graphical methods for selecting factorial effects
TRANSCRIPT
1
Making the most of this learning opportunityStat-Ease worldwide webinars attract many attendees, so, to prevent audio disruptions, all must be muted by the presenter. Also, to avoid interruptions and keep the presentation to about one hour, please hold all questions until afterwards and address them to [email protected].
– Mark
P.S. Find slides posted now at www.statease.com/webinar.html and, barring technical issues, a recording put up afterwards.
By Mark J. Anderson, PE, CQEStat-Ease, Inc., Minneapolis, MN
Graphical methods for selecting factorial effects: What’s In It for You
Graphical Methods for Selecting Factorial Effects
0.00 3.60 7.21 10.81 14.42 18.02 21.63
0
10
20
30
50
70
80
90
95
99
Half-Normal Plot
|Standardized Effect|
Half-N
orm
al %
Pro
bability
A-Temperature
C-ConcentrationD-Stir Rate
AC
AD
0.00
1.40
2.80
4.20
5.59
6.99
8.39
9.79
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Pareto Chart
Rank
t-V
alu
e o
f |E
ffe
ct|
Bonferroni Limit 3.82734
t-Value Limit 2.22814
A-Temperature
AC
AD
D-Stir Rate
C-Concentration
©2017
Stat
-Eas
e, Inc
.
The WIIFM for this Webinar
By way of a variety of case studies, this webinar on design of experiments (DOE) provides insights into graphical approaches—half-normal and Pareto plots—that assess effects at a glance—a huge advantage for experimenters who get overwhelmed by esoteric statistical reports. It details an impressive number of enhancements incorporated in Design-Expert® and Design-Ease® software to:
deploy these graphical tools over a broad array of factorial designs (including multilevel categoric and split plot),
make them more robust to missing data and non-orthogonal levels, and
provide interactive layouts that lead to good decisions on sorting out the vital few effects from the trivial many.
See how Stat-Ease makes selection of factor effects easy for its users!
Graphical Methods for Selecting Factorial Effects
2
Please press the raise hand now if you are with me.©2017
Stat
-Eas
e, Inc
.
Graphical Methods for Selecting Factorial Effects
3
Reference: DOE Simplified
FormulationSIMPLIFIED
Finding the Sweet Spot via Design and Analysis of
Experiments with Mixtures
A Primer on Mixture Design: What’s In It
for Formulators?www.statease.com/pubs/MIXprimer.pdf
2nd edition 2016.
* Productivity Press CRC, Taylor & FrancisNew York, June 2015.
Now in 3rd edition.*
A: Apple
B: Cinnamon C: Lemon
Taste
100
100
100
75
75
75
50
50
50
25
25
25
0
0
0
3
3.5
3.5
4
4
4.55
5.56
6.57
2
2 2
2
1st edition 2017!
©2017
Stat
-Eas
e, Inc
.
The Half-Normal (aka “Daniel”) Plot (1/2)
In 1959, Cuthbert Daniel introduced an innovative way to graphically judge the “reality of the observed effects” from two-level factorial designs. This proved to be especially helpful for unreplicated experiments such as the one illustrated here.
Graphical Methods for Selecting Factorial Effects
4
0.00 3.60 7.21 10.81 14.42 18.02 21.63
0
10
20
30
50
70
80
90
95
99
Half-Normal Plot
|Standardized Effect|
Ha
lf-N
orm
al %
Pro
bab
ility
A-Temperature
C-ConcentrationD-Stir Rate
AC
AD
Filtrate (hold off demo until Pareto)
©2017
Stat
-Eas
e, Inc
.
The Half-Normal (aka “Daniel”) Plot (2/2)A funny story…
Daniel laid out a complete primer (pictured) on how apply these innovative plots in his 1976 text (Wiley) Applications of Statistics to Industrial Experimentation. It featured full-normal plots, which he then favored (vs the half-normal plots in Daniel’s original paper of ’59).
Box, Hunter & Hunter, in their classic 1979 (Wiley) book Statistics for Experimenters (note my Box-autographed copy), popularized “Daniel” plots. At the 2001 ASQ World Quality Conference in Charlotte, I gave a talk featuring half-normal plots that Stu Hunter attended. He told me that BHH wrote their manuscript using half-normals, but Daniel talked them into going to the fullnormal, so they laboriously reworked all their effects plots. Just before publication, Daniel changed his mind and recommended the half normal. By then it was too late to change.
Stat-Ease favors half-normal plots as I will explain later….
Graphical Methods for Selecting Factorial Effects
5
©2017
Stat
-Eas
e, Inc
.
Backlash against graphical methods
IMO a graph is worth 1000 numbers (apologies to Confucious). However, resistance can be mighty from statisticians such as those who fought me introducing the half-normal to ASTM 1169-14 Standard Practice for Conducting Ruggedness Tests.
Russel V. Lenth brought this to a head recently in “The Case Against Normal Plots of Effects”.* A number of discussants came to the defense.
Graphical Methods for Selecting Factorial Effects
6
Cartoon by John Landers, Nov ’16 Cause Web captioning contest**** www.causeweb.org/cause/caption-contest/november/2016/results
*(Journal of Quality Technology, V47, N2, April 2015, pp 91-97) ©20
17 S
tat-E
ase,
Inc.
Keeping to the positive plane:
How Half-Normal Plots Help Experimenters
7Graphical Methods for Selecting Factorial Effects
Half-Normal plots (aka “Daniel”) provide many advantages as enumerated* by David Steinberg, Prof Stats, Tel Aviv U, who achieved his PhD at U Wisconsin under supervision of George Box:
Easy to produce Gives a quick visual of important effects Effects can be compared to a reference—the near-zero
contrasts that emanate from the origin Adapts to split-plot experiments
“The greatest single advantage of the Daniel plot is its ability to stimulate discussion by encapsulating, in a single display, such a variety of information.”
* (“Discussion: On Daniel Plots,” Journal of Quality Technology, V47, N2, April 2015, p110.)©2017
Stat
-Eas
e, Inc
.
Pareto Chart of Effects (1/2)
A perfect companion to the half-normal plot
Graphical Methods for Selecting Factorial Effects
8
I used Pareto plots in the the ’70s to focus my quality team on the 20% of causes for the 80% of problems. However, my first exposure to this tool being applied to effect selection came from Don Wheeler’s scree plot*--the only difference from a Pareto being the form of graphing, line (Scree) vs bar (Pareto). (“Scree” refers to a slope of loose stones, such as that pictured.) Wheeler advised that effects above the “elbow” (point of inflection) be considered significant, with the remaining “rubble” relegated to the error pool.
*Understanding Industrial Experimentation, Statistical Process Controls, Knoxville, TN, 1987. ©2017
Stat
-Eas
e, Inc
.
Pareto Chart of Effects (2/2)
A perfect companion to the half-normal plot
Graphical Methods for Selecting Factorial Effects
9
0.00
1.40
2.80
4.20
5.59
6.99
8.39
9.79
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
Pareto Chart
Rank
t-V
alu
e o
f |E
ffe
ct|
Bonferroni Limit 3.82734
t-Value Limit 2.22814
A-Temperature
AC
AD
D-Stir Rate
C-Concentration
Stat-Ease software makes it far easier to interpret the ordered chart of effects by it’s innovative enhancements: Two limit lines Color coding* Interactivity*
* These features also on half-normal
Filtrate (demo Half-normal & Pareto selection tools)©2017
Stat
-Eas
e, Inc
.
Completing the trifecta for effect selection:
ANOVA’s p-Values
Graphical Methods for Selecting Factorial Effects
10
Analysis of variance table
Sum of Mean F p-value
Source Squares df Square Value Prob > F
Model 5535.81 5 1107.16 56.74 < 0.0001
A-Temperature 1870.56 1 1870.56 95.86 < 0.0001
C-Concentration 390.06 1 390.06 19.99 0.0012
D-Stir Rate 855.56 1 855.56 43.85 < 0.0001
AC 1314.06 1 1314.06 67.34 < 0.0001
AD 1105.56 1 1105.56 56.66 < 0.0001
Residual 195.12 10 19.51
Cor Total 5730.94 15
Graphs and stats can all get along—best of both worlds!©2017
Stat
-Eas
e, Inc
.
Going back a step to fill in:
Full-Normal Plot Enhanced by New Pre-Set
Graphical Methods for Selecting Factorial Effects
11
Go to www.youtube.com/channels & search on Stat-EasePlay “mini-Tip 1 - Graphical selection tools for factorial effects”.
Re-do live in slow-motion in DX with Filtrate file showing 100% vs 50% preset.
Version 10 of Design-Ease and Design-Expert presets the line to the smallest 50% of effects on half and full normal plots. * This is most helpful on the full normal plot where the line is not anchored at the origin, thus, otherwise (without preset), making it unfriendly to unwary users.
*(Originally suggested by Doug Montgomery in his rebuttal to “The Case Against Normal Plots of Effects” by Russell Lenth. See Journal of Quality Technology, Vol. 47, No. 2, April 2015, “Discussion 3”, pp 105-106.)
©2017
Stat
-Eas
e, Inc
.
Other Upgrades to Graphical Tools
Stat-Ease, led by our founder Pat Whitcomb, has been “all in” on graphical selection of effects from the Company’s beginning in 1982. With invaluable input from advisors Kinley Larntz and Gary Oehlert, the Stat-Ease team of consultants and programmers have put together an impressive array of enhancements that make the statistics easy for experimenters using Design-Ease and Design-Expert. My favorites are:
Pure error on half/full normal (answer to those who say these plots are only useful for unreplicated factorials)
Preselected half-normal plots for multilevel categoric factorials
Dual half/full normal plots for split plots (an approach I first saw presented by Box & Bisgaard at a ’95 DOE workshop in Madison, WI)
Graphical Methods for Selecting Factorial Effects
12
BHH325C (w Cp’s), Slinky2 (GenFac), Paper-helicopter (Split Plot)
©2017
Stat
-Eas
e, Inc
.
More Innovative Features
On all effect-selection graphs (i.e., half/full normal & Pareto) the error pools update based on terms picked—an interactive feature that provides a major advantage over manual graphs. (Not to be taken for granted!)
Effects standardized to contend with missing data and/or altered independent factors, which can cause the variance associated with the estimated effects to differ. (Also handles non-orthogonal two-level designs such as the amazing minimum-run resolution IV (MR4) and MR5s for screening and characterization; respectively.)
Selection of effects by order (i.e., main effects then two factor interactions, etc.), going forward when degrees of freedom do not suffice to estimate an entire group of terms. Using forward selection when an order cannot be completely estimated, thus providing robustness to missing data.
Graphical Methods for Selecting Factorial Effects
13
Reactor half fraction (1 run ignored, yet DE emerges)©20
17 S
tat-E
ase,
Inc.
Some Other Nifty Touches
Ctrl and shift click and right-click to reveal effect sizes. (This effects finders also copies to the clipboard.)
Right click on a selected effect for aliasing.
Substitute aliased effects.
Graphical Methods for Selecting Factorial Effects
14
©2017
Stat
-Eas
e, Inc
.
Not to be overlooked: Scatterplots
“Of all the graphic forms used today, the scatterplot is arguably the most versatile…and useful invention in the history of statistical graphics.”
Stat-Ease software provides via its “graph columns” a great way to see effects simply displayed on scatterplots. Let’s circle back to a prior example and see how this tool makes the effects graphs easy.
*Michael Friendly & Daniel Denis, “The Early Origins and Development of the Scatterplot”, Journal of the History of Behavioral Sciences, V41(2), 103-120, Spring 2005. (They credit J.F.W. Herschel for inventing this plot in 1833.)
Graphical Methods for Selecting Factorial Effects
15
Filtrate
©2017
Stat
-Eas
e, Inc
.
The WIIFM for this Webinar
By way of a variety of case studies, this webinar on design of experiments (DOE) provides insights into graphical approaches—half-normal and Pareto plots—that assess effects at a glance—a huge advantage for experimenters who get overwhelmed by esoteric statistical reports. It details an impressive number of enhancements incorporated in Design-Expert® and Design-Ease® software to:
deploy these graphical tools over a broad array of factorial designs (including multilevel categoric and split plot),
make them more robust to missing data and non-orthogonal levels, and
provide interactive layouts that lead to good decisions on sorting out the vital few effects from the trivial many.
See how Stat-Ease makes selection of factor effects easy for its users!
Now you know.
Graphical Methods for Selecting Factorial Effects
16
©2017
Stat
-Eas
e, Inc
.
Where to Get Statistical Details
1. Larntz, K. and Whitcomb, P. (1992). “The Role of Pure Error on Normal Probability Plots” Transactions of Annual Quality Congress of the American Society of Quality, Milwaukee, WI. www.statease.com/pubs/role-of-pure-error.pdf
2. Larntz, K. and Whitcomb, P. (1998). “Use of Replication in Almost UnreplicatedFactorials,” Fall Technical Conference, Corning, NY. www.statease.com/pubs/use-of-rep.pdf
3. Whitcomb, P. and Oehlert, G. (2007). “Graphical Selection of Effects in General Factorials,” Fall Technical Conference, Jacksonville, FL. www.statease.com/pubs/2023%20-%20Graphical%20selection%20of%20effects.pdf
4. Oehlert, G. (2009). “Graphical Methods for Selecting Effects in Factorial Models,” Statistics Workshop at Yale University. www.stat.yale.edu/Conferences/Stats2009/GOSLIDES.pdf
5. Design-Expert® software version 10 Help topic: “Calculation of Effects”.
Graphical Methods for Selecting Factorial Effects
17
©2017
Stat
-Eas
e, Inc
.
181818
Stat-Ease Training: Sharpen Up via Computer-Intensive Workshops
Shari Kraber,Workshop Manager & Master [email protected]
PreDOEWeb-Based (optional)
Stat-Ease Academy
Experiment DesignMade Easy
Response Surface Methods for Process Optimization
Mixture and CombinedDesigns for
Optimal Formulations
Robust Designand
Tolerance Analysis
Basic Statistics forDesign of Experiments
Designed Experiments forPharma, Life Sciences,
Assay Optimization, Food Science
Factorial Split-Plot Designs
Stat-Ease Academy
www.statease.com/training/stat-ease-academy.html
Plus more classes here!Modern DOE for Process Optimization
Graphical Methods for Selecting Factorial Effects
©2017
Stat
-Eas
e, Inc
.
Statistics Made Easy®
Graphical Methods for Selecting Factorial Effects
19
Best of luck for your experimenting!
Thanks for listening!
-- Mark
“The greatest value of a picture is when it forces us to notice what we never expected to see.”
- John Tukey, Exploratory Data Analysis, 1977, p. vi.©2017
Stat
-Eas
e, Inc
.