kumaraswamy weibull-generated family of … kumaraswamy weibull-generated family of distributions...
TRANSCRIPT
Advances and Applications in Statistics © 2016 Pushpa Publishing House, Allahabad, India Published Online: May 2016 http://dx.doi.org/10.17654/AS048030205 Volume 48, Number 3, 2016, Pages 205-239 ISSN: 0972-3617
Received: February 14, 2016; Revised: March 1, 2016; Accepted: March 20, 2016 2010 Mathematics Subject Classification: 62-XX. Keywords and phrases: Kumaraswamy Weibull distribution, Weibull-G family, moments, order statistics, maximum likelihood estimation.
Communicated by K. K. Azad
KUMARASWAMY WEIBULL-GENERATED FAMILY OF DISTRIBUTIONS WITH APPLICATIONS
Amal S. Hassan1 and M. Elgarhy1,2 1Institute of Statistical Studies and Research (ISSR) Department of Mathematical Statistics Cairo University Egypt e-mail: [email protected]
2Department of Mathematical Statistics Burraydah Colleges Qassim, Saudi Arabia e-mail: [email protected]
Abstract
In this paper, we introduce and study a new family of continuous distributions called Kumaraswamy Weibull-generated ( )GKwW -
family of distributions which is an extension of the Weibull-G family of distributions proposed by Bourguignon in [3]. The new family includes several known models. We obtain general explicit expressions for the quantile function, moments, probability weighted moments, generating function, mean deviation and order statistics. Making use of maximum likelihood method, we discuss the model
Amal S. Hassan and M. Elgarhy 206
parameters. Furthermore, besides showing the usefulness of our proposed family of distributions by considering a three real data set, evidence of this family to outperform other classes of lifetime models has been noticed.
1. Introduction
Over the past decades, several statistical distributions have been extensively used and applied in several areas such as engineering, actuarial, medical sciences, demography, etc. However, in many situations, the classical distributions are not suitable for describing and predicting real world phenomena. Recently, there has been an increased interest in defining new generators or generalized classes of univariate continuous distributions by introducing additional shape parameter(s) to baseline model. The new generating families extended the well-known distributions and at the same time provide great flexibility in modeling data in practice.
Eugene et al. [8] introduced a new class of generated distributions based on beta distribution, called the beta-generated family of distributions. Zografos and Balakrishnan [14] proposed gamma generated family. Cordeiro and de Castro [4], introduced Kumaraswamy generalized distributions. From an arbitrary parent cumulative distribution function (cdf) ( ),xG Ristic and
Balakrishnan [13] proposed an alternative gamma generator for any parent distribution ( ).xG Pescim et al. [11] proposed the Kummer beta-generated
family of distributions for any parent cdf. Cordeiro et al. [5], proposed the exponentiated generalized class of distributions. Alzaatreh et al. [1], extended the beta-generated family of distributions by using any non-negative continuous random variable T as the generator, in place of the beta random variable. They defined this class as follows:
( ) ( )( )( )
∫=xGW
dttrxF0
, (1)
where ( )tr is the probability density function (pdf) of a non-negative
Kumaraswamy Weibull-generated Family of Distributions … 207
continuous random variable T. In this new class, the distribution of the random variable T is the generator and the upper limit is the transformation
( )( ) ( )[ ].1log xGxGW −−= This generated family of distributions is called
“T-X distribution”. It is clear that one can define a different upper limit for generating different types of T-X distributions. The transformation ( )( )xGW
satisfies the following two conditions: ( )( ) [ )∞∈ ,0xGW and it is a
monotonic non decreasing function.
Recently, Bourguignon et al. [3], proposed the Weibull- ( )GWG - family
of distributions. They considered a continuous distribution ( )⋅G with density
function ( )⋅g and the Weibull cdf ( )βα−−= tetR 1 ( )0for >t with positive
parameters α and .β Based on this density, by replacing ( )( )xGW in (1)
with ( ) ( ) [ ( ) ( )],;1;,;; ξ−=ξξξ xGxGxGxG they defined cdf of their
family by
( )( )( )
( )( )∫ ξ−ξ β
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−α−−β −=αβ=ξβα
β;1;
0;1
;1 ,1,,; xG
xGxG
xGt edtetxF
,0,, >βα⊆∈ RDx
where ( )ξ;xG is a baseline cdf, which depends on a parameter vector .ξ
In this paper, a new extension of the GW - family of distributions called
Kumaraswamy Weibull-G family of distributions is proposed. The rest of this article is organized as follows: In Section 2, the Kumaraswamy Weibull-G family of distributions is defined. Some general mathematical properties of the family are discussed in Section 3. Some members of the proposed family are discussed in Section 4. In Section 5, the estimation of the model parameters is performed by the method of maximum likelihood. An illustrative application based on real data is examined in Section 6. Finally, concluding remarks are presented in Section 7.
Amal S. Hassan and M. Elgarhy 208
2. The Kumaraswamy Weibull-generated Family
In this section, the new class of distributions, called GKwW - family of
distributions is introduced.
Cordeiro et al. [6] presented the KwW distribution with the following cdf and pdf:
( ) [ [ ] ] ,0,,,;0;111 >βα>−−−=βα− batetR bat
and
( ) [ ] [ [ ] ] ;111 111 −α−−α−α−−β βββ−−−αβ= batatt eeetabtr
.0,,,;0 >βα> bat (2)
We obtain the new family by replacing the generator ( )tr defined in (1)
by the pdf generator defined in (2) as follows:
( ) [ ] [ [ ] ]( )( )∫ ξ−ξ
−α−−α−α−−β βββ−−−αβ= ;1
;
0111 .111xG
xGbatatt dteeetabxF
Then, the distribution function of GKwW - family takes the following form:
( ) [ (( )( ) ) ] ,0,,,;0;111 ;1;
>βα>−−−=
β
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−baxexF baxG
xG
(3)
where 0,, >βba are the three shape parameters and 0>α is the scale
parameter. The cdf (3) provides a wider family of continuous distributions. The pdf corresponding to (3) is given by
( ) ( )( ) ( )( )( )
( )( ) [
( )( ) ] 1;1;
;1;
1
11
;1;; −⎥⎦
⎤⎢⎣⎡
ξ−ξ
α−⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−
+β
−βββ
−ξ−
ξξαβ= axGxG
xGxG
eexG
xgxGabxf
[ (( )( ) ) ] .0,,,;0;11 1;1;
>βα>−−× −⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−β
baxe baxGxG
(4)
Hereafter, a random variable X with pdf (4) is denoted by .-~ GKwWX
Kumaraswamy Weibull-generated Family of Distributions … 209
Note that.
(i) For ,1=b the GKwW - distribution reduces to a new family, called
exponentiated Weibull-generated ( )GEW - family of distributions.
(ii) For 1== ba the GKwW - distribution reduces to GW - family of
distributions which is obtained by Bourguignon et al. [3].
The survival, hazard, reversed hazard and cumulative hazard rate functions are obtained, respectively, as follows:
( ) [ (( )( ) ) ] ,0,,,;0;11 ;1;
>βα>−−=
β
⎥⎦⎤
⎢⎣⎡
ξ−ξα−
baxexF baxGxG
( ) ( ) ( )( )( ) [
( )( ) ]
( )( ) [ (( )( ) ) ]
,
11;1
1;;
;1;
1
1;1;
;1;
1
axGxG
axGxG
xGxG
exG
eexgxGabxh β
ββ
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−+β
−⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−−β
−−ξ−
−ξξαβ=
( )
( ) ( )( )( ) [
( )( ) ]
[ (( )( ) ) ]
( )( ) { [ (( )( ) ) ] }
,
111;1
11
1;;
;1;
1
1;1;
1;1;
;1;
1
baxGxG
baxGxG
axGxG
xGxG
exG
e
eexgxGab
x β
β
ββ
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−+β
−⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−
−⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−−β
−−−ξ−
−−×
−ξξαβ
=τ
and
( ) ( ) [ (( )( ) ) ] .11ln1ln ;1;
baxGxG
exFxH
β
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−−−−=−−=
3. Statistical Properties
In this section, some properties of the GKwW - family will be obtained
as follows:
3.1. Quantile and median
Quantile functions are used in theoretical aspects of probability theory,
Amal S. Hassan and M. Elgarhy 210
statistical applications and simulations. Simulation methods utilize quantile function to produce simulated random variables for classical and new
continuous distributions. The quantile function, say ( ) ( )uFuQ 1−= of X is
given by
[ (( )( )( )( ) ) ]bauQGuQG
eu
β
⎥⎦⎤
⎢⎣⎡−
α−−−−= 1111
after some simplifications, it reduces to the following form:
( ) { [ [ ( ) ] ] }
{ [ [ ( ) ] ] },
111ln1
111ln1111
11111
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡
−−−+
−−−=βα
−βα
−−
ab
ab
u
uGuQ (5)
where u is considered as a uniform random variable on the unit interval (0, 1)
and ( )⋅−1G is the inverse function of ( ).⋅G
In particular, the median can be derived from (5) by setting ,5.0=u i.e., the median is given by
{ [ [ ( ) ] ] }
{ [ [ ( ) ] ] }.
5.011ln1
5.011ln1111
11111
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡
−−+
−−=βα
−βα
−−
ab
abGMedian
3.2. Expansions for distribution and density functions
In this section, some representations of the cdf and pdf for Kumaraswamy Weibull-G family of distributions will be presented. The mathematical relation given below will be useful in this section.
It is well-known that, if 0>β is real non integer and ,1<z the
generalized binomial theorem is written as follows:
( ) ( )∑∞
=
−β ⎟⎠⎞
⎜⎝⎛ −β
−=−0
1 .1
11i
ii zi
z (6)
Then, by applying the binomial theorem (6) in (4), the probability density function of GKwW - family, where b is real non integer becomes
Kumaraswamy Weibull-generated Family of Distributions … 211
( ) ( ) ( )( )( )
( )( ) 1
;1;
1
;1;;
+β
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−−β
ξ−
ξβαβ=
β
xGexgxGabxf
xGxG
( ) [( )( ) ] ( )∑
∞
=
−+⎥⎦⎤
⎢⎣⎡
ξ−ξα−
β
−⎟⎠⎞
⎜⎝⎛ −
−×0
11;1;
.11
1i
iaxGxG
i ei
b
Then, using binomial expansion again in the previous equation, where a is real non integer, leads to:
( ) ( ) ( )( )( ) 1
1
;1;;
+β
−β
ξ−
ξβαβ=xG
xgxGabxf
( )( ) ( ) ( )
( )∑∞
=
⎥⎦⎤
⎢⎣⎡
ξ−ξ+α−
+
β
⎟⎠⎞
⎜⎝⎛ −+⎟⎠⎞
⎜⎝⎛ −
−×0,
;1;1
,111
1ji
xGxGjji e
jia
ib
(7)
by using the power series for the exponential function, we obtain
( ) ( )( ) ( ) ( ) ( )
( )∑∞
=
β⎥⎦⎤
⎢⎣⎡
ξ−ξ
+α−
⎥⎦⎤
⎢⎣⎡
ξ−ξ+α−=
β
0
;1;1
.;1;
!11
k
kkkkxG
xGj
xGxG
kje (8)
Inserting expansion (8) in (7) we have
( ) ( ) ( ) ( ) ( )∑∞
=
++
⎟⎟⎠
⎞⎜⎜⎝
⎛ −+⎟⎟⎠
⎞⎜⎜⎝
⎛ −+α−ξαβ=0,,
111!
11;kji
kkkji
j
ia
i
bk
jxgabxf
( )[ ] ( ) ( )[ ] ( )[ ].;1; 1111 ++β−−+β ξ−ξ× kk xGxG (9)
Now, using the generalized binomial theorem, we can write
( )[ ] ( )[ ] ( )( )( )∑
∞
=
++β− ξ⎟⎠⎞
⎜⎝⎛ ++β
=ξ−0
11 .;1
;1 xGk
xG k (10)
Inserting (10) in (9), the probability density function of GKwW − can
Amal S. Hassan and M. Elgarhy 212
be expressed as an infinite linear combination of exponentiated-G (exp-G for short) density functions, i.e.,
( ) ( ) ( ) ( )∑∞
=
−++βξξη=0,,,
11,,, ,;;
kji
kkji xGxgxf (11)
then
( ) ( ) ( )∑∞
=+ξ+β ξ=
0,,,,,, ,;
kjikkji xhWxf (12)
where
( ) ( )( )[ ]
( ) ( ),
1111!1
111,,, ⎟⎟
⎠
⎞⎜⎜⎝
⎛ ++β⎟⎟⎠
⎞⎜⎜⎝
⎛ −+⎟⎟⎠
⎞⎜⎜⎝
⎛ −++β
+−α=+++ k
jia
ib
kkjabW
kkjikkji
and
( )[ ] ( ) ( ) ( ) .;;;,11,,,
,,,−ξξ=ξ
++βη
= aa
kjikji xGxagxhkW
If ba, and β are integers, the index i stops at jb ,1− stops at
( ) 11 −+ia and stops at ( ) .1 ++β k
Another form can be yielded when β is real non integer by adding and
subtracting 1 to ( ) ( ) 11; −++βξ kxG into (10) as follows:
( ) ( ) ( )[ ][ ] ( )∑∞
=
−++βξ−−ξη=0,,,,
11,,, .;11;
ukji
kkji xGxgxf
Using binomial expansion yields:
( ) ( )( )
( ) ( )[ ]∑∞
=
ξ−ξ⎟⎟⎠
⎞⎜⎜⎝
⎛ ++βη−=
0,,,,,,, .;1;
11
ukji
ukji
u xGxgu
kxf
Using binomial expansion more time yields:
Kumaraswamy Weibull-generated Family of Distributions … 213
( ) ( )( )
( ) ( )( )∑ ∑∞
= =
+ ⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛ ++β
η−=0,,,, 0
,,, .1
1ukji
u
m
mkji
mu xGxgmu
uk
xf
So the expansion form of the pdf is as follows:
( ) ( ) ( )( )∑ ∑∞
= =
η=0,,,, 0
,,,,, ,ukji
u
m
mmukji xGxgxf (13)
where
( )( )
.1
1 ,,,,,,,, ⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛ ++βη−=η +
mu
uk
kjimu
mukji
Furthermore, an expansion for the cumulative function is derived as follows:
Using binomial expansion for ( )[ ] ,hxF where h is an integer, leads to:
( )[ ] ( ) [ (( )( ) ) ]∑
=
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−β
−−⎟⎟⎠
⎞⎜⎜⎝
⎛−=
h
g
bgaxGxG
gh egh
xF0
;1;
.111
Using binomial expansion another time, where b is a real non integer, leads to
( )[ ] ( ) (( )( ) )∑∑
=
∞
=
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−+
β
−⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛−=
h
g p
apxGxG
pgh ep
bggh
xF0 0
;1;
.11
Using binomial expansion again, where a is a real non integer, leads to
( )[ ] ( )( )( )∑ ∑
=
∞
=
⎥⎦⎤
⎢⎣⎡
ξ−ξ
α−++
β
⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛−=
h
g qp
xGxGqqpgh e
q
ap
p
bg
g
hxF
0 0,
;1;
,1
by using the power series for the exponential function (8) in the previous equation, we obtain
Amal S. Hassan and M. Elgarhy 214
( )[ ] ( ) ( ) ( )( )∑ ∑
=
∞
=
β+++
⎥⎦⎤
⎢⎣⎡
ξ−ξ
⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛α−=h
g tqp
tttqpgh
xGxG
q
ap
p
bg
g
ht
qxF0 0,,
,;1;
!1
by using the relation (10) in the previous equation where β is real non
integer, we obtain
( )[ ] ( ) ( ) ( )( )∑ ∑=
∞
=
+β+++
ξ⎟⎟⎠
⎞⎜⎜⎝
⎛β⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛α−=h
g tuqp
utttqpg
h xGut
qap
pbg
gh
tqxF
0 0,,,,;!
1
( )[ ] ( ) ( ) ( )[ ][ ]∑ ∑=
∞
=
+β+++
ξ−−⎟⎟⎠
⎞⎜⎜⎝
⎛β⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛α−=h
g tuqp
utttqpg
h xGut
qap
pbg
gh
tqxF
0 0,,,.;11!
1
Using binomial expansion again, leads to
( )[ ] ( ) ( ) ( )[ ]∑ ∑=
∞
=
++++ξ−⎟
⎠⎞
⎜⎝⎛ +β⎟⎠⎞
⎜⎝⎛β⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛α−=
h
g lutqp
ltltqpg
h xGl
utut
qap
pbg
gh
tqxF
0 0,,,,.;1!
1
Using binomial expansion again, where l is a real integer, leads to
( )[ ] ( ) ( )∑ ∑ ∑=
∞
= =
+++++ α−=ξh
g lutqp
l
z
tzltqpgh
tqxF
0 0,,,, 0!
1;
( ) .; zxGzl
lut
ut
qap
pbg
gh
ξ⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛ +β⎟⎠⎞
⎜⎝⎛β⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛×
Replacing ∑ ∑∞
= =0 0l
l
zwith ∑ ∑
∞
=
∞
=zl z 0yielding
( )[ ] ( ) ( )∑ ∑ ∑∑=
∞
=
∞
=
∞
=
+++++ α−=h
g utqp zl z
tzltqpgh
tqxF
0 0,,, 0!
1
( ) .; zxGzl
lut
ut
qap
pbg
gh
ξ⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛ +β⎟⎠⎞
⎜⎝⎛β⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛⎟⎠⎞
⎜⎝⎛×
Finally,
Kumaraswamy Weibull-generated Family of Distributions … 215
( )[ ] ( )∑∞
=
ξ=0
,;z
zz
h xGsxF (14)
where
( ) ( )∑ ∑ ∑=
∞
=
∞
=
+++++
⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛ +β⎟⎟⎠
⎞⎜⎜⎝
⎛β⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎟⎠
⎞⎜⎜⎝
⎛α−=h
g utqp zl
tzltqpgz
z
l
l
ut
u
t
q
ap
p
bg
g
ht
qs0 0,,,
.!1
3.3. Moments
If X has the pdf (11), then its rth moment can be obtained through the following relation:
( ) ( )∫∞
∞−==μ′ .dxxfxXE rr
r (15)
First, when β is an integer, then substituting (11) into (15) yields:
( ) ( ) ( )( ) ( )∑ ∫∞
=
∞
∞−−++βξξη==μ′
0,,,
11,,, .;;
kji
krkji
rr dxxGxgxXE
Then
( )∑∞
=−++βτη=μ′
0,,,11,,, ,
kjikrkir
where ( ) 11, −++βτ kr is the probability weighted moments (PWMs) of the
baseline distribution. Second, when β is a real non integer, then substituting
(13) into (15) yields:
( ) ( ) ( )( )∑ ∑ ∫∞
= =
∞
∞−ξξη==μ′
0,,,, 0,,,,, .;;
ukji
u
m
mrmukji
rr dxxGxgxXE
Then, ∑ ∑∞
= =τη=μ′
0,,,, 0,,,,, .
ukji
u
mmrmukir
Amal S. Hassan and M. Elgarhy 216
Another form based on the parent quantile function:
First, when β is an integer yields
( )( ) ( )∑ ∫∞
=
−++βη=μ′0,,,
1
011
,,, .kji
krGkjir duuuQ
Second, when β is a real non integer yields
( )( )∑ ∑ ∫∞
= =
η=μ′0,,,, 0
1
0,,,,, .ukji
u
m
mrGmukjir duuuQ
3.4. The probability weighted moments
A general theory for PWMs covers the summarization and description of the theoretical probability distributions, estimation of parameters, quantiles of probability distributions and hypothesis testing for probability distributions. The PWMs method can generally be used for estimating parameters of a distribution whose inverse form cannot be expressed explicitly. The probability weighted moments can be obtained from the following relation:
[ ( ) ] ( ) ( )( )∫∞
∞−==τ ., dxxFxfxxFXE srsr
sr (16)
First, when β is an integer, by substituting (11) and (14) into (16), replacing
h with s, leads to:
( ) ( )( ) ( )∑ ∑ ∫∞
=
∞
=
∞
∞−−++β+ξξη=τ
0,,,, 0
11,,,, .;;
ukji z
kzrkjizsr dxxGxgxs
Then,
( )∑ ∑∞
=
∞
=−++β+τη=τ
0,,, 011,,,,, .
kji zkzrkjizsr s
Kumaraswamy Weibull-generated Family of Distributions … 217
Second, when β is real non integer, by substituting (13) and (14) into (16),
replacing h with s, leads to:
( ) ( )∑ ∑∑ ∫∞
= =
∞
=
∞
∞−+ξξη=τ
0,,,, 0 0,,,,,, .;;
ukji
u
m z
zmrmukjizsr dxxGxgxs
Then
∑ ∑∑∞
= =
∞
=+τη=τ
0,,,, 0 0,,,,,,, .
ukji
u
m zmzrmukjizsr s
Another form based on the parent quantile function:
First, when β is an integer yields
( )( ) ( )∑ ∑ ∫∞
=
∞
=
−++β+η=τ0,,, 0
1
011
,,,, .kji z
kzrGkjizsr duuuQs
Second, when β is real non integer yields
( )( )∑ ∑∑ ∫∞
= =
∞
=
+η=τ0,,,, 0 0
1
0,,,,,, .ukji
u
m z
mzrGmukjizsr duuuQs
3.5. The moment generating function
Generally, the moment generating function is ( ) ( ),tXX eEtM = by
using the exponential expansion, it can written as ( ) ( )∑∞
==
0.!r
rr
X XErttM
Another form based on the parent quantile function: first, when β is an
integer yields
( ) ( )( ) ( )∑ ∫∞
=
−++βη=0,,,
1
011
,,, .kji
kutQkjiu duuetM G
Amal S. Hassan and M. Elgarhy 218
Second, when β is real non integer yields
( ) ( )( )∑ ∑ ∫∞
= =
η=0,,,, 0
1
0,,,,, .ukji
u
m
mutQmukjiu duuetM G
3.6. The mean deviation
Basically, the mean deviation is a measure for the amount of scatter in X. Generally, the mean deviation is expressed by:
( ) ( ) ( )μ−μμ=δ TFX 221 and ( ) ( ),22 MTX −μ=δ
where ( ) ( )∫ ∞−=
qdxxxfqT which is the first incomplete moment.
Another form based on the parent quantile function:
First, when β is an integer yields
( ) ( ) ( )∑ ∫∞
=
+++βη=0,,,
011
,,, .kji
q kGkji duuuQqT
Second, when β is real non integer yields
( ) ( )∑ ∑ ∫∞
= =
η=0,,,, 0
1
0,,,,, .ukji
u
m
mGmukji duuuQqT
3.7. Order statistics
Let ( ) ( ) ( )nXXX <<< 21 be the order statistics of a random sample
of size n following the Kumaraswamy Weibull-G family of distributions, with parameters ,,, αba and .β The pdf of the kth order statistic, as
mentioned in David [7], can be written as follows:
( ) ( ( ) )( ( ) )
( ) ( ) ( ( ) )∑−
=ν
−+νν⎟⎟⎠
⎞⎜⎜⎝
⎛ν
−−
+−=
knk
kk
kX xFkn
knkBxf
xf k0
1.11, (17)
Kumaraswamy Weibull-generated Family of Distributions … 219
First, when β is an integer substituting (11) and (14) in (17), replacing h
with ,1−+ν k leads to
( ) ( ( ) )( ( ) )
( )1,;+−ξ
= knkBxg
xf kkX k
( ( ) ) ( )∑∑ ∑−
=ν
∞
=
∞
=
−++β+ξη×kn
z kji
kzkvzkji xGp
0 0 0,,,
11,,,, ,; (18)
where
( ) .1, zz skn
p ⎟⎟⎠
⎞⎜⎜⎝
⎛ν
−−= ν
ν
Second, when β is real non integer substituting (13) and (14) in (17),
replacing h with ,1−+ν k leads to
( ) ( ( ) )( ( ) )
( ) ∑∑−
=ν
∞
=−−ξ
=kn
z
kkX knkB
xgxf k
0 01,
;
( ( ) )∑ ∑∞
= =
+ξη×0,,,, 0
,,,,,, ,;ukji
u
m
zmkvzmukji xGp (19)
where ( )⋅g and ( )⋅G are the density and cumulative functions of the
GKwW - distributions, respectively.
Moreover, rth moment of kth order statistics is obtained as follows:
( ( ) ) ( ) ( ( ) ) ( )∫∞
∞−= .kk
rk
rk dxxfxXE (20)
First, when β is an integer substituting (18) in (20), leads to
( ( ) ) ( ) ( )∑∑ ∑−
=ν
∞
=
∞
=−++β+τη
+−=
kn
z kjikzrvzkji
rk pknkBXE
0 0 0,,,11,,,,, .1,
1
Amal S. Hassan and M. Elgarhy 220
Second, when β is real non integer substituting (19) in (20), leads to
( ( ) ) ( ) ∑∑ ∑ ∑−
=ν
∞
=
∞
= =+τη
+−=
kn
z kji
u
mmzrvzmukji
rk pknkBXE
0 0 0,,, 0,,,,,,, .1,
1
4. Some Special Sub-models
Here, we discuss a few examples of GKwW - family of distributions.
4.1. KwW-uniform distribution
As a first example, suppose that the parent distribution is uniform in
the interval ∞<θ<< x0 and ( ) .1;θ
=θxg Therefore, the KwW-uniform
(KwWU) distribution has the following cdf, pdf by direct substituting
( ) ,;θ
=θ xxG in (3) and (4) as follows:
( ) [ ( ) ] ,0,0,,,,111 θ<<>βα−−−=
β⎟⎠⎞⎜
⎝⎛
−θα−
xbaexF baxx
( )( )
[ ] [ [ ] ] ,111 111
1−⎟
⎠⎞⎜
⎝⎛
−θα−−⎟
⎠⎞⎜
⎝⎛
−θα−⎟
⎠⎞⎜
⎝⎛
−θα−
+β
−ββββ
−−−−θαβθ= bax
xax
xx
x
eeex
xabxf
.0,0,,, θ<<>βα xba
Furthermore, the survival, hazard rate and reversed hazard rate functions are obtained, respectively, as follows:
( ) [ ( ) ] ,11 baxx
exF
β⎟⎠⎞⎜
⎝⎛
−θα−
−−=
( ) [ ]
( ) [ [ ] ]
,
11
1
1
11
axx
axx
xx
ex
eexabxh β
ββ
⎟⎠⎞⎜
⎝⎛
−θα−+β
−⎟⎠⎞⎜
⎝⎛
−θα−⎟
⎠⎞⎜
⎝⎛
−θα−
−β
−−−θ
−αβθ=
Kumaraswamy Weibull-generated Family of Distributions … 221
and
( ) [ ] [ [ ] ]
( ) [ [ [ ] ] ]
.
111
111
1
111
baxx
baxx
axx
xx
ex
eeexabx β
βββ
⎟⎠⎞⎜
⎝⎛
−θα−+β
−⎟⎠⎞⎜
⎝⎛
−θα−−⎟
⎠⎞⎜
⎝⎛
−θα−⎟
⎠⎞⎜
⎝⎛
−θα−
−β
−−−−θ
−−−αβθ=τ
4.2. KwW-BurrXII distribution
Let us consider the parent BurrXII distribution with pdf and cdf given by
( ) 0,,1,,;1
1 >σμ⎥⎦
⎤⎢⎣
⎡⎟⎠⎞⎜
⎝⎛μ
+σμ=μσ−σ−
−− cxxccxgc
cc and ( ) =μσ,,; cxG
,11σ−
⎥⎦
⎤⎢⎣
⎡⎟⎠⎞⎜
⎝⎛μ
+−cx respectively. Then the KwW-BurrXII (KwWBurrXII)
distribution, has the following cdf, pdf, survival, hazard rate and reversed hazard rate functions:
( ) [ [ ] ] ,0,0,,,,,,,11111
>>σμβα−−−=
βσ
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−
xcbaexF ba
x c
( ) [ ] [ [ ] ] ,
1
111
11
1
111
111
1111
σ−
−⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−−⎥
⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−
−βσ⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−−−
⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+
−−−×
⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟⎟
⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+αβσμ
=
βσβσ
βσ
c
ba
x
a
x
cx
cc
x
ee
xexabc
xf
cc
c
( ) [ [ ] ] ,1111
ba
x c
exF
βσ
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−
−−=
Amal S. Hassan and M. Elgarhy 222
( ) [ ]
[ [ ] ]
,
111
1
11
111
111
1111
a
xc
a
x
cx
cc
c
c
c
ex
e
xexabc
xh βσ
βσ
βσ
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−σ−
−⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−
−βσ⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−−−
−−⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+
−×
⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟⎟
⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+αβσμ
=
and
( )
[ ]
[ [ ] ]
[ [ [ ] ] ]
,
1111
11
111
111
111
111111
1
ba
xc
ba
x
a
xc
x
cc
c
c
cc
ex
e
exexabc
x βσ
βσ
βσβσ
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−σ−
−⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−
−⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−−βσ⎥⎥
⎦
⎤
⎢⎢
⎣
⎡−⎟
⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+α−−−
−−−⎟⎟⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+
−−
−⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟⎟
⎠
⎞⎜⎜⎝
⎛⎟⎠⎞⎜
⎝⎛μ
+αβσμ
=τ
where 0,0 >μ>x is scale parameter, 0,,,,, >σβα cba are shape
parameters.
4.3. KwW-Weibull distribution
If the random variable X follows the Weibull distribution with cdf given
by ( ) ,0,,,1,; >γλ−=γλγλ− xexG x where λ and γ are scale and shape
parameters. The pdf and cdf of KwW-Weibull (KwWW) is derived from the KwW-G family of distributions as follows:
( ) [ ( ( ) ) ] ,0,0,,,,,111 1 >>γλβα−−−=βγλ −α− xbaexF bae x
( ) [ ] { ( ) }( ( ) ) 111111 1 −−α−λ−−α−−β−λ−γ βγλγβγλγ−αβλγ= aexex xx
eeexabxf
[ ( ( ) ) ] .11 11 −−α− βγλ−−× bae x
e
Kumaraswamy Weibull-generated Family of Distributions … 223
Additionally, survival, hazard rate and reversed hazard rate functions of KwWW take the following forms:
( ) [ ( ( ) ) ] ,11 1 bae xexF
βγλ −α−−−=
( ) [ ] { ( ) }( ( ) )
( ( ) ),
11
111
11111
ae
aexex
x
xx
e
eeexabxhβγλ
βγλγβγλγ
−α−
−−α−λ−−α−−βλ−γ
−−
−−αβλγ=
and
( )
[ ] { ( ) }
( ( ) ) [ ( ( ) ) ]
[ ( ( ) ) ],
111
111
1
1
1111
111
bae
baeae
xex
x
xx
x
e
ee
eexab
xβγλ
βγλβγλ
γβγλγ
−α−
−−α−−−α−
λ−−α−−βλ−γ
−−−
−−−
−αβλγ
=τ
respectively. Note that for 1=γ the KwW-Weibull distribution reduces to
KwW-exponential distribution.
4.4. KwW-quasi Lindley distribution
Quasi Lindley distribution is introduced by Shanker and Mishra [12]. They defined the cdf and pdf of the quasi Lindley with two parameters as follows:
( ) ,111,; ⎥⎦⎤
⎢⎣⎡
+θ+−=θ θ−
pxepxG x
( ) ( ) .1,; xexpppxg θ−θ++θ=θ
The cdf and pdf for Kumaraswamy Weibull quasi Lindley (KwWQL) distribution are derived from (3) and (4), respectively, as follows:
( ) [ ( ) ] ,0,1,0,,,,,111
1
11>−>>θβα−−−=
β−θ
⎥⎥⎦
⎤
⎢⎢⎣
⎡⎟⎠⎞⎜
⎝⎛
+θ+α−
xpbaexF bapxe x
Amal S. Hassan and M. Elgarhy 224
( )
( )
( ) [ ([ ]
) ]
( )2
111111
11111
111
1
111
111
111
1
⎟⎠⎞⎜
⎝⎛
+θ++
−×
⎥⎦
⎤⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+×θ+αβθ
=−
β−−⎟⎠⎞⎜
⎝⎛
+θ+θα−
β−θ
β−θ
−−−⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+α−
−β−θ⎥
⎥
⎦
⎤
⎢⎢
⎣
⎡θ−
⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+α−
pxp
e
pxeexpab
xfbap
xxex
x
eapxe
xxp
xe
[ ( ) ] .11 1111
−⎥⎥⎦
⎤
⎢⎢⎣
⎡⎟⎠⎞⎜
⎝⎛
+θ+α−
β−θ
−−× bapxe x
e Furthermore, the survival, hazard rate and reversed hazard rate functions
are obtained as follows:
( ) [ ( [ ] ) ] ,111
111
bae pxx
exFβ
−−⎟⎠⎞⎜
⎝⎛
+θ+θ
α−−−=
( )
( ) ( [ ] )
( ) [ ( [ ] ) ]
,
11111
1111
11
11
1
11
11
2
111
111
ae
xpxe
aex
pxx
x
pxx
epxp
e
epxexpab
xhβ
−−⎟⎠⎞⎜
⎝⎛
+θ+θ
β−θ
β−
−⎟⎠⎞⎜
⎝⎛
+θ+θ
α−
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡θ−
⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+α−
−α−−β−
θ
−−⎟⎠⎞⎜
⎝⎛
+θ++
−⎥⎦
⎤⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+θ+αβθ
=
and
( )
( )
[ [ [ ] ) ] ( [ ] )
( ) [ [ ( [ ] ) ] ]bae
aebae
pxexp
xe
pxx
pxxp
xx
xx
epxp
ee
expab
x
β
−⎟⎠⎞⎜
⎝⎛
+θ+θ
β
−⎟⎠⎞⎜
⎝⎛
+θ+θ
β
−⎟⎠⎞⎜
⎝⎛
+θ+θ
×β−θ
β−θ
−α−
−−α−−−α−
×⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+
⎥⎥
⎦
⎤
⎢⎢
⎣
⎡θ−
⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+α−
−−−⎟⎠⎞⎜
⎝⎛
+θ++
−−−
θ+αβθ
=τ
12
1111
111111
1
11
1
111
11
111
111111
111
[ ( ) ] ,11 1111
1
−⎥⎥⎦
⎤
⎢⎢⎣
⎡−⎟
⎠⎞⎜
⎝⎛
+θ+α−
β−θ
−−× bapxe x
e respectively. For θ=p the KwW-Lindley distribution will be obtained.
Kumaraswamy Weibull-generated Family of Distributions … 225
Plots of the pdf and hazard rate function for the new four special different distributions deduced from KwW-G family for some parameter values are displayed in Figures 1 and 2, respectively.
(a)
(b)
(c)
(d)
Figure 1. Density plots (a) KwWU, (b) KwWBurrXII, (c) KwWW (d) KwWQL.
Amal S. Hassan and M. Elgarhy 226
(a)
(b)
(c)
(d)
Figure 2. Hazard plots (a) KwWU, (b) KwWBurrXII, (c) KwWW, (d) KwWQL.
Kumaraswamy Weibull-generated Family of Distributions … 227
5. Maximum Likelihood Estimation
The maximum likelihood estimators of the unknown parameters of the KwW-G family from complete samples are determined. Let nXX ...,,1 be
observed values from the KwW-G distribution with parameters βα,,, ba
and .ξ Let ( )Tba ξβα=Θ ,,,, be the 1×p parameter vector. The total
log-likelihood function for the vector of parameters Θ can be expressed as
( )ΘLln
( ) ( )[ ]∑=
ξ−β+β+α++=n
iixGnnbnan
1,ln1lnlnlnln
( ) ( )[ ] ( )[ ] ( )[ ]∑ ∑ ∑= = =
βξα−ξ+ξ−+β−n
i
n
i
n
iiii xHxgxG
1 1 1,,ln,1ln1
( ) [ ( )[ ] ] ( ) [ ( ( )[ ] ) ],11ln11ln11
,
1
, ∑∑=
ξα−
=
ξα− ββ−−−+−−+
n
i
axHn
i
xH ii ebea
where ( ) ( )( ) .,1,,
ξ−ξ
=ξi
ii xG
xGxH The elements of the score function ( ) =ΘU
( )ξβα UUUUU ba ,,,, are given by
[ ( )[ ] ] ( ) ( ( )[ ] ) ( ( )[ ] )
( ( )[ ] )∑ ∑= =
ξα−
ξα−ξα−ξα−
β
βββ
−−
−−−−−+=n
i
n
i axH
xHaxHxH
ai
iii
e
eebeanU
1 1,
,,, ,
11
1ln111ln
[ ( ( )[ ] ) ]∑=
ξα− β−−+=
n
i
axHb ieb
nU1
, ,11ln
( )[ ] ( ) ( )[ ] ( )[ ]
( )[ ]∑ ∑= =
ξα−
ξα−ββ
α β
β
−
ξ−+ξ−
α=
n
i
n
ixH
xHi
ii
i
e
exHaxHnU1 1
,
,
1
,1,
( ) ( )[ ] ( )[ ] [ ( )[ ] ]
[ ( )[ ] ]∑=
ξα−
−ξα−ξα−β
β
ββ
−−
−ξ−−
n
i axH
axHxHi
i
ii
e
eexHba1
,
1,,,
11
1,1
Amal S. Hassan and M. Elgarhy 228
( )[ ] ( )[ ] ( )[ ]∑ ∑= =
ββ ξξα−ξ+
β=
n
i
n
iiii xHxHxHnU
1 1,ln,,ln
( )( )[ ]
( )[ ] ( )[ ]( )[ ]∑
=ξα−
β
β
βξα−
−
ξξ−α+
n
ixH
ii
i
ixH
e
xHxHea1
,1
,ln,1,
( ) ( )[ ] ( )[ ] ( )[ ] [ ( )[ ] ]
[ ( )[ ] ]∑=
ξα−
−ξα−ξα−β
β
ββ
−−
−ξξ−α−
n
i axH
axHxHii
i
ii
e
eexHxHba1 ,
1,,,
11
1,ln,1
and
kUξ ( ) ( )( ) ( ) ( )
( )∑ ∑= =
ξ−ξ∂ξ∂
+β+ξ
ξ∂ξ∂−β=
n
i
n
i iki
iki
xGxG
xGxG
1 1,1
,1,,1
( )( ) ( )[ ] ( )∑ ∑
= =
−β ξ∂ξ∂ξαβ−ξξ∂ξ∂
+n
i
n
ikii
iki xHxHxg
xg
1 1
1 ,,,,
( ) ( )[ ] ( )[ ] ( )( )[ ]∑
=ξα−
ξα−−β
β
β
−
ξ∂ξ∂ξ−αβ+
n
ixH
kixH
i
i
i
e
xHexHa1
,
,1
1
,,1
( )
( )[ ] ( )[ ]
[ ( )[ ] ] ( )
[ ( )[ ] ]∑=
ξα−
−ξα−
ξα−−β
β
β
β
−−
ξ∂ξ∂−×
ξ
−βα−n
i axHki
axH
xHi
i
i
i
e
xHe
exH
ba1
,
1,
,1
.11
,1
,
1
Setting βα UUUU ba ,,, and ξU equal to zero and solving the
equations simultaneously yields the maximum likelihood estimator =Θ
( )ξβα ,,,, ba of ( ) .,,,, Tba ξβα=Θ These equations cannot be solved
analytically and statistical software can be used to solve them numerically.
For interval estimation of the parameters, the 55 × observed information matrix ( ) { }ν=Θ ,uII for ( ),,,,,, ξβα= bavu whose elements are the
Kumaraswamy Weibull-generated Family of Distributions … 229
second derivatives of the log-likelihood ( )Lln function. Under the regularity
conditions, the known asymptotic properties of the maximum likelihood
method ensure that: ( ) ( ( ))Θ→Θ−Θ −1,0 INnd
as ,∞→n where d→
means the convergence in distribution, with mean zero and covariance matrix
( )Θ−1I then, the ( )%1100 ω− confidence interval for ( )ξβα=Θ ,,,, ba
is given as follows:
( ),var2 Θ±Θ ωZ
where 2ωZ is the upper ( )2100 ω percentile of the standard normal
distribution and ( )⋅var denote the diagonal elements of ( )Θ−1I
corresponding to the model parameters.
6. Applications
In this section, we explain the flexibility and superiority of new KwWE as special distribution from the Kumaraswamy Weibull-G family proposed here. We examine three real data sets to compare the fits of KwWE with exponentiated Weibull exponential (EWE) distribution and Weibull exponential (WE) distribution with the following corresponding densities:
( ) [ ] { ( ) }( ( ) ) 1111 11 −−α−λ−−α−−βλ βλβλ−−αβλ= aexex
KwWExx
eeeabxf
[ ( ( ) ) ] ,0,0,,,,,11 11 >>λβα−−× −−α− βλxbae bae x
( ) [ ] { ( ) }( ( ) ) ,11 1111 −−α−λ−−α−−βλ βλβλ−−αβλ= aexex
EWExx
eeeaxf
,0,0,,, >>λβα xa
and
( ) [ ] { ( ) } .0,0,,,1 11 >>λβα−αβλ= λ−−α−−βλ βλxeexf xex
WEx
Amal S. Hassan and M. Elgarhy 230
In each real data set, the parameters are estimated by maximum likelihood method. Akaike information criterion (AIC), the correct Akaike information criterion (CAIC), Bayesian information criterion (BIC), Hannan information criterion (HQIC), –2 ln, L, the Kolmogorov-Smirnov ( )SK - and
p-value statistics are considered to compare the three models. The formula for these criteria is as follows:
( ) ,112,ln22−−++=−= kn
kkAICCAICLkAIC
( ) ( )[ ] ,ln2lnln2andln2ln LnkHQICLnkBIC −=−=
where k is the number of parameters in the statistical model, n is the sample size and ln L is the maximized value of the log-likelihood function under the
considered model. Also, ( ) ( )[ ],sup yFyFsk ny −=− where ( ) nyFn1=
(number of observation ,)y≤ and ( )yF denotes the cdf.
In general, the best distribution corresponds to the smallest values of SKHQICCAICBICAICL -,,,,,ln2− and the biggest value of p-value
criteria.
6.1. The first real data set
The first data set was taken from Hinkley [9], which consists of thirty successive values of March precipitation (in inches) in Minneapolis/St Paul. The data set is as follows:
0.77, 1.74, 0.81, 1.20, 1.95, 1.20, 0.47, 1.43, 3.37, 2.20, 3.00, 3.09, 1.51, 2.10, 0.52, 1.62, 1.31, 0.32, 0.59, 0.81, 2.81, 1.87, 1.18, 1.35, 4.75, 2.48, 0.96, 1.89, 0.90, 2.05.
Table 1 provides the maximum likelihood estimates (MLEs) of the model parameters. Also, the numerical values of statistics, AIC, BIC, CAIC, HQIC and Kolmogorov-Smirnov values as well as the p-value statistics are listed in Table 1. Figures 3 and 4 provide the plots of estimated cumulative and estimated densities of the fitted KwWE, EWE and WE models for the first data set.
Kumaraswamy Weibull-generated Family of Distributions … 231
Table 1. The S-KHQICBICCAICAICMLEs ,,,,, and p-value of the
models based on the first data set
Model
KwWE EWE WE
901.3ˆ =α 402.12ˆ =α -
001.32ˆ =b - -
583.8ˆ =α 016.39ˆ =α 708.168ˆ =α
578.0ˆ =β 581.0ˆ =β 572.1ˆ =β
Parameter Estimates
310328.4ˆ −×=λ 310066.8ˆ −×=λ 02.0ˆ =λ
–2 ln L 108.217 115.109 117.798
AIC 118.217 123.109 123.798
CAIC 120.717 124.709 124.721
BIC 115.603 121.017 122.229
HQIC 120.458 124.902 125.142
K-S 0.06 0.089 0.0796
p-value 0.9999 0.971 0.991
Based on the values of AIC, CAIC, BIC, HQIC, K-S and p-value in Table 1, the KwWE model provides better fit than the EWE and WE models.
Amal S. Hassan and M. Elgarhy 232
Figure 3. Estimated cumulative densities for the first data.
Figure 4. Estimated densities of models for the first data.
Kumaraswamy Weibull-generated Family of Distributions … 233
It is clear from Figure 4 that the fitted density for KwWE model is closer to the empirical histogram than EWE and WE models.
6.2. The second real data set
The second data set was taken from Murthy et al. [10]. The corresponding data is referring to the time between failures for a repairable item. The data set is as follows:
1.43, 0.11, 0.71, 0.77, 2.63, 1.49, 3.46, 2.46, 0.59, 0.74, 1.23, 0.94, 4.36, 0.40, 1.74, 4.73, 2.23, 0.45, 0.70, 1.06, 1.46, 0.30, 1.82, 2.37, 0.63, 1.23, 1.24, 1.97, 1.86, 1.17.
The following Table 2 provides MLEs, the AIC, BIC, CAIC, HQIC and K-S values as well as the p-value statistics for the second data set.
Table 2. The MLEs, AIC, CAIC, BIC, HQIC, K-S and p-value of the models based on second data set
Model
KwWE EWE WE
797.2ˆ =α 355.6ˆ =α -
55.29ˆ =b - -
521.6ˆ =α 227.30ˆ =α 933.127ˆ =α
593.0ˆ =β 58.0ˆ =β 318.1ˆ =β
Parameter estimates
310326.4ˆ −×=λ 310347.9ˆ −×=λ 015.0ˆ =λ
–2 ln L 106.964 113.344 111.901
AIC 116.964 121.344 117.901
CAIC 119.464 122.944 118.824
BIC 114.35 119.252 116.332
HQIC 119.205 123.137 119.246
K-S 0.062 0.078 0.083
p-value 0.9998 0.993 0.987
Amal S. Hassan and M. Elgarhy 234
In order to assess whether the KwWE model is appropriate, Figures 5 and 6 provide some plots of the estimated cdf as well as the estimated probability densities of the fitted KwWE, EWE and WE models for the second data set.
Figure 5. Estimated cumulative densities for the second data.
Figure 6. Estimated densities of models for the second data.
Kumaraswamy Weibull-generated Family of Distributions … 235
From Figure 6, it is concluded that the KwWE distribution is quite suitable for this data set.
6.3. The third real data set
The following data represent the survival times (in days) of 72 guinea pigs infected with virulent tubercle bacilli, observed and reported by Bjerkedal [2]. The data is as follows:
0.1, 0.33, 0.44, 0.56, 0.59, 0.72, 0.74, 0.77, 0.92, 0.93, 0.96, 1, 1, 1.02, 1.05, 1.07, 07, .08, 1.08, 1.08, 1.09, 1.12, 1.13, 1.15, 1.16, 1.2, 1.21, 1.22, 1.22, 1.24, 1.3, 1.34, 1.36, 1.39, 1.44, 1.46, 1.53, 1.59, 1.6, 1.63, 1.63, 1.68, 1.71, 1.72, 1.76, 1.83, 1.95, 1.96, 1.97, 2.02, 2.13, 2.15, 2.16, 2.22, 2.3, 2.31, 2.4, 2.45, 2.51, 2.53, 2.54, 2.54, 2.78, 2.93, 3.27, 3.42, 3.47, 3.61, 4.02, 4.32, 4.58, 5.55.
The following Table 3 gives the MLEs, AIC, BIC, CAIC, HQIC and K-S values as well as the p-value statistics for the third data set.
Table 3. The MLEs, AIC, CAIC, BIC, HQIC, K-S and p-value of the models based on third data set
Model
KwWE EWE WE
494.4ˆ =α 973.10ˆ =α -
374.22ˆ =b - -
739.9ˆ =α 229.32ˆ =α 738.163ˆ =α
556.0ˆ =β 601.0ˆ =β 584.1ˆ =β
Parameter Estimates
31037.4ˆ −×=λ 011.0ˆ =λ 02.0ˆ =λ
–2 ln L 265.787 283.242 292.659
AIC 275.787 291.242 298.659
Amal S. Hassan and M. Elgarhy 236
CAIC 278.287 292.842 299.582
BIC 275.074 290.671 298.036
HQIC 280.319 294.867 301.378
K-S 0.105 0.111 0.1394
p-value 0.4015 0.338 0.122
The values in Table 3, indicate that the KwWE model is a strong competitor among other distributions that was used here to fit this data set. Figures 7 and 8 provide some plots of the estimated cdf and pdf of the fitted KwWE, EWE and WE models for the third data set.
Figure 7. Estimated cumulative densities for the third data.
Kumaraswamy Weibull-generated Family of Distributions … 237
Figure 8. Estimated densities of models for the third data.
Based on the plots of Figure 8 one can notice that the fitted density of the KwWE model is closer to the empirical histogram than the corresponding densities of the EWE and WE models.
7. Conclusion
In view of an idea of the family of Weibull-G distributions and Kumaraswamy distribution, we introduce a new family of distributions, called Kumaraswamy Weibull-G. The Kumaraswamy Weibull-G family extends several widely known distributions. An expansion for the density function and explicit expressions for the moments, probability weighted moments, generating function, mean deviation, quantile function and order statistics are obtained. The estimation of parameters is approached by the method of maximum likelihood. Some applications of the new family to three real data are given to demonstrate the importance and usefulness of the suggested family. In this study, we wish that this new generalization will
Amal S. Hassan and M. Elgarhy 238
attract a broadly applications in several areas such as engineering, survival and lifetime data, hydrology, etc.
Acknowledgment
The authors are thankful to the editor and a reviewer whose comments led to improvement of the article.
References
[1] A. Alzaatreh, C. Lee and F. Famoye, A new method for generating families of continuous distributions, Metron 71 (2013), 63-79.
[2] T. Bjerkedal, Acquisition of resistance in guinea pies infected with different doses of virulent tubercle bacilli, American Journal of Hygiene 72 (1960), 130-148.
[3] M. Bourguignon, R. B. Silva and G. M. Cordeiro, The Weibull-G family of probability distributions, Journal of Data Science 12 (2014), 53-68.
[4] G. M. Cordeiro and M. de Castro, A new family of generalized distribution, J. Statist. Comput. Simul. 81 (2011), 883-898.
[5] G. M. Cordeiro, E. M. M. Ortega and D. C. C. da Cunha, The exponentiated generalized class of distributions, Journal of Data Science 11 (2013), 1-27.
[6] G. M. Cordeiro, E. M. M. Ortega and S. Nadarajah, The Kumaraswamy Weibull distribution with application to failure data, Journal of the Franklin Institute 347 (2010), 1399-1429.
[7] H. A. David, Order Statistics, 2nd ed., Wiley, New York, 1981.
[8] N. Eugene, C. Lee and F. Famoye, The beta-normal distribution and its applications, Commun. Statist. Theory Methods 31 (2002), 497-512.
[9] D. Hinkley, On quick choice of power transformations, Journal of the Royal Statistical Society, Series (c), Applied Statistics 26 (1977), 67-69.
[10] D. N. P. Murthy, M. Xie and R. Jiang, Weibull Models, John Wiley, New Jersey, 2004.
[11] R. R. Pescim, G. M. Cordeiro, C. G. B. Demetrio, E. M. M. Ortega and S. Nadarajah, The new class of Kummer beta generalized distributions, Statistics and Operations Research Transactions 36 (2012), 153-180.
Kumaraswamy Weibull-generated Family of Distributions … 239
[12] R. Shanker and A. Mishra, A quasi Lindley distribution, African J. Math. Comput. Sci. Res. 6 (2013), 64-71.
[13] M. M. Ristic and N. Balakrishnan, The gamma-exponentiated exponential distribution, J. Statist. Comput. Simul. 82 (2012), 1191-1206.
[14] K. Zografos and N. Balakrishnan, On families of beta- and generalized gamma-generated distributions and associated inference, Statist. Method. 6 (2009), 344-362.