amongΩ values (Figures 2(a)ndash2(d))are statistically significant at a high confidence level thatis these differences are much greater than the simulationerrors
Alternatively following Krishnamoorthy and Lian [17]the simulation error for the replications of 20000000 usedroutinely in our work can be estimated approximately as2 times radic05 times 0520000000 = 000022
Because we carried out 102 independent simulationexperiments each with 20000000 replications our sim-ulation errors were even less than the above value Thusthe Monte Carlo simulations can be considered highly pre-cise They can also be said to be highly accurate becauseour procedure was modified after the highly precise andaccurate method of Verma and Quiroz-Ruiz [21] Theseauthors had shown high precision and accuracy of each119880(0 1) and 119873(0 1) experiments and had also applied allkinds of simulated data quality tests suggested by Lawand Kelton [25] Besides in the present work a largenumber of such experiments (204 streams of 119880(0 1) and102 streams of 119873(0 1)) have been carried out Thereforeas an innovation in Monte Carlo simulations we presentthe mean (119909) values as well as the total uncertainty (119906
99)
of 102 independent experiments in terms of the confi-dence interval of the mean at the strict 99 confidencelevel
Finally in order to evaluate the test performance test N2was used as a reference and differences inmean (Δ119909N119895) valuesof the other three tests were calculated as
Δ119909N119895 = (119909N119895 minus 119909N2
119909N2) times 100 (7)
where the subscript N119895 stands for N8 N14 or N15
6 Results and Discussion
61 119862 Type and Contaminant-Absent Events According toBarnett and Lewis [1] this type of events is of no majorconcern because the contaminant 119909
119888occupies an inner
position in the ordered array and the extremeobservation119909(119899)
or 119909(1)
under evaluation from discordancy tests is a legitimateobservation An inner position of the contaminant wouldaffect much less the sample mean and standard deviation [1]For small values of 120575 or 120576 close to 0 or plusmn1 respectively mostevents generated from the Monte Carlo simulation are of 119862type The 120587
119863119862and 120587
119863119862values for 119899 = 5 to 119899 = 20 as a
function of 120575 are presented in Figures 3(a)ndash3(d) and Figures4(a)ndash4(d) respectively For 120576 these parameters behave verysimilarly and therefore the corresponding diagrams are notpresented
When the contaminant is absent (120575 = 0 or 120576 = plusmn1)the 120587
119863119862and 120587
119863119862values are close to the expected values
of 099 and 001 respectively because the discordancy testswere applied at the 99 confidence level (open circles inFigures 3(a)ndash3(d) and Figures 4(a)ndash4(d)) As 120575 changes from0 to about plusmn25 the 120587
119863119862values slightly increase from 099
to about 0996 for 119899 = 5 (Figure 3(a)) 0996 for 119899 = 10
(Figure 3(b)) 0994-0995 for 119899 = 15 (Figure 3(c)) and 0993-0994 for 119899 = 20 (Figure 3(d)) The 120587
119863119862values show the
complementary behavior (Figures 4(a)ndash4(d)) Because in thistype of events (119862) a legitimate extreme observation is beingtested our best desire is that the 120587
119863119862and 120587
119863119862values remain
close to the theoretical values of 099 and 001 respectivelyfor contaminant-absent events This is actually observed inFigures 3 and 4
6 The Scientific World Journal
Table2Non
Spurious
type
IIerrorp
robability(120587119863119862)p
aram
eter
forfou
rsinglee
xtremeo
utlierd
iscordancytests
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
50
09900
03000
00057
0990026
000
00058
00023
09900
05000
00057
000
020989994
000
00058
minus000
095
plusmn01
0989948
00000139
0989975
00000140
00027
0989950
00000140
00002
0989940
00000144
minus000
095
plusmn2
0977285
00000113
0977358
00000125
00075
0977290
00000111
00005
0977267
00000116
minus00018
5plusmn3
0961486
000
00148
0961632
000
00142
00152
0961497
000
00146
00012
0961459
000
00146
minus00028
5plusmn5
0891864
00000205
0892491
00000194
0070
0891901
00000200
000
430891823
00000208
minus000
465
plusmn6
0834635
000
00222
0835779
000
00231
0137
0834704
000
00226
00084
08346
02000
00225
minus00038
5plusmn7
07646
60000
00240
0766559
000
00257
0248
0764766
000
00250
00139
07646
61000
00237
000
025
plusmn8
0685336
00000253
0688202
00000255
0418
0685496
00000249
00234
0685401
00000247
00095
5plusmn10
0514122
00000280
0519336
00000282
101
05144
0700000292
0056
0514394
00000287
0053
5plusmn12
035110
3000
00258
0358468
000
00275
210
0351500
000
00257
0113
0351629
000
00258
0150
5plusmn15
0165615
000
00194
01740
53000
00215
51
0166070
000
00196
0275
0166367
000
00196
045
5plusmn18
0062959
000
00126
0069566
000
00148
105
0063321
000
00127
058
0063618
000
00134
105
5plusmn20
0029322
000
00082
0033997
000
00089
159
0029577
000
00084
087
0029815
000
00081
168
100
09900
08000
00057
09900
41000
00057
00034
09900
03000
00054
minus000
060990025
000
00059
00017
10plusmn01
0989851
00000197
0989888
00000173
00037
0989844
00000207
minus000
060989869
000
00195
00019
10plusmn1
0978810
00000183
0979321
00000174
0052
0978911
00000179
00103
0978879
00000188
00070
10plusmn2
0949169
00000168
0951501
00000163
0246
0949704
00000165
0056
0949433
00000175
00279
10plusmn3
0878974
00000199
0886717
00000192
088
0880804
00000197
0208
0879762
00000202
0090
10plusmn4
0736953
000
00241
0756911
000
00242
271
0741679
000
00266
064
0738849
000
00253
0257
10plusmn5
0523889
000
00273
0561084
000
00257
710532523
000
00263
165
0527197
000
00266
063
10plusmn6
0300940
000
00248
0349468
000
00258
161
0311799
000
00260
361
0304921
000
00249
132
10plusmn7
0136560
000
00182
0181585
000
00226
330
0146058
000
00174
700139899
000
00185
244
10plusmn8
004
8488
00000122
0079155
00000144
6300544
2200000118
122
0050495
00000130
414
10plusmn9
0013431
000
00074
0029234
000
00092
1180016139
000
00075
202
0014300
000
00074
65
10plusmn10
0002900
000
00031
000
9248
000
00056
219
0003821
000
00036
318
0003180
000
00030
9610
plusmn11
000
0489
000
00013
0002529
000
00029
418
000
0726
000
00016
49000
0556
000
00014
138
150
0989998
0000006
0990015
00000062
00017
098998
00000061
minus00018
0989989
0000006
minus00010
15plusmn01
0989768
00000204
0989811
00000210
00044
0989766
00000209
minus00002
0989759
00000200
minus000
0915
plusmn05
0985307
00000236
0985665
00000202
00364
0985507
00000249
00204
0985345
00000240
00039
15plusmn2
0974943
00000225
0976229
00000186
0132
0975702
00000241
0078
0975113
00000236
00174
15plusmn5
0932551
000
00202
0938851
000
00184
068
09360
96000
00207
0380
0933417
000
00206
0093
15plusmn4
0827672
000
00245
0848433
000
00236
251
0838586
000
00251
132
0830350
000
00236
0324
15plusmn45
0621647
000
00288
0669717
000
00270
77064
4852
000
00332
373
0627283
000
00285
091
15plusmn5
0487591
000
00276
0550018
000
00319
128
0516243
000
00322
59
0494476
000
00290
141
15plusmn5
0351577
000
00269
0423601
000
00266
205
0382834
000
00290
89
0358972
000
00287
210
15plusmn6
0137443
000
00198
0204194
000
00211
490162735
000
00196
184
0143150
000
00201
415
15plusmn7
0035757
000
00093
0074779
00000144
109
004
8090
00000112
345
0038352
00000098
7315
plusmn8
000
6107
000
00043
002118
4000
00094
249
000
9822
000
00054
61000
6819
000
0004
2117
15plusmn9
000
0682
000
00015
000
4748
000
00039
600
0001393
000
00019
104
000
0802
000
00017
176
The Scientific World Journal 7
Table2Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn10
000
0049
000
0000
4000
0858
000
00017
1640
000
0138
000
00007
180
000
0062
000
00005
257
200
0990017
000
00057
09900
04000
0006
minus00014
0990026
000
0006
000
0909900
04000
00059
minus00014
20plusmn01
0989743
00000259
0989771
00000248
00028
0989786
00000243
000
430989734
00000277
minus000
0920
plusmn05
09846
0000000243
0985127
00000258
0054
0985093
00000265
0050
09846
8200000246
00083
20plusmn1
0972754
00000237
0974729
00000256
0203
09744
3900000213
0173
0973084
00000225
00339
20plusmn2
0922859
000
00259
0932611
000
00219
106
0930457
000
00234
082
09244
63000
00266
0174
20plusmn3
0798422
000
00265
0829677
000
00230
391
0820494
000
00248
276
0803167
000
00283
059
20plusmn4
0561409
000
00283
0628839
000
00267
120
060
4043
000
00273
760570474
000
00288
161
20plusmn45
0415514
00000280
0498225
00000261
199
046
4616
00000262
118
0425784
00000285
247
20plusmn5
0276626
000
00244
0365289
000
00265
321
0325675
000
00297
177
0286659
000
00245
363
20plusmn6
0085535
000
00163
0152881
000
00198
790116
833
000
00185
366
0091506
000
00175
7020
plusmn7
0015809
000
00070
004
6048
000
00127
191
0026842
000
00087
700017699
000
00079
120
20plusmn8
0001716
000
00021
0010242
000
00053
500
0003921
000
00036
128
0002041
000
00025
189
20plusmn9
000
0109
000
0000
60001733
000
00024
1500
000
0366
000
00011
237
000
0139
000
00007
285
20plusmn10
000
0004
000
00001
000
0229
000
00010
5600
000
0022
000
00002
450
000
0006
000
00001
413
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
4 The Scientific World Journal
04858
04860
04862
04864
04866
0 5E6 1E7 15E7 2E7
M
N2 (E5 minus 55E6)
N2 (6E6 minus 2E7)
u99
Ω
(a) 119899 = 5 and 120575 = 10
04806
04808
04810
04812
04814
0 5E6 1E7 15E7 2E7
M
N8 (E5 minus 55E6)
N8 (6E6 minus 2E7)
u99
Ω
(b) 119899 = 5 and 120575 = 10
0M
04856
04858
04860
04862
04864
N14 (E5 minus 55E6)
N14 (6E6 minus 2E7)
5E6 1E7 15E7 2E7
u 99
Ω
(c) 119899 = 5 and 120575 = 10
04856
04858
04860
04862
0M
5E6 1E7 15E7 2E7
Ω
N15 (E5 minus 55E6)
N15 (6E6 minus 2E7)
u99
(d) 119899 = 5 and 120575 = 10
Figure 1 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for sample size 119899 = 5 andcontaminant parameter 120575 = 10 symbols are explained in each figure the vertical error bars represent uncertainty (119906
99) at 99 confidence
level from 102 simulations (a) test N2 (b) test N8 (c) test N14 and (d) test N15
The Scientific World Journal 5
Table 1 Sample simulation and test outcome (modified after Hayes and Kinsella [7])
Discordancy testresult
Contaminant 119909119888(119872 replications)
Present (120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1)Absent
(120575 = 0 or 120576 = 1)Contaminant 119909119888not used in a
discordancy test(119862 type event)
Contaminant 119909119888used in a
discordancy test(119862 type event)
Not significant(TN le CV99)
Spurious type II errorprobability (120587
119863119862)
(Figure 3)
Nonspurious type II errorprobability (120587
119863119862)
(Figures 5 and 7)
True negative probability(120587119863)
(Figures 3 5 and 7)
Significant(TN gt CV99)
Spurious power probability(120587119863119862
)(Figure 4)
Nonspurious power probability(120587119863119862
)(Figures 6 and 8)
Type I error probability(120587119863)
(Figures 4 6 and 8)TN calculated test statistic for a sample (TN2 TN8 TN14 or TN15) CV99 critical value for a given discordancy test at 99 confidence level
function of the replications (119872 = 100 000 to 20000000) forN2 N8 N14 and N15 AlthoughΩmean values remain prac-tically constant (within the confidence limits of the mean)for replications of about 8000000 still higher replicationsof 20000000 (Figures 1 and 2) were used in all simulationexperiments
SimilarlyΩ for all four tests as a function of replications isalso shown in Figures 2(a)ndash2(d) which allows a visual com-parison of this performance parameter for different samplesizes and 120575 values Error bars (119906
99) for the 102 simulation
experiments are not shown for simplicity but for replicationslarger than 10000000 they were certainly within the size ofthe symbols The replications of 20000000 routinely usedfor comparing the performance of discordancy tests clearlyshow that the differences amongΩ values (Figures 2(a)ndash2(d))are statistically significant at a high confidence level thatis these differences are much greater than the simulationerrors
Alternatively following Krishnamoorthy and Lian [17]the simulation error for the replications of 20000000 usedroutinely in our work can be estimated approximately as2 times radic05 times 0520000000 = 000022
Because we carried out 102 independent simulationexperiments each with 20000000 replications our sim-ulation errors were even less than the above value Thusthe Monte Carlo simulations can be considered highly pre-cise They can also be said to be highly accurate becauseour procedure was modified after the highly precise andaccurate method of Verma and Quiroz-Ruiz [21] Theseauthors had shown high precision and accuracy of each119880(0 1) and 119873(0 1) experiments and had also applied allkinds of simulated data quality tests suggested by Lawand Kelton [25] Besides in the present work a largenumber of such experiments (204 streams of 119880(0 1) and102 streams of 119873(0 1)) have been carried out Thereforeas an innovation in Monte Carlo simulations we presentthe mean (119909) values as well as the total uncertainty (119906
99)
of 102 independent experiments in terms of the confi-dence interval of the mean at the strict 99 confidencelevel
Finally in order to evaluate the test performance test N2was used as a reference and differences inmean (Δ119909N119895) valuesof the other three tests were calculated as
Δ119909N119895 = (119909N119895 minus 119909N2
119909N2) times 100 (7)
where the subscript N119895 stands for N8 N14 or N15
6 Results and Discussion
61 119862 Type and Contaminant-Absent Events According toBarnett and Lewis [1] this type of events is of no majorconcern because the contaminant 119909
119888occupies an inner
position in the ordered array and the extremeobservation119909(119899)
or 119909(1)
under evaluation from discordancy tests is a legitimateobservation An inner position of the contaminant wouldaffect much less the sample mean and standard deviation [1]For small values of 120575 or 120576 close to 0 or plusmn1 respectively mostevents generated from the Monte Carlo simulation are of 119862type The 120587
119863119862and 120587
119863119862values for 119899 = 5 to 119899 = 20 as a
function of 120575 are presented in Figures 3(a)ndash3(d) and Figures4(a)ndash4(d) respectively For 120576 these parameters behave verysimilarly and therefore the corresponding diagrams are notpresented
When the contaminant is absent (120575 = 0 or 120576 = plusmn1)the 120587
119863119862and 120587
119863119862values are close to the expected values
of 099 and 001 respectively because the discordancy testswere applied at the 99 confidence level (open circles inFigures 3(a)ndash3(d) and Figures 4(a)ndash4(d)) As 120575 changes from0 to about plusmn25 the 120587
119863119862values slightly increase from 099
to about 0996 for 119899 = 5 (Figure 3(a)) 0996 for 119899 = 10
(Figure 3(b)) 0994-0995 for 119899 = 15 (Figure 3(c)) and 0993-0994 for 119899 = 20 (Figure 3(d)) The 120587
119863119862values show the
complementary behavior (Figures 4(a)ndash4(d)) Because in thistype of events (119862) a legitimate extreme observation is beingtested our best desire is that the 120587
119863119862and 120587
119863119862values remain
close to the theoretical values of 099 and 001 respectivelyfor contaminant-absent events This is actually observed inFigures 3 and 4
6 The Scientific World Journal
Table2Non
Spurious
type
IIerrorp
robability(120587119863119862)p
aram
eter
forfou
rsinglee
xtremeo
utlierd
iscordancytests
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
50
09900
03000
00057
0990026
000
00058
00023
09900
05000
00057
000
020989994
000
00058
minus000
095
plusmn01
0989948
00000139
0989975
00000140
00027
0989950
00000140
00002
0989940
00000144
minus000
095
plusmn2
0977285
00000113
0977358
00000125
00075
0977290
00000111
00005
0977267
00000116
minus00018
5plusmn3
0961486
000
00148
0961632
000
00142
00152
0961497
000
00146
00012
0961459
000
00146
minus00028
5plusmn5
0891864
00000205
0892491
00000194
0070
0891901
00000200
000
430891823
00000208
minus000
465
plusmn6
0834635
000
00222
0835779
000
00231
0137
0834704
000
00226
00084
08346
02000
00225
minus00038
5plusmn7
07646
60000
00240
0766559
000
00257
0248
0764766
000
00250
00139
07646
61000
00237
000
025
plusmn8
0685336
00000253
0688202
00000255
0418
0685496
00000249
00234
0685401
00000247
00095
5plusmn10
0514122
00000280
0519336
00000282
101
05144
0700000292
0056
0514394
00000287
0053
5plusmn12
035110
3000
00258
0358468
000
00275
210
0351500
000
00257
0113
0351629
000
00258
0150
5plusmn15
0165615
000
00194
01740
53000
00215
51
0166070
000
00196
0275
0166367
000
00196
045
5plusmn18
0062959
000
00126
0069566
000
00148
105
0063321
000
00127
058
0063618
000
00134
105
5plusmn20
0029322
000
00082
0033997
000
00089
159
0029577
000
00084
087
0029815
000
00081
168
100
09900
08000
00057
09900
41000
00057
00034
09900
03000
00054
minus000
060990025
000
00059
00017
10plusmn01
0989851
00000197
0989888
00000173
00037
0989844
00000207
minus000
060989869
000
00195
00019
10plusmn1
0978810
00000183
0979321
00000174
0052
0978911
00000179
00103
0978879
00000188
00070
10plusmn2
0949169
00000168
0951501
00000163
0246
0949704
00000165
0056
0949433
00000175
00279
10plusmn3
0878974
00000199
0886717
00000192
088
0880804
00000197
0208
0879762
00000202
0090
10plusmn4
0736953
000
00241
0756911
000
00242
271
0741679
000
00266
064
0738849
000
00253
0257
10plusmn5
0523889
000
00273
0561084
000
00257
710532523
000
00263
165
0527197
000
00266
063
10plusmn6
0300940
000
00248
0349468
000
00258
161
0311799
000
00260
361
0304921
000
00249
132
10plusmn7
0136560
000
00182
0181585
000
00226
330
0146058
000
00174
700139899
000
00185
244
10plusmn8
004
8488
00000122
0079155
00000144
6300544
2200000118
122
0050495
00000130
414
10plusmn9
0013431
000
00074
0029234
000
00092
1180016139
000
00075
202
0014300
000
00074
65
10plusmn10
0002900
000
00031
000
9248
000
00056
219
0003821
000
00036
318
0003180
000
00030
9610
plusmn11
000
0489
000
00013
0002529
000
00029
418
000
0726
000
00016
49000
0556
000
00014
138
150
0989998
0000006
0990015
00000062
00017
098998
00000061
minus00018
0989989
0000006
minus00010
15plusmn01
0989768
00000204
0989811
00000210
00044
0989766
00000209
minus00002
0989759
00000200
minus000
0915
plusmn05
0985307
00000236
0985665
00000202
00364
0985507
00000249
00204
0985345
00000240
00039
15plusmn2
0974943
00000225
0976229
00000186
0132
0975702
00000241
0078
0975113
00000236
00174
15plusmn5
0932551
000
00202
0938851
000
00184
068
09360
96000
00207
0380
0933417
000
00206
0093
15plusmn4
0827672
000
00245
0848433
000
00236
251
0838586
000
00251
132
0830350
000
00236
0324
15plusmn45
0621647
000
00288
0669717
000
00270
77064
4852
000
00332
373
0627283
000
00285
091
15plusmn5
0487591
000
00276
0550018
000
00319
128
0516243
000
00322
59
0494476
000
00290
141
15plusmn5
0351577
000
00269
0423601
000
00266
205
0382834
000
00290
89
0358972
000
00287
210
15plusmn6
0137443
000
00198
0204194
000
00211
490162735
000
00196
184
0143150
000
00201
415
15plusmn7
0035757
000
00093
0074779
00000144
109
004
8090
00000112
345
0038352
00000098
7315
plusmn8
000
6107
000
00043
002118
4000
00094
249
000
9822
000
00054
61000
6819
000
0004
2117
15plusmn9
000
0682
000
00015
000
4748
000
00039
600
0001393
000
00019
104
000
0802
000
00017
176
The Scientific World Journal 7
Table2Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn10
000
0049
000
0000
4000
0858
000
00017
1640
000
0138
000
00007
180
000
0062
000
00005
257
200
0990017
000
00057
09900
04000
0006
minus00014
0990026
000
0006
000
0909900
04000
00059
minus00014
20plusmn01
0989743
00000259
0989771
00000248
00028
0989786
00000243
000
430989734
00000277
minus000
0920
plusmn05
09846
0000000243
0985127
00000258
0054
0985093
00000265
0050
09846
8200000246
00083
20plusmn1
0972754
00000237
0974729
00000256
0203
09744
3900000213
0173
0973084
00000225
00339
20plusmn2
0922859
000
00259
0932611
000
00219
106
0930457
000
00234
082
09244
63000
00266
0174
20plusmn3
0798422
000
00265
0829677
000
00230
391
0820494
000
00248
276
0803167
000
00283
059
20plusmn4
0561409
000
00283
0628839
000
00267
120
060
4043
000
00273
760570474
000
00288
161
20plusmn45
0415514
00000280
0498225
00000261
199
046
4616
00000262
118
0425784
00000285
247
20plusmn5
0276626
000
00244
0365289
000
00265
321
0325675
000
00297
177
0286659
000
00245
363
20plusmn6
0085535
000
00163
0152881
000
00198
790116
833
000
00185
366
0091506
000
00175
7020
plusmn7
0015809
000
00070
004
6048
000
00127
191
0026842
000
00087
700017699
000
00079
120
20plusmn8
0001716
000
00021
0010242
000
00053
500
0003921
000
00036
128
0002041
000
00025
189
20plusmn9
000
0109
000
0000
60001733
000
00024
1500
000
0366
000
00011
237
000
0139
000
00007
285
20plusmn10
000
0004
000
00001
000
0229
000
00010
5600
000
0022
000
00002
450
000
0006
000
00001
413
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 5
Table 1 Sample simulation and test outcome (modified after Hayes and Kinsella [7])
Discordancy testresult
Contaminant 119909119888(119872 replications)
Present (120575 gt 0 120575 lt 0 120576 gt 1 or 120576 lt minus1)Absent
(120575 = 0 or 120576 = 1)Contaminant 119909119888not used in a
discordancy test(119862 type event)
Contaminant 119909119888used in a
discordancy test(119862 type event)
Not significant(TN le CV99)
Spurious type II errorprobability (120587
119863119862)
(Figure 3)
Nonspurious type II errorprobability (120587
119863119862)
(Figures 5 and 7)
True negative probability(120587119863)
(Figures 3 5 and 7)
Significant(TN gt CV99)
Spurious power probability(120587119863119862
)(Figure 4)
Nonspurious power probability(120587119863119862
)(Figures 6 and 8)
Type I error probability(120587119863)
(Figures 4 6 and 8)TN calculated test statistic for a sample (TN2 TN8 TN14 or TN15) CV99 critical value for a given discordancy test at 99 confidence level
function of the replications (119872 = 100 000 to 20000000) forN2 N8 N14 and N15 AlthoughΩmean values remain prac-tically constant (within the confidence limits of the mean)for replications of about 8000000 still higher replicationsof 20000000 (Figures 1 and 2) were used in all simulationexperiments
SimilarlyΩ for all four tests as a function of replications isalso shown in Figures 2(a)ndash2(d) which allows a visual com-parison of this performance parameter for different samplesizes and 120575 values Error bars (119906
99) for the 102 simulation
experiments are not shown for simplicity but for replicationslarger than 10000000 they were certainly within the size ofthe symbols The replications of 20000000 routinely usedfor comparing the performance of discordancy tests clearlyshow that the differences amongΩ values (Figures 2(a)ndash2(d))are statistically significant at a high confidence level thatis these differences are much greater than the simulationerrors
Alternatively following Krishnamoorthy and Lian [17]the simulation error for the replications of 20000000 usedroutinely in our work can be estimated approximately as2 times radic05 times 0520000000 = 000022
Because we carried out 102 independent simulationexperiments each with 20000000 replications our sim-ulation errors were even less than the above value Thusthe Monte Carlo simulations can be considered highly pre-cise They can also be said to be highly accurate becauseour procedure was modified after the highly precise andaccurate method of Verma and Quiroz-Ruiz [21] Theseauthors had shown high precision and accuracy of each119880(0 1) and 119873(0 1) experiments and had also applied allkinds of simulated data quality tests suggested by Lawand Kelton [25] Besides in the present work a largenumber of such experiments (204 streams of 119880(0 1) and102 streams of 119873(0 1)) have been carried out Thereforeas an innovation in Monte Carlo simulations we presentthe mean (119909) values as well as the total uncertainty (119906
99)
of 102 independent experiments in terms of the confi-dence interval of the mean at the strict 99 confidencelevel
Finally in order to evaluate the test performance test N2was used as a reference and differences inmean (Δ119909N119895) valuesof the other three tests were calculated as
Δ119909N119895 = (119909N119895 minus 119909N2
119909N2) times 100 (7)
where the subscript N119895 stands for N8 N14 or N15
6 Results and Discussion
61 119862 Type and Contaminant-Absent Events According toBarnett and Lewis [1] this type of events is of no majorconcern because the contaminant 119909
119888occupies an inner
position in the ordered array and the extremeobservation119909(119899)
or 119909(1)
under evaluation from discordancy tests is a legitimateobservation An inner position of the contaminant wouldaffect much less the sample mean and standard deviation [1]For small values of 120575 or 120576 close to 0 or plusmn1 respectively mostevents generated from the Monte Carlo simulation are of 119862type The 120587
119863119862and 120587
119863119862values for 119899 = 5 to 119899 = 20 as a
function of 120575 are presented in Figures 3(a)ndash3(d) and Figures4(a)ndash4(d) respectively For 120576 these parameters behave verysimilarly and therefore the corresponding diagrams are notpresented
When the contaminant is absent (120575 = 0 or 120576 = plusmn1)the 120587
119863119862and 120587
119863119862values are close to the expected values
of 099 and 001 respectively because the discordancy testswere applied at the 99 confidence level (open circles inFigures 3(a)ndash3(d) and Figures 4(a)ndash4(d)) As 120575 changes from0 to about plusmn25 the 120587
119863119862values slightly increase from 099
to about 0996 for 119899 = 5 (Figure 3(a)) 0996 for 119899 = 10
(Figure 3(b)) 0994-0995 for 119899 = 15 (Figure 3(c)) and 0993-0994 for 119899 = 20 (Figure 3(d)) The 120587
119863119862values show the
complementary behavior (Figures 4(a)ndash4(d)) Because in thistype of events (119862) a legitimate extreme observation is beingtested our best desire is that the 120587
119863119862and 120587
119863119862values remain
close to the theoretical values of 099 and 001 respectivelyfor contaminant-absent events This is actually observed inFigures 3 and 4
6 The Scientific World Journal
Table2Non
Spurious
type
IIerrorp
robability(120587119863119862)p
aram
eter
forfou
rsinglee
xtremeo
utlierd
iscordancytests
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
50
09900
03000
00057
0990026
000
00058
00023
09900
05000
00057
000
020989994
000
00058
minus000
095
plusmn01
0989948
00000139
0989975
00000140
00027
0989950
00000140
00002
0989940
00000144
minus000
095
plusmn2
0977285
00000113
0977358
00000125
00075
0977290
00000111
00005
0977267
00000116
minus00018
5plusmn3
0961486
000
00148
0961632
000
00142
00152
0961497
000
00146
00012
0961459
000
00146
minus00028
5plusmn5
0891864
00000205
0892491
00000194
0070
0891901
00000200
000
430891823
00000208
minus000
465
plusmn6
0834635
000
00222
0835779
000
00231
0137
0834704
000
00226
00084
08346
02000
00225
minus00038
5plusmn7
07646
60000
00240
0766559
000
00257
0248
0764766
000
00250
00139
07646
61000
00237
000
025
plusmn8
0685336
00000253
0688202
00000255
0418
0685496
00000249
00234
0685401
00000247
00095
5plusmn10
0514122
00000280
0519336
00000282
101
05144
0700000292
0056
0514394
00000287
0053
5plusmn12
035110
3000
00258
0358468
000
00275
210
0351500
000
00257
0113
0351629
000
00258
0150
5plusmn15
0165615
000
00194
01740
53000
00215
51
0166070
000
00196
0275
0166367
000
00196
045
5plusmn18
0062959
000
00126
0069566
000
00148
105
0063321
000
00127
058
0063618
000
00134
105
5plusmn20
0029322
000
00082
0033997
000
00089
159
0029577
000
00084
087
0029815
000
00081
168
100
09900
08000
00057
09900
41000
00057
00034
09900
03000
00054
minus000
060990025
000
00059
00017
10plusmn01
0989851
00000197
0989888
00000173
00037
0989844
00000207
minus000
060989869
000
00195
00019
10plusmn1
0978810
00000183
0979321
00000174
0052
0978911
00000179
00103
0978879
00000188
00070
10plusmn2
0949169
00000168
0951501
00000163
0246
0949704
00000165
0056
0949433
00000175
00279
10plusmn3
0878974
00000199
0886717
00000192
088
0880804
00000197
0208
0879762
00000202
0090
10plusmn4
0736953
000
00241
0756911
000
00242
271
0741679
000
00266
064
0738849
000
00253
0257
10plusmn5
0523889
000
00273
0561084
000
00257
710532523
000
00263
165
0527197
000
00266
063
10plusmn6
0300940
000
00248
0349468
000
00258
161
0311799
000
00260
361
0304921
000
00249
132
10plusmn7
0136560
000
00182
0181585
000
00226
330
0146058
000
00174
700139899
000
00185
244
10plusmn8
004
8488
00000122
0079155
00000144
6300544
2200000118
122
0050495
00000130
414
10plusmn9
0013431
000
00074
0029234
000
00092
1180016139
000
00075
202
0014300
000
00074
65
10plusmn10
0002900
000
00031
000
9248
000
00056
219
0003821
000
00036
318
0003180
000
00030
9610
plusmn11
000
0489
000
00013
0002529
000
00029
418
000
0726
000
00016
49000
0556
000
00014
138
150
0989998
0000006
0990015
00000062
00017
098998
00000061
minus00018
0989989
0000006
minus00010
15plusmn01
0989768
00000204
0989811
00000210
00044
0989766
00000209
minus00002
0989759
00000200
minus000
0915
plusmn05
0985307
00000236
0985665
00000202
00364
0985507
00000249
00204
0985345
00000240
00039
15plusmn2
0974943
00000225
0976229
00000186
0132
0975702
00000241
0078
0975113
00000236
00174
15plusmn5
0932551
000
00202
0938851
000
00184
068
09360
96000
00207
0380
0933417
000
00206
0093
15plusmn4
0827672
000
00245
0848433
000
00236
251
0838586
000
00251
132
0830350
000
00236
0324
15plusmn45
0621647
000
00288
0669717
000
00270
77064
4852
000
00332
373
0627283
000
00285
091
15plusmn5
0487591
000
00276
0550018
000
00319
128
0516243
000
00322
59
0494476
000
00290
141
15plusmn5
0351577
000
00269
0423601
000
00266
205
0382834
000
00290
89
0358972
000
00287
210
15plusmn6
0137443
000
00198
0204194
000
00211
490162735
000
00196
184
0143150
000
00201
415
15plusmn7
0035757
000
00093
0074779
00000144
109
004
8090
00000112
345
0038352
00000098
7315
plusmn8
000
6107
000
00043
002118
4000
00094
249
000
9822
000
00054
61000
6819
000
0004
2117
15plusmn9
000
0682
000
00015
000
4748
000
00039
600
0001393
000
00019
104
000
0802
000
00017
176
The Scientific World Journal 7
Table2Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn10
000
0049
000
0000
4000
0858
000
00017
1640
000
0138
000
00007
180
000
0062
000
00005
257
200
0990017
000
00057
09900
04000
0006
minus00014
0990026
000
0006
000
0909900
04000
00059
minus00014
20plusmn01
0989743
00000259
0989771
00000248
00028
0989786
00000243
000
430989734
00000277
minus000
0920
plusmn05
09846
0000000243
0985127
00000258
0054
0985093
00000265
0050
09846
8200000246
00083
20plusmn1
0972754
00000237
0974729
00000256
0203
09744
3900000213
0173
0973084
00000225
00339
20plusmn2
0922859
000
00259
0932611
000
00219
106
0930457
000
00234
082
09244
63000
00266
0174
20plusmn3
0798422
000
00265
0829677
000
00230
391
0820494
000
00248
276
0803167
000
00283
059
20plusmn4
0561409
000
00283
0628839
000
00267
120
060
4043
000
00273
760570474
000
00288
161
20plusmn45
0415514
00000280
0498225
00000261
199
046
4616
00000262
118
0425784
00000285
247
20plusmn5
0276626
000
00244
0365289
000
00265
321
0325675
000
00297
177
0286659
000
00245
363
20plusmn6
0085535
000
00163
0152881
000
00198
790116
833
000
00185
366
0091506
000
00175
7020
plusmn7
0015809
000
00070
004
6048
000
00127
191
0026842
000
00087
700017699
000
00079
120
20plusmn8
0001716
000
00021
0010242
000
00053
500
0003921
000
00036
128
0002041
000
00025
189
20plusmn9
000
0109
000
0000
60001733
000
00024
1500
000
0366
000
00011
237
000
0139
000
00007
285
20plusmn10
000
0004
000
00001
000
0229
000
00010
5600
000
0022
000
00002
450
000
0006
000
00001
413
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
6 The Scientific World Journal
Table2Non
Spurious
type
IIerrorp
robability(120587119863119862)p
aram
eter
forfou
rsinglee
xtremeo
utlierd
iscordancytests
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
50
09900
03000
00057
0990026
000
00058
00023
09900
05000
00057
000
020989994
000
00058
minus000
095
plusmn01
0989948
00000139
0989975
00000140
00027
0989950
00000140
00002
0989940
00000144
minus000
095
plusmn2
0977285
00000113
0977358
00000125
00075
0977290
00000111
00005
0977267
00000116
minus00018
5plusmn3
0961486
000
00148
0961632
000
00142
00152
0961497
000
00146
00012
0961459
000
00146
minus00028
5plusmn5
0891864
00000205
0892491
00000194
0070
0891901
00000200
000
430891823
00000208
minus000
465
plusmn6
0834635
000
00222
0835779
000
00231
0137
0834704
000
00226
00084
08346
02000
00225
minus00038
5plusmn7
07646
60000
00240
0766559
000
00257
0248
0764766
000
00250
00139
07646
61000
00237
000
025
plusmn8
0685336
00000253
0688202
00000255
0418
0685496
00000249
00234
0685401
00000247
00095
5plusmn10
0514122
00000280
0519336
00000282
101
05144
0700000292
0056
0514394
00000287
0053
5plusmn12
035110
3000
00258
0358468
000
00275
210
0351500
000
00257
0113
0351629
000
00258
0150
5plusmn15
0165615
000
00194
01740
53000
00215
51
0166070
000
00196
0275
0166367
000
00196
045
5plusmn18
0062959
000
00126
0069566
000
00148
105
0063321
000
00127
058
0063618
000
00134
105
5plusmn20
0029322
000
00082
0033997
000
00089
159
0029577
000
00084
087
0029815
000
00081
168
100
09900
08000
00057
09900
41000
00057
00034
09900
03000
00054
minus000
060990025
000
00059
00017
10plusmn01
0989851
00000197
0989888
00000173
00037
0989844
00000207
minus000
060989869
000
00195
00019
10plusmn1
0978810
00000183
0979321
00000174
0052
0978911
00000179
00103
0978879
00000188
00070
10plusmn2
0949169
00000168
0951501
00000163
0246
0949704
00000165
0056
0949433
00000175
00279
10plusmn3
0878974
00000199
0886717
00000192
088
0880804
00000197
0208
0879762
00000202
0090
10plusmn4
0736953
000
00241
0756911
000
00242
271
0741679
000
00266
064
0738849
000
00253
0257
10plusmn5
0523889
000
00273
0561084
000
00257
710532523
000
00263
165
0527197
000
00266
063
10plusmn6
0300940
000
00248
0349468
000
00258
161
0311799
000
00260
361
0304921
000
00249
132
10plusmn7
0136560
000
00182
0181585
000
00226
330
0146058
000
00174
700139899
000
00185
244
10plusmn8
004
8488
00000122
0079155
00000144
6300544
2200000118
122
0050495
00000130
414
10plusmn9
0013431
000
00074
0029234
000
00092
1180016139
000
00075
202
0014300
000
00074
65
10plusmn10
0002900
000
00031
000
9248
000
00056
219
0003821
000
00036
318
0003180
000
00030
9610
plusmn11
000
0489
000
00013
0002529
000
00029
418
000
0726
000
00016
49000
0556
000
00014
138
150
0989998
0000006
0990015
00000062
00017
098998
00000061
minus00018
0989989
0000006
minus00010
15plusmn01
0989768
00000204
0989811
00000210
00044
0989766
00000209
minus00002
0989759
00000200
minus000
0915
plusmn05
0985307
00000236
0985665
00000202
00364
0985507
00000249
00204
0985345
00000240
00039
15plusmn2
0974943
00000225
0976229
00000186
0132
0975702
00000241
0078
0975113
00000236
00174
15plusmn5
0932551
000
00202
0938851
000
00184
068
09360
96000
00207
0380
0933417
000
00206
0093
15plusmn4
0827672
000
00245
0848433
000
00236
251
0838586
000
00251
132
0830350
000
00236
0324
15plusmn45
0621647
000
00288
0669717
000
00270
77064
4852
000
00332
373
0627283
000
00285
091
15plusmn5
0487591
000
00276
0550018
000
00319
128
0516243
000
00322
59
0494476
000
00290
141
15plusmn5
0351577
000
00269
0423601
000
00266
205
0382834
000
00290
89
0358972
000
00287
210
15plusmn6
0137443
000
00198
0204194
000
00211
490162735
000
00196
184
0143150
000
00201
415
15plusmn7
0035757
000
00093
0074779
00000144
109
004
8090
00000112
345
0038352
00000098
7315
plusmn8
000
6107
000
00043
002118
4000
00094
249
000
9822
000
00054
61000
6819
000
0004
2117
15plusmn9
000
0682
000
00015
000
4748
000
00039
600
0001393
000
00019
104
000
0802
000
00017
176
The Scientific World Journal 7
Table2Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn10
000
0049
000
0000
4000
0858
000
00017
1640
000
0138
000
00007
180
000
0062
000
00005
257
200
0990017
000
00057
09900
04000
0006
minus00014
0990026
000
0006
000
0909900
04000
00059
minus00014
20plusmn01
0989743
00000259
0989771
00000248
00028
0989786
00000243
000
430989734
00000277
minus000
0920
plusmn05
09846
0000000243
0985127
00000258
0054
0985093
00000265
0050
09846
8200000246
00083
20plusmn1
0972754
00000237
0974729
00000256
0203
09744
3900000213
0173
0973084
00000225
00339
20plusmn2
0922859
000
00259
0932611
000
00219
106
0930457
000
00234
082
09244
63000
00266
0174
20plusmn3
0798422
000
00265
0829677
000
00230
391
0820494
000
00248
276
0803167
000
00283
059
20plusmn4
0561409
000
00283
0628839
000
00267
120
060
4043
000
00273
760570474
000
00288
161
20plusmn45
0415514
00000280
0498225
00000261
199
046
4616
00000262
118
0425784
00000285
247
20plusmn5
0276626
000
00244
0365289
000
00265
321
0325675
000
00297
177
0286659
000
00245
363
20plusmn6
0085535
000
00163
0152881
000
00198
790116
833
000
00185
366
0091506
000
00175
7020
plusmn7
0015809
000
00070
004
6048
000
00127
191
0026842
000
00087
700017699
000
00079
120
20plusmn8
0001716
000
00021
0010242
000
00053
500
0003921
000
00036
128
0002041
000
00025
189
20plusmn9
000
0109
000
0000
60001733
000
00024
1500
000
0366
000
00011
237
000
0139
000
00007
285
20plusmn10
000
0004
000
00001
000
0229
000
00010
5600
000
0022
000
00002
450
000
0006
000
00001
413
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 7
Table2Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn10
000
0049
000
0000
4000
0858
000
00017
1640
000
0138
000
00007
180
000
0062
000
00005
257
200
0990017
000
00057
09900
04000
0006
minus00014
0990026
000
0006
000
0909900
04000
00059
minus00014
20plusmn01
0989743
00000259
0989771
00000248
00028
0989786
00000243
000
430989734
00000277
minus000
0920
plusmn05
09846
0000000243
0985127
00000258
0054
0985093
00000265
0050
09846
8200000246
00083
20plusmn1
0972754
00000237
0974729
00000256
0203
09744
3900000213
0173
0973084
00000225
00339
20plusmn2
0922859
000
00259
0932611
000
00219
106
0930457
000
00234
082
09244
63000
00266
0174
20plusmn3
0798422
000
00265
0829677
000
00230
391
0820494
000
00248
276
0803167
000
00283
059
20plusmn4
0561409
000
00283
0628839
000
00267
120
060
4043
000
00273
760570474
000
00288
161
20plusmn45
0415514
00000280
0498225
00000261
199
046
4616
00000262
118
0425784
00000285
247
20plusmn5
0276626
000
00244
0365289
000
00265
321
0325675
000
00297
177
0286659
000
00245
363
20plusmn6
0085535
000
00163
0152881
000
00198
790116
833
000
00185
366
0091506
000
00175
7020
plusmn7
0015809
000
00070
004
6048
000
00127
191
0026842
000
00087
700017699
000
00079
120
20plusmn8
0001716
000
00021
0010242
000
00053
500
0003921
000
00036
128
0002041
000
00025
189
20plusmn9
000
0109
000
0000
60001733
000
00024
1500
000
0366
000
00011
237
000
0139
000
00007
285
20plusmn10
000
0004
000
00001
000
0229
000
00010
5600
000
0022
000
00002
450
000
0006
000
00001
413
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
8 The Scientific World Journal
Table3Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0020034
000
00163
0019986
000
00163
minus0243
0020031
000
00165
minus00191
0020053
000
00167
0092
5plusmn2
0028537
000
00125
0028456
000
00128
minus0283
0028531
000
00124
minus00224
0028563
000
00127
0089
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
075
plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
000
00247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
000
00196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
00000081
minus0051
10plusmn01
0020121
000
00216
0020050
000
00181
minus0352
0020131
00000218
0052
0020087
00000211
minus0169
10plusmn1
0029615
000
00189
0029124
000
00180
minus16
60029522
000
00185
minus0316
0029546
000
00193
minus0233
10plusmn2
0056182
00000169
0053967
000
00177
minus394
0055608
00000161
minus10
20056005
00000176
minus0314
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
000
00074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0020212
000
00214
0020157
000
00214
minus0272
0020236
000
00217
0117
0020233
000
00212
0106
15plusmn05
0024306
000
00241
0023963
000
00203
minus14
10024143
000
00259
minus067
0024284
00000250
minus0091
15plusmn1
0033660
000
00236
0032450
000
00192
minus360
0032982
000
00251
minus201
0033545
000
00245
minus0341
1525
0113807
000
00218
0102067
000
00195
minus103
0107680
000
00210
minus54
0113112
000
00226
minus061
15plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus48
064
1028
000
00287
minus114
15plusmn6
0862557
000
00198
0795806
000
00211
minus774
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 9
Table3Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0020222
000
00253
0020213
000
00242
minus004
60020175
000
00238
minus0237
0020248
000
00270
0126
20plusmn05
0025049
000
00251
0024566
000
00260
minus19
30024578
000
00267
minus18
80024992
000
00255
minus0227
20plusmn1
0036024
000
00249
0034154
000
00266
minus52
0034454
000
00228
minus436
0035781
000
00235
minus067
2025
0132180
000
00257
0113955
000
00238
minus138
0119
739
000
00238
minus94
0130861
000
00269
minus10
020
plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
204
0438591
000
00283
0374722
000
00286
minus146
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
10 The Scientific World Journal
Table4TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120575
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn01
0010052
000
00139
0010025
000
00140
minus0262
0010050
000
00140
minus00153
00100
60000
00144
0085
5plusmn2
0022715
000
00113
0022642
000
00125
minus0322
0022710
000
00111
minus00212
0022733
00000116
0079
5plusmn3
0038514
000
00148
0038368
000
00142
minus0380
0038503
000
00146
minus00291
0038541
00000146
0070
5plusmn4
0065920
000
00155
0065606
000
00154
minus048
0065900
000
00155
minus00309
0065955
000
00157
0054
5plusmn5
0108136
000
00205
0107509
000
00194
minus058
0108099
000
00200
minus00351
0108177
000
00208
00379
5plusmn6
0165365
000
00222
0164221
000
00231
minus069
0165296
000
00226
minus004
220165398
000
00225
00194
5plusmn7
0235340
000
00240
0233441
000
00257
minus081
0235234
000
00250
minus0045
0235339
000
00237
minus000
066
5plusmn8
03146
64000
00253
0311798
000
00255
minus091
0314504
000
00249
minus0051
0314599
00000247
minus00206
5plusmn10
0485878
000
00280
04806
64000
00282
minus10
70485593
000
00292
minus0059
0485606
00000287
minus0056
5plusmn1017
050
044
000
00279
049
503
000
0028
7minus10
8050
014
000
0028
0minus005
9050
015
000
0028
0minus005
85
plusmn12
064
8897
000
00258
064
1532
000
00275
minus114
064
8500
000
00257
minus0061
064
8371
000
00258
minus0081
5plusmn15
0834385
000
00194
0825947
000
00215
minus10
10833930
000
00196
minus0055
0833633
00000196
minus0090
5plusmn18
0937041
000
00126
0930434
000
00148
minus071
0936679
000
00127
minus00387
0936382
000
00134
minus0070
5plusmn20
0970678
00000082
09660
0300000089
minus048
09704
2300000084
minus00263
0970185
000
00081
minus0051
10plusmn01
0010149
000
00197
0010112
000
00173
minus0364
0010156
000
00207
0061
0010131
000
00195
minus0181
10plusmn1
002119
0000
00183
0020679
000
00174
minus241
0021089
000
00179
minus048
0021121
000
00188
minus0323
10plusmn2
0050831
00000168
004
8499
00000163
minus46
0050296
00000165
minus10
50050567
00000175
minus052
10plusmn3
0121026
000
00199
0113283
000
00192
minus64
0119
196
000
00197
minus15
10120238
000
00202
minus065
10plusmn4
0263047
000
00241
0243089
000
00242
minus76
0258321
000
00266
minus18
00261151
000
00253
minus072
10plusmn5
0476111
000
00273
0438916
000
00257
minus78
046
7477
000
00263
minus18
10472803
000
00266
minus069
10plusmn5105
050
0474
000
0028
5046
1588
000
0026
1minus78
049
1473
000
00273
minus18
0049
7048
000
00272
minus068
10plusmn6
0699060
000
00248
0650532
000
00258
minus69
0688201
000
00260
minus15
50695079
000
00249
minus057
10plusmn7
0863440
000
00182
0818415
000
00226
minus52
0853942
000
00174
minus110
0860101
000
00185
minus0387
10plusmn8
0951512
000
00122
0920845
000
00144
minus322
0945578
000
00118
minus062
0949505
000
00130
minus0211
10plusmn9
0986569
000
00074
0970766
00000092
minus16
00983861
00000075
minus0274
0985700
00000074
minus0088
10plusmn10
0997100
000
00031
0990752
000
00056
minus064
0996179
000
00036
minus0092
0996820
000
00030
minus00280
10plusmn11
0999511
000
00013
0997471
000
00029
minus0204
0999274
000
00016
minus00237
0999444
000
00014
minus000
6715
plusmn01
0010232
000
00204
0010189
000
00210
minus0425
0010234
000
00209
00204
0010241
000
00200
0089
15plusmn05
0014693
000
00236
0014335
000
00202
minus244
00144
93000
00249
minus13
70014655
000
00240
minus0260
15plusmn1
0025057
000
00225
0023771
000
00186
minus51
0024298
000
00241
minus303
0024887
000
00236
minus068
15plusmn25
0109047
000
00212
0097138
000
00185
minus109
0102549
000
00212
minus60
0107459
000
00213
minus14
615
plusmn3
0172328
000
00245
0151567
000
00236
minus120
0161414
000
00251
minus63
0169650
000
00236
minus15
515
plusmn4
0378353
000
00288
0330283
000
00270
minus127
0355148
000
00332
minus61
0372717
000
00285
minus14
915
plusmn446
050
1373
000
00279
043
9974
000
00313
minus122
04730
72000
00321
minus56
049
4564
000
0029
7minus13
615
plusmn5
064
8423
000
00269
0576399
000
00266
minus111
0617166
000
00290
minus482
064
1028
000
00287
minus114
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 11
Table4Con
tinued
119899120575
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn6
0862557
000
00198
0795806
000
00211
minus77
0837265
000
00196
minus293
0856850
000
00201
minus066
15plusmn7
0964243
000
00093
0925221
000
00144
minus405
0951910
00000112
minus12
80961648
000
00098
minus0269
15plusmn8
0993893
00000043
0978816
00000094
minus15
20990178
000
00054
minus0374
0993181
000
0004
2minus0072
15plusmn9
0999318
000
00015
0995252
000
00039
minus040
70998607
000
00019
minus0071
0999198
00000017
minus00120
15plusmn10
0999951
000
0000
40999142
000
00017
minus0081
0999862
000
00007
minus000
890999938
000
00005
minus000127
20plusmn01
0010257
000
00259
0010229
000
00248
minus0271
0010214
000
00243
minus0414
0010266
00000277
0086
20plusmn05
0015400
000
00243
0014873
000
00258
minus342
0014907
000
00265
minus320
0015318
000
00246
minus053
20plusmn1
0027246
000
00237
0025271
000
00256
minus72
0025561
000
00213
minus62
0026916
000
00225
minus12
120
plusmn25
01266
09000
00249
0108344
000
00227
minus144
0113053
000
00241
minus107
0123707
000
00263
minus229
20plusmn3
0201578
000
00265
0170323
000
00230
minus155
0179506
000
00248
minus109
0196833
000
00283
minus235
20plusmn4
0438591
000
00283
037116
1000
00267
minus154
0395957
000
00273
minus97
0429526
000
00288
minus207
20plusmn421
049
9301
000
00272
042
4653
000
0026
6minus150
0453296
000
0026
8minus92
048
9576
000
00279
minus19
520
plusmn5
0723374
000
00244
0634711
000
00265
minus123
0674325
00000297
minus68
0713341
000
00245
minus13
920
plusmn6
09144
65000
00163
0847119
000
00198
minus74
0883167
00000185
minus342
0908494
000
00175
minus065
20plusmn7
0984191
00000070
0953952
00000127
minus307
0973158
00000087
minus112
0982301
00000079
minus0192
20plusmn8
0998284
00000021
0989758
00000053
minus085
0996079
000
00036
minus0221
0997959
000
00025
minus00325
20plusmn9
0999891
000
0000
60998267
000
00024
minus0162
0999634
000
00011
minus00258
0999861
000
00007
minus000309
20plusmn10
0999996
000
00001
0999771
000
00010
minus00225
0999978
000
00002
minus000180
0999994
000
00001
minus000
017
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
12 The Scientific World Journal
Table5Po
wer
ofTest(Ω
)valuesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
0020850
000
00310
0020803
000
00314
minus0227
0020844
000
00307
minus00313
0020869
000
00314
0087
5plusmn3
006
4711
000
00419
006
4385
000
00419
minus050
006
4691
000
00425
minus00304
006
4743
00000
412
0050
5plusmn5
0154521
000
00496
0153274
000
00481
minus081
01544
49000
00483
minus004
60154511
000
00480
minus000
665
plusmn7
0258413
000
00570
0256108
000
00574
minus089
0258284
000
00555
minus0050
0258324
00000568
minus00344
5plusmn10
0397926
000
00695
0394635
000
00687
minus083
0397744
000
00695
minus004
60397724
000
00691
minus0051
5plusmn129
050
175
000
00610
049
817
000
0060
9minus071
050
155
000
0059
4minus003
89050
151
000
0062
0minus004
75
plusmn15
0560321
000
00690
0556782
000
00652
minus063
0560129
000
00682
minus00341
0560080
000
00690
minus004
295
plusmn20
066
0616
000
0060
00657405
000
0060
0minus049
066
0442
000
0060
0minus00264
066
0386
000
00588
minus00347
5plusmn30
0771476
000
00629
0769007
000
00655
minus0320
0771337
000
00637
minus00181
0771296
000
0064
4minus00233
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0023252
000
0040
00023052
000
00335
minus086
0023228
000
0040
0minus0106
0023210
000
00386
minus0181
10plusmn3
02300
07000
00688
0215369
000
00615
minus64
0226613
000
00671
minus14
80228723
000
00691
minus056
10plusmn5
046
6178
000
00675
044
4482
000
00696
minus47
0461346
000
00693
minus10
37046
4412
000
00674
minus0379
10plusmn54
050
1202
000
0066
8047
9497
000
0068
2minus433
049
6385
000
0062
9minus096
049
9458
000
00674
minus034
810
plusmn10
060
9736
000
00680
0589504
000
0064
2minus332
0605299
000
0066
8minus073
0608161
000
00726
minus0258
10plusmn10
0729058
000
00551
0712878
000
00551
minus222
0725561
000
00548
minus048
0727835
000
00575
minus0168
10plusmn15
0824592
000
00517
0813173
000
00531
minus13
80822123
000
00498
minus0299
0823755
000
00530
minus0101
10plusmn20
0872018
000
00545
0863336
000
00533
minus10
00870142
000
00543
minus0215
0871397
000
00552
minus0071
10plusmn30
0918594
000
00550
0912810
000
00579
minus063
0917335
000
00538
minus0137
0918194
000
00540
minus00435
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
000
00184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
000
00161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0024941
000
00507
0024516
000
00434
minus17
00024714
000
00454
minus091
0024907
000
00523
minus0137
15plusmn3
0315624
000
00816
0285482
000
00707
minus96
0301908
000
00791
minus435
0312616
000
00768
minus095
15plusmn44
050
5574
000
00745
047
0626
000
0067
7minus69
049
0624
000
0069
7minus296
0502393
000
0070
5minus063
15plusmn5
0563420
000
00689
0529313
000
00721
minus61
0549047
000
00748
minus255
05604
07000
00698
minus053
15plusmn7
0692305
000
00627
0663682
000
00610
minus413
0680599
000
00638
minus16
90689963
000
00629
minus0338
15plusmn10
0791754
000
00591
0770289
000
00590
minus271
0783169
000
00580
minus10
80790166
000
00564
minus0201
15plusmn15
0867777
000
00506
0853207
000
00541
minus16
80862030
000
00516
minus066
0866870
000
00525
minus0105
15plusmn20
0904654
000
00486
0893754
000
00511
minus12
00900431
000
00428
minus047
0904100
000
00549
minus0061
15plusmn30
0940391
000
0044
80933225
000
00533
minus076
0937686
000
00478
minus0288
0940194
000
00497
minus00209
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0229
0949688
000
00258
minus0053
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 13
Table5Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
00000200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
00000137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0025959
000
0060
00025296
000
00570
minus2551
0025396
000
00534
minus217
0025877
000
00610
minus0313
20plusmn3
0362436
000
00753
0321879
000
00782
minus112
0338086
000
00708
minus67
0357843
000
00754
minus12
720
plusmn4
050
9029
000
0067
7046
5277
000
0066
1minus86
048
4093
000
0069
1minus49
050
4506
000
0064
9minus089
20plusmn5
060
9022
000
00669
0567957
000
00706
minus67
0586331
000
00707
minus373
0605042
000
00639
minus065
20plusmn7
0728947
000
00692
0695836
000
00693
minus45
0711240
000
0064
0minus243
0726084
000
00696
minus0393
20plusmn10
0818758
000
00462
07944
79000
00533
minus297
0806135
000
00486
minus15
40816978
000
00476
minus0217
20plusmn15
0886139
00000424
0869861
00000430
minus18
40877932
000
0046
8minus093
0885261
00000
465
minus0099
20plusmn20
0918508
000
00399
0906382
000
0044
6minus13
20912559
000
00407
minus065
0918085
000
00427
minus004
620
plusmn30
0949667
000
00432
0941715
000
00507
minus084
0945966
000
00481
minus0390
0949717
000
00506
00053
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
100000210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
00000170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
00000128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
14 The Scientific World Journal
Table6TestPerfo
rmance
Criterio
nP5
(120587119863|119862)v
aluesfor
four
singlee
xtremeo
utlierd
iscordancytests
asafun
ctionof
120576
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
5plusmn11
001110
6000
00274
0011080
000
00273
minus0237
001110
1000
00274
minus004
250011115
00000280
0076
5plusmn3
0056787
00000393
0056474
00000387
minus055
0056768
00000397
minus00321
0056810
000
00382
004
085
plusmn5
0146913
000
00462
0145680
000
00452
minus084
0146842
000
00447
minus004
80146893
000
00452
minus00131
5plusmn7
0250890
000
00523
0248598
000
00537
minus091
0250762
000
00512
minus0051
0250790
00000524
minus00397
5plusmn10
03904
62000
00624
0387183
000
00608
minus084
0390280
000
00627
minus0047
0390249
000
00624
minus0055
5plusmn131
050
043
000
0054
4049
686
000
00559
minus071
050
023
000
0054
4minus004
06050
018
000
00553
minus005
05
plusmn15
0552883
000
00554
0549351
000
00535
minus064
0552692
000
00547
minus00346
0552630
000
00551
minus004
65
plusmn20
0653179
000
00491
0649982
000
00501
minus049
0653009
000
00491
minus00262
0652941
000
00480
minus00365
5plusmn30
07640
71000
00476
0761610
000
0046
4minus0322
0763935
000
00480
minus00178
0763882
000
00487
minus00248
5plusmn40
0822288
000
00430
0820357
000
0044
2minus0235
0822181
000
00431
minus00130
0822138
000
00428
minus00182
5plusmn60
0881517
00000389
0880180
00000385
minus0152
0881443
00000392
minus00084
0881408
00000394
minus00124
5plusmn80
0911274
000
00321
0910263
000
00327
minus0111
0911218
000
00324
minus000
61091119
5000
00311
minus00087
5plusmn120
0941005
000
00307
0940328
000
00307
minus0072
0940970
00000306
minus000371
0940951
000
00306
minus00058
5plusmn160
0955830
000
00281
0955330
000
00269
minus0052
0955801
00000285
minus000302
0955792
000
00281
minus000395
5plusmn200
0964715
000
00244
0964312
000
00248
minus00417
0964693
000
00241
minus000226
096
4681
000
00242
minus000345
10plusmn11
0013593
000
00385
0013418
000
00340
minus12
90013562
000
00384
minus0227
0013564
000
00378
minus0209
10plusmn3
0222304
000
00659
0207623
000
00591
minus66
0218930
000
00661
minus15
20220981
000
00656
minus060
10plusmn5
0458753
000
00661
0437008
000
0066
8minus47
0453946
000
00689
minus10
50456939
000
00667
minus0395
10plusmn541
049
9554
000
0066
8047
7805
000
0068
2minus435
049
4764
000
0062
9minus096
049
776
000
00674
minus0359
10plusmn7
0602388
000
00651
0582111
000
00625
minus337
0597979
000
00649
minus073
060
0766
000
00684
minus0269
10plusmn10
0721760
000
00501
0705530
000
00527
minus225
0718290
000
00492
minus048
07204
82000
00517
minus0177
10plusmn15
0817312
000
00418
0805848
000
00426
minus14
00814868
000
00429
minus0299
08164
20000
00412
minus0109
10plusmn20
0864739
000
00433
0856024
000
0040
8minus10
10862890
000
00426
minus0214
0864058
000
00439
minus0079
10plusmn30
0911335
000
00299
0905514
000
00319
minus064
09100
96000
00294
minus0136
0910883
000
00309
minus0050
10plusmn40
0934126
000
00272
0929776
000
00294
minus047
0933203
000
00279
minus0099
0933789
000
00276
minus00360
10plusmn50
0947611
000
00234
0944143
000
00241
minus0366
0946875
000
00242
minus0078
0947342
00000238
minus00284
10plusmn100
0974142
00000183
0972421
00000184
minus0177
0973777
00000185
minus00374
0974010
00000184
minus00136
10plusmn150
0982843
00000161
0981700
00000168
minus0116
0982600
00000166
minus00247
0982756
00000161
minus00089
10plusmn200
0987171
00000142
0986318
00000144
minus0086
0986991
00000138
minus00183
0987105
00000143
minus000
6715
plusmn11
0015232
000
00509
0014804
000
00414
minus281
0014971
000
0044
2minus17
10015180
000
00518
minus0340
15plusmn3
0307670
000
00774
0277454
000
00694
minus98
0293796
000
00758
minus45
0304363
000
00734
minus10
815
plusmn45
050
8306
000
0069
30473352
000
0062
7minus69
049
3247
000
0068
6minus296
050
4762
000
00674
minus070
15plusmn5
0555712
000
0064
40521524
000
00708
minus62
054114
7000
00721
minus262
0552297
000
0066
4minus061
15plusmn7
06846
63000
00590
0655963
000
00589
minus419
0672765
00000603
minus174
0681886
000
00602
minus040
615
plusmn10
0784156
000
00518
0762607
000
00574
minus275
0775363
000
00548
minus112
0782114
000
00522
minus0260
15plusmn15
0860198
000
00416
0845545
000
0044
4minus17
00854249
000
00423
minus069
0858831
000
00422
minus0159
15plusmn20
0897091
00000348
0886105
00000350
minus12
20892656
000
00352
minus049
08960
63000
00350
minus0115
15plusmn30
0932834
000
00271
0925572
000
00285
minus078
0929908
000
00291
minus0314
0932159
000
00271
minus0072
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 15
Table6Con
tinued
119899120576
Disc
ordancytests
N2
N8
N14
N15
119909N2
(11990699) N
2119909N8
(11990699) N
8Δ119909N8
119909N14
(11990699) N
14Δ119909N14
119909N15
(11990699) N
15Δ119909N15
15plusmn40
0950189
000
00256
0944782
000
00263
minus057
0948017
000
00256
minus0228
0949688
000
00258
minus0053
15plusmn60
0967191
000
00212
0963613
000
00224
minus0370
0965751
000
00218
minus0149
0966859
000
00207
minus00343
15plusmn80
0975546
00000199
0972878
00000213
minus0274
09744
7500000198
minus0110
0975299
000
00200
minus00253
15plusmn140
0986132
00000140
0984621
00000148
minus0153
0985524
000
00143
minus0062
0985992
000
00137
minus00142
15plusmn200
0990334
00000113
0989278
00000127
minus0107
0989908
00000116
minus004
290990235
000
00116
minus000
9920
plusmn11
0016228
000
00577
0015531
000
00546
minus429
0015647
000
00516
minus358
0016119
00000580
minus067
20plusmn3
0354224
000
00726
0313612
000
00769
minus115
0329474
000
00698
minus70
0349056
000
00732
minus14
620
plusmn4
050
0954
000
0063
9045
7154
000
00653
minus87
0475548
000
0065
1minus51
049
5727
000
0062
0minus10
420
plusmn5
0601015
000
00617
0559904
000
00661
minus68
0577809
000
00665
minus386
0596262
000
00581
minus079
20plusmn7
0721004
000
00628
0687842
000
00622
minus46
0702741
000
00577
minus253
0717289
000
00637
minus052
20plusmn10
0810844
000
0044
80786525
000
0046
8minus30
0797652
000
00469
minus16
30808171
000
00433
minus0330
20plusmn15
0878244
000
00370
0861918
000
00398
minus18
60869456
000
00419
minus10
008764
61000
00382
minus0203
20plusmn20
0910617
000
00332
0898443
000
00353
minus13
409040
86000
00333
minus072
0909299
000
00325
minus0145
20plusmn30
0941779
000
00246
0933781
000
00270
minus085
0937486
000
00248
minus046
0940911
000
00255
minus0092
20plusmn40
0956868
000
00229
0950924
000
00258
minus062
0953687
000
00219
minus0332
0956225
000
00226
minus0067
20plusmn60
0971612
00000209
0967685
00000218
minus040
40969514
000
00209
minus0216
097119
1000
00210
minus00434
20plusmn80
0978856
00000170
0975928
00000185
minus0299
0977289
00000179
minus0160
0978541
000
00170
minus00322
20plusmn140
0988015
00000127
0986356
00000127
minus0168
0987127
00000139
minus0090
0987837
000
00128
minus00181
20plusmn200
0991646
000
00107
09904
87000
00113
minus0117
0991025
000
00109
minus0063
0991520
000
00113
minus00126
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
16 The Scientific World Journal
0480
0482
0484
0486
Ω
0 5E6 1E7 15E7 2E7
M
(a) 119899 = 5 and 120575 = 10
0966
0968
0970
0972
0 5E6 1E7 15E7 2E7
M
Ω
(b) 119899 = 5 and 120575 = 20
056
058
060
062
064
066
Ω
0 5E6 1E7 15E7 2E7
M
N2N8
N14N15
(c) 119899 = 15 and 120575 = 5
N2N8
N14N15
0 5E6 1E7 15E7 2E7
M
10002
09999
09996
09993
09990
Ω
(d) 119899 = 15 and 120575 = 10
Figure 2 Determination of optimum simulation replication (119872) for Power of Test (Ω) as a function of replications for all tests N2 N8 N14and N15 symbols are explained in each figure (a) Sample size 119899 = 5 and contaminant parameter 120575 = 10 (b) 119899 = 5 and 120575 = 20 (c) 119899 = 15 and120575 = 5 and (d) 119899 = 15 and 120575 = 10
62 119862 Type and Contaminant-Absent Events The 119862 typeevents are of major consequence for sample statistical param-eters In such events because the contaminant 119909
119888occupies
an extreme outlying position (119909(119899)
or 119909(1)) in an ordered data
array it is desirable that the discordancy tests detect thiscontaminant observation as discordant The 120587
119863119862and 120587
119863119862
values for 119899 = 5 to 119899 = 20 as a function of 120575 are presentedin Figures 5(a)ndash5(d) and Figures 6(a)ndash6(d) respectivelySimilarly these values as a function of 120576 are shown in Figures7(a)ndash7(d) and Figures 8(a)ndash8(d) respectively
For uncontaminated samples (120575 = 0 in Figures 5(a)ndash5(d)and Table 2 or 120576 = plusmn1 in Figures 7(a)ndash7(d)) the probability
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 17
0 1 2
0990
0992
0994
0996
0998
minus2 minus1
120575
120587DC
(a) 119899 = 5
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
120587DC
(b) 119899 = 10
0990
0992
0994
0996
0998
0 1 2minus2 minus1
N2N8
N14N15
120575
120587DC
(c) 119899 = 15
0990
0992
0994
0996
0998
0 1 2minus2 minus1
120575
N2N8
N14N15
120587DC
(d) 119899 = 20
Figure 3 Spurious type II error probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
120587119863119862
values were close to the theoretical value of 099 (whichcorresponds to the confidence level used for each test)Similarly for such samples 120587
119863119862values for all sample sizes
were close to the theoretical value of 001 (complement of 099is 001 Figures 6 and 8)
A complementary behavior of 120587119863119862
and 120587119863119862
exists forall other 120575 or 120576 values as well (Figures 5 and 7 or Figures 6and 8) Thus for all tests 120587
119863119862decreases sharply from 099
for 120575 = 0 to very small values of about 003 for 120575 = plusmn20
and 119899 = 5 to about 001ndash003 for 120575 = plusmn9 and 119899 = 10to about 0006ndash002 for 120575 = plusmn8 and 119899 = 15 and to about0001ndash001 for 120575 = plusmn8 and 119899 = 20 (Table 2 Figures 5(a)ndash5(d))On the contrary 120587
119863119862increases very rapidly from very small
values of 001 to close to the maximum theoretical value of099 (see the complementary behavior 120587
119863119862in Figures 6(a)ndash
6(d) and Figures 5(a)ndash5(d))These probability (120587119863119862
and 120587119863119862
)
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
18 The Scientific World Journal
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120587DC
120575
(a) 119899 = 5
0 1 2minus2 minus1
120575
0000
0002
0004
0006
0008
0010
120587DC
(b) 119899 = 10
0000
0002
0004
0006
0008
0010
0 1 2minus2 minus1
120575
N8N14N15
N2
120587DC
(c) 119899 = 15
0000
0002
0004
0006
0008
0010
N8N14N15
0 1 2minus2 minus1
120575
N2
120587DC
(d) 119899 = 20
Figure 4 Spurious power probability (120587119863119862
) as a function of 120575 from minus25 to +25 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
values show a similar behavior for larger values of 120576 thanfor 120575 (compare Figures 7 and 8 with Figures 5 and 6 resp)There are some differences in these probability values amongthe different tests (Table 2 Figures 5ndash8) but theywill be betterdiscussed for the test performance criteria
63 Test Performance Criteria (Ω and 120587119863|119862) These two
parameters are plotted as a function of 120575 and 120576 in Figures 9
10 11 and 12 and the most important results are summarizedin Tables 3ndash6 For a good test both Ω (120587
119863119862+ 120587119863119862
(5)) and120587119863|119862
(6) should be large [1 7] Values of both performancecriteria (Ω and 120587
119863|119862) increase as 120575 or 120576 values depart from
the uncontaminated values of 120575 = 0 or 120576 = plusmn1 (Figures 9ndash12 Tables 3ndash6) HoweverΩ and 120587
119863|119862increase less rapidly for
smaller 119899 than for larger 119899 For 119899 = 5 even for 120575 = plusmn20 or120576 = plusmn200 none of the two parameters truly reaches the max-imum theoretical value of 099 (Figure 9(a) to Figure 12(a))
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 19
0 5 10 15 20
00
02
04
06
08
10120587DC
120575
minus20 minus15 minus10 minus5
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20
120587DC
120575
minus20 minus15 minus10 minus5
N8N14N15
N2
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 5 Nonspurious type II error probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For larger 119899 (10ndash20) however both Ω and 120587119863|119862
get close tothis value for all tests and for much smaller values of 120575 or 120576
than themaximumvalues of 20 and 200 respectively (Figures9(b)ndash9(d) to Figures 12(b)ndash12(d) Tables 3ndash6)
The performance differences of the four tests are nowbriefly discussed in terms of both 120575 and 120576 as well as 119899The totaluncertainty 119906
99values of the simulations are extremely small
(the error is at the fifth or even sixth decimal place Tables 3ndash6) Therefore most differences among the tests (Δ119909N8 for test
N8 Δ119909N14 for test N14 and Δ119909N15 for test N15 all percentdifferences are with respect to test N2 see (7)) are statisticallysignificant (Tables 3ndash6) A negative value of Δ119909N119895 (where N119895
stands for N8 N14 or N15) means that Ω or 120587119863|119862
value fora given test (N8 N14 or N15) is less than that of test N2implying a worse performance of the given test as comparedto test N2 whereas a positive value of Δ119909N119895 signifies justthe opposite Note that test N2 is chosen as a reference testbecause it shows generally the best performance (values of
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
20 The Scientific World Journal
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(a) 119899 = 5
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
120587DC
(b) 119899 = 10
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(c) 119899 = 15
00
02
04
06
08
10
0 5 10 15 20120575
minus20 minus15 minus10 minus5
N8N14N15
N2
120587DC
(d) 119899 = 20
Figure 6 Nonspurious power probability (120587119863119862
) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 120587119863119862
values foruncontaminated samples (120575 = 0) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Δ119909N119895 aremostly negative in Tables 3ndash6) Additional fine-scalesimulations were also carried out for which both Ω and 120587
119863|119862
become about 05 for the reference test N2 (05 is about thehalf of the maximum value of one for Ω or 120587
119863|119862) Hence the
values of Ω and 120587119863|119862
can be visually compared in Tables 3ndash6(see the rows in italic font)
For 119899 = 5 all tests show rather similar performancebecause the maximum difference (Δ119909N119895) is only about minus11for N8 (as compared to N2) and lt minus01 for N14 and N15
(see the first set of rows corresponding to 119899 = 5 in Tables3ndash6) Test N2 shows Ω = 050044 for 120575 = plusmn1017 whereastests N8 N14 and N15 haveΩ values of 049503 050014 and050015 respectively (Table 3)The respectiveΔ119909N119895 values areabout minus11 minus006 and minus006 (Table 3) Practically thesame results are valid for 120587
119863|119862as well (see the row in italic
font in Table 4) Similar results were documented for Ω and120587119863|119862
as a function of 120576 (rows for 120576 = plusmn129 or plusmn131 in Tables5 and 6 resp)
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 21
0 100 200
00
02
04
06
08
10120587DC
120576
minus200 minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
(b) 119899 = 10
0 100 200120576
minus200 minus100
00
02
04
06
08
10
120587DC
N8N14N15
N2
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
120587DC
120576
minus200 minus100
N8N14N15
N2
(d) 119899 = 20
Figure 7 Nonspurious type II error probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
For 119899 = 10 Dixon test N8 becomes considerablyless efficient than Grubbs test N2 because the Δ119909N8 valuesbecome as low as minus78 for 120575 = plusmn5 or minus64 for 120576 = plusmn3
(Tables 3ndash6) Skewness test N14 also shows slightly lowerΩ and 120587
119863|119862than N2 (Δ119909N14 = minus18 for 120575 = plusmn4- plusmn 5
or Δ119909N14 = minus15 for 120576 = plusmn3 Tables 3ndash6) Kurtosis testN15 shows a similar performance as test N2 the maximumdifference Δ119909N15 is about 07 (Tables 3ndash6) For 119899 = 10 test
N2 shows Ω = 05 (or 120587119863|119862
= 05) for 120575 = plusmn5105for this case the other three tests (N8 N14 and N15) showΔ119909N119895 values of about minus78 minus18 and minus07 (Tables 3and 4) Similarly for such cases Ω and 120587
119863|119862show Δ119909N8
Δ119909N14 and Δ119909N15 values of about minus43 minus10 and minus04respectively
For 119899 = 15 and 119899 = 20 test N8 shows the worstperformance and the Δ119909N8 values become as large as minus122
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
22 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100minus200
120587DC
120576
N2N8
N14N15
(c) 119899 = 15
N2N8
N14N15
0 100 200minus100minus200
120576
00
02
04
06
08
10
120587DC
(d) 119899 = 20
Figure 8 Nonspurious power probability (120587119863119862
) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 120587119863119862
values for uncontaminated samples (120576 = plusmn1) are shown by open circles (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
to minus155 for 120575 (Tables 3 and 4) or minus98 to minus115 for 120576
(Tables 5 and 6) For these sample sizes test N14 also showsa worse performance as compared to N2 because the max-imum differences represented by Δ119909N14 values are aboutminus63 tominus109 for 120575 (Tables 3 and 4) orminus45 tominus70 for 120576(Tables 5 and 6) Test N15 shows a comparable performancebecause the maximum differences (Δ119909N15 values) are aboutminus15 to minus24 for 120575 (Tables 3 and 4) or minus11 to minus15 for
120576 (Tables 5 and 6) For 119899 = 15 and 119899 = 20 when test N2shows Ω = 05 or 120587
119863|119862= 05 the Δ119909N8 Δ119909N14 and Δ119909N15
values range from about minus69 to minus150 minus30 to minus92and minus06 to minus19 respectively
The significantly lower Ω and 120587119863|119862
values of the Dixontest N8 as compared to the Grubbs test N2 skewness test N14and kurtosis test N15 may be related to the masking effect ofthe penultimate observation 119909
(119899minus1)on 119909(119899)
or of 119909(2)
on 119909(1)
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 23
0 5 10 15 20
00
02
04
06
08
10
minus20 minus15
Ω
minus5minus10
120575
(a) 119899 = 5
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
(b) 119899 = 10
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
Ω
0 5 10 15 20minus20 minus15 minus5minus10
120575
N2N8
N14N15
(d) 119899 = 20
Figure 9 Power of Test (Ω) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d)119899 = 20
as documented by Barnett and Lewis [1] The masking effectmay also be responsible for a somewhat worse performanceof N14 as compared to N2
64 Final Remarks The two performance criteria (Ω and120587119863|119862
) [1 7] used in this work provide similar estimates(Tables 3ndash6) and more importantly similar conclusionsTherefore any of them can be used to evaluate numer-ous other discordancy tests for single or multiple outliers[1 26ndash28] The main result of Monte Carlo simulations
concerning the performance of the single extreme outlierdiscordancy tests could be stated as follows N2 cong N15 gt
N14 gt N8Additional simulation work is required to evaluate other
discordancy tests such as the single upper or lower outliertests as well as more complex statistical contaminationinvolving two or more discordant outliers and the compar-ison of consecutive application of single outlier discordancytests with multiple outlier tests [1 7 26ndash28] Then the multi-ple test method initially proposed by Verma [29] and used by
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
24 The Scientific World Journal
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
(a) 119899 = 5
0 100 20000
02
04
06
08
10
minus100
120576
minus200
Ω
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(c) 119899 = 15
0 100 200
00
02
04
06
08
10
minus100
120576
minus200
Ω
N2N8
N14N15
(d) 119899 = 20
Figure 10 Power of Test (Ω) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8 N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c)119899 = 15 and (d) 119899 = 20
many researchers [30ndash35] would be substantially improvedfor subsequent applicationsThese performance results couldthen be incorporated in new versions of the computerprograms DODESSYS [36] TecD [37] and UDASYS [38]
7 Conclusions
Our simulation study clearly shows that Dixon test N8performs less well than the other three extreme single
outlier tests (Grubbs N2 skewness N14 and kurtosis N15)Both performance parameters (the Power of Test Ω and TestPerformance Criterion 120587
119863|119862) have up to about 16 less values
for N8 than test N2 Test N8 therefore shows the worstperformance for outlier detection For certain values of 120575 or120576 test N14 also shows lesser values of Ω and 120587
119863|119862than N2
which means that N14 is also somewhat worse than N2 Theother two tests (N2 andN15) could be considered comparablein their performance
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 25
0 5 10 15 20
00
02
04
06
08
10
minus20 minus10minus15
120575
120587DC
minus5
(a) 119899 = 5
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
(b) 119899 = 10
00
02
04
06
08
10
120587DC
0 5 10 15 20minus20 minus10minus15
120575
minus5
N2N8
N14N15
(c) 119899 = 15
00
02
04
06
08
10
120587DC
N2N8
N14N15
0 5 10 15 20minus20 minus10minus15
120575
minus5
(d) 119899 = 20
Figure 11 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120575 from minus20 to +20 for all tests N2 N8 N14 and N15(a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
The computing facilities for this work were partly fromthe DGAPA-PAPIIT project IN104813 The second author
(Lorena Dıaz-Gonzalez) acknowledges PROMEP support tothe project ldquoEstadıstica computacional para el tratamientode datos experimentalesrdquo (PROMEP103-5107332) Thethird author (Mauricio Rosales-Rivera) thanks the SistemaNacional de Investigadores (Mexico) for a scholarship thatenabled him to participate in this research as Ayudantes deInvestigadorNacionalNivel III of the first author (Surendra PVerma)
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
26 The Scientific World Journal
00
02
04
06
08
10120587DC
0 100 200minus200
120576
minus100
(a) 119899 = 5
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
(b) 119899 = 10
0 100 200
00
02
04
06
08
10
minus200
120576
120587DC
minus100
N2N8
N14N15
(c) 119899 = 15
N2
0 100 200minus200
120576
00
02
04
06
08
10
120587DC
minus100
N8N14N15
(d) 119899 = 20
Figure 12 Test Performance Criterion (120587119863|119862
or Conditional Power P5) as a function of 120576 from minus1 to minus200 and +1 to +200 for all tests N2 N8N14 and N15 (a) 119899 = 5 (b) 119899 = 10 (c) 119899 = 15 and (d) 119899 = 20
References
[1] V Barnett and T Lewis Outliers in Statistical Data John Wileyamp Sons Chichester UK 3rd edition 1994
[2] W J Dixon ldquoAnalysis of extreme valuesrdquo Annals of Mathemati-cal Statistics vol 21 no 4 pp 488ndash506 1950
[3] T S Ferguson ldquoRules for rejection of outliersrdquo Revue delrsquoInstitut International de Statistique vol 29 no 3 pp 29ndash431961
[4] S S Shapiro M B Wilk and H J Chen ldquoA comparative studyof various tests for normalityrdquo Journal of American StatisticalAssociation vol 63 no 324 pp 1343ndash1371 1968
[5] D M Hawkins ldquoAnalysis of three tests for one or two outliersrdquoStatistica Nederlandica vol 32 no 3 pp 137ndash148 1978
[6] P Prescott ldquoExamination of the behavior of tests for outlierswhen more than one outlier is presentrdquo Applied Statistics vol27 no 1 pp 10ndash25 1978
[7] K Hayes and T Kinsella ldquoSpurious and non-spurious powerin performance criteria for tests of discordancyrdquo Journal of theRoyal Statistical Society Series D vol 52 no 1 pp 69ndash82 2003
[8] G L Tietjen and R H Moore ldquoSome Grubbs-type statistics forthe detection of several outliersrdquo Technometrics vol 14 no 3pp 583ndash597 1972
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013
The Scientific World Journal 27
[9] B Rosner ldquoOn the detection of many outliersrdquo Technometricsvol 17 no 2 pp 221ndash227 1975
[10] J P Royston ldquoThe W test for normalityrdquo Journal of the RoyalStatistical Society C vol 31 no 2 pp 176ndash180 1982
[11] J S Simonoff ldquoThe breakdown and influence properties ofoutlier rejection-plus-mean proceduresrdquo Communications inStatistics vol 16 no 6 pp 1749ndash1760 1987
[12] C E Efstathiou ldquoEstimation of type I error probability fromexperimental Dixonrsquos ldquoQrdquo parameter on testing for outlierswithin small size data setsrdquoTalanta vol 69 no 5 pp 1068ndash10712006
[13] R Gottardo A E Raftery K Y Yeung and R E BumgarnerldquoQuality control and robust estimation for cDNA microarrayswith replicatesrdquo Journal of the American Statistical Associationvol 101 no 473 pp 30ndash40 2006
[14] S Khedhiri andG EMontasser ldquoThe effects of additive outlierson the seasonal KPSS test a Monte Carlo analysisrdquo Journal ofStatistical Computation and Simulation vol 80 no 6 pp 643ndash651 2010
[15] P A Patel and J S Patel ldquoA monte carlo comparison ofsome variance estimators of the Horvitz-Thompson estimatorrdquoJournal of Statistical Computation and Simulation vol 80 no 5pp 489ndash502 2010
[16] H A Noughabi and N R Arghami ldquoMonte carlo comparisonof seven normality testsrdquo Journal of Statistical Computation andSimulation vol 81 no 8 pp 965ndash972 2011
[17] K Krishnamoorthy andX Lian ldquoClosed-form approximate tol-erance intervals for some general linearmodels and comparisonstudiesrdquo Journal of Statistical Computation and Simulation vol82 no 4 pp 547ndash563 2012
[18] S P Verma ldquoGeochemometricsrdquo Revista Mexicana de CienciasGeologicas vol 29 no 1 pp 276ndash298 2012
[19] S P Verma A Quiroz-Ruiz and L Dıaz-Gonzalez ldquoCriticalvalues for 33 discordancy test variants for outliers in normalsamples up to sizes 1000 and applications in quality control inEarth Sciencesrdquo Revista Mexicana de Ciencias Geologicas vol25 no 1 pp 82ndash96 2008
[20] S P Verma and A Quiroz-Ruiz ldquoCorrigendum to Criticalvalues for 22 discordancy test variants for outliers in normalsamples up to sizes 100 and applications in science andengineering [Revista Mexicana de Ciencias Geologicas vol 23pp 302ndash319 2006]rdquo Revista Mexicana de Ciencias Geologicasvol 28 no 1 p 202 2011
[21] S P Verma and A Quiroz-Ruiz ldquoCritical values for six Dixontests for outliers in normal samples up to sizes 100 andapplications in science and engineeringrdquo Revista Mexicana deCiencias Geologicas vol 23 no 2 pp 133ndash161 2006
[22] R Cruz-Huicochea and S P Verma ldquoNew critical values forF and their use in the ANOVA and Fisherrsquos F tests for evalu-ating geochemical reference material granite G-2 (USA) andigneous rocks from the Eastern Alkaline Province (Mexico)rdquoJournal of Iberian Geology vol 39 no 1 pp 13ndash30 2013
[23] S P Verma and R Cruz-Huicochea ldquoAlternative approach forprecise and accurate Studentrsquos t critical values in geosciencesand application in geosciencesrdquo Journal of Iberian Geology vol39 no 1 pp 31ndash56 2013
[24] F E Grubbs ldquoProcedures for detecting outlying observations insamplesrdquo Technometrics vol 11 no 1 pp 1ndash21 1969
[25] A M Law andW D Kelton Simulation Modeling and AnalysisMcGraw Hill Boston Mass USA 3rd edition 2000
[26] S P Verma L Dıaz-Gonzalez and R Gonzalez-Ramırez ldquoRel-ative efficiency of single-outlier discordancy tests for processinggeochemical data on reference materials and application toinstrumental calibrations by a weighted least-squares linearregression modelrdquo Geostandards and Geoanalytical Researchvol 33 no 1 pp 29ndash49 2009
[27] S P Verma Estadıstica Basica para el Manejo de Datos Exper-imentales Aplicacion en la Geoquımica (Geoquimiometrıa)UNAM Mexico DF 2005
[28] R Gonzalez-Ramırez L Dıaz-Gonzalez and S P VermaldquoEficiencia relativa de 15 pruebas de discordancia con 33variantes aplicadas al procesamiento de datos geoquımicosrdquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 501ndash515 2009
[29] S P Verma ldquoSixteen statistical tests for outlier detectionand rejection in evaluation of international geochemical refer-ence materials example of microgabbro PM-Srdquo GeostandardsNewsletter vol 21 no 1 pp 59ndash75 1997
[30] E Gomez-Arias J Andaverde E Santoyo and G UrquizaldquoDeterminacion de la viscosidad y su incertidumbre en fluidosde perforacion usados en la construccion de pozos geotermicosaplicacion en el campo de Los Humeros Puebla MexicordquoRevista Mexicana de Ciencias Geologicas vol 26 no 2 pp 516ndash529 2009
[31] S G Marroquın-Guerra F Velasco-Tapia and L Dıaz-Gonzalez ldquoEvaluacion estadıstica de materiales de referen-cia geoquımica del Centre de Recherches Petrographiques etGeochimiques (Francia) aplicando un esquema de deteccion yeliminacion de valores desviadosrdquoRevistaMexicana de CienciasGeologicas vol 26 no 2 pp 530ndash542 2009
[32] K Pandarinath ldquoEvaluation of geochemical sedimentary refer-ence materials of the Geological Survey of Japan (GSJ) by anobjective outlier rejection statistical methodrdquo Revista Mexicanade Ciencias Geologicas vol 26 no 3 pp 638ndash646 2009
[33] K Pandarinath ldquoClay minerals in SW Indian continental shelfsediment cores as indicators of provenance and palaeomon-soonal conditions a statistical approachrdquo International GeologyReview vol 51 no 2 pp 145ndash165 2009
[34] K Pandarinath and S K Verma ldquoApplication of four setsof tectonomagmatic discriminant function based diagrams tobasic rocks from northwest Mexicordquo Journal of Iberian Geologyvol 39 no 1 pp 181ndash195 2013
[35] M P Verma ldquoIAEA inter-laboratory comparisons of geother-mal water chemistry critiques on analytical uncertainty accu-racy and geothermal reservoir modelling of Los AzufresMexicordquo Journal of Iberian Geology vol 39 no 1 pp 57ndash722013
[36] S P Verma and L Dıaz-Gonzalez ldquoApplication of the dis-cordant outlier detection and separation system in the geo-sciencesrdquo International Geology Review vol 54 no 5 pp 593ndash614 2012
[37] S P Verma andM A Rivera-Gomez ldquoNew computer programTecD for tectonomagmatic discrimination from discriminantfunction diagrams for basic and ultrabasic magmas and itsapplication to ancient rocksrdquo Journal of Iberian Geology vol 39no 1 pp 167ndash179 2013
[38] S P Verma R Cruz-Huicochea and L Dıaz-Gonzalez ldquoUni-variate data analysis system deciphering mean compositions ofisland and continental arcmagmas and influence of underlyingcrustrdquo International Geology Review vol 55 no 15 pp 1922ndash1940 2013