>>>>>>>>>>>>>>>>>>>>>>>>>>DOE FAQ Alert<<<<<<<<<<<<<<<<<<<<<<<<<< Issue: Volume 1, Number 3 Date: May 2001 From: Mark J. Anderson, Stat-Ease, Inc. (http://www.statease.com) "Statistics Made Easy" (tm) Dear Experimenter, Here's our third issue in an ongoing series of e-mails with answers to frequently asked questions (FAQs) about doing design of experiments (DOE), plus alerts to timely information and free software updates. If you missed the prior DOE FAQ Alert (or earlier ones), go to http://www.statease.com/doealert.html . Feel free to forward this newsletter to your colleagues. It's easy to subscribe (or unsubscribe) - just follow the instructions at the end of this message. Before I get into the meat of this message, I offer this appetizer: click on http://www.stat.ucla.edu/history/ for links (not all of which work, unfortunately) to lots of interesting historical tidbits on statistics, one of which (the picture of a Latin square) I referred to in my last issue. (Can you suggest any links that experimenters might find fun and interesting? If so, send me an e-mail with the link embedded.) Here's what I cover in the body text of this DOE FAQ Alert (topics that delve into statistical detail are rated "X"): 1. X-FAQ: The negative impact of ignored run(s) on alias structure, a sequel to FAQ 1-2-1 (previous issue) - Botched runs in fractional two-level factorial design 2. FAQs (two): Import/export of data to/from Design-Expert(R) software, specifically for setting up a customized design 3. X-FAQ (Six Sigma): Using a Response Surface Method (RSM) optimization feature called propagation of error (POE) to minimize variability 4. Workshop alert: Coming soon to Philadelphia and Detroit 5. Events alert: A heads-up on DOE talks and demos. 6. Info alert: Link to educational article (very basic) on DOE ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 1 - FAQ: Botched runs in fractional two-level factorial design 1b. X-FAQ: The negative impact of ignored run(s) on alias structure and how to repair the damage. -----Original Question (Part b)----- From: California (sequel to FAQ 1-2-1 (previous issue)) "We conducted a 4-factor [8 run] half-fractional factorial. During one run the temperature rose above the upper limit. I threw it [the run with temperature too high] out because the results got peculiar. The initial alias list indicated that there would be confounding on the two-way interactions, etc, but the main effects would be free. However, the design [evaluation] now indicates that there is confounding among the main effects as well. I do not understand what happened here." Answer: To keep things simple, let's assume you perform a standard 2^4-1 (half-fraction of four factors at two-levels) with no centerpoints or blocking. The alias structure for this design is: [A] = A + BCD [B] = B + ACD [C] = C + ABD [D] = D + ABC [AB] = AB + CD [AC] = AC + BD [AD] = AD + BC which is Resolution IV. This is acceptable for screening designs, assuming you can follow up if any two-factor interactions (2fi's) come out significant. However, after throwing out a "peculiar" result, only 7 runs remain to estimate 7 effects and the overall average, which can't work - you've used up all your degrees of freedom. As a result your resolution degrades. For example, let's say you ignore standard order #8. The alias structure for this botched design is now: [A] = A - AD - BC + BCD [B] = B - AD - BC + ACD [C] = C - AD - BC + ABD [D] = D - AD - BC + ABC [AB] = AB - AD - BC + CD [AC] = AC - AD - BC + BD which is Resolution III (main effects confounded with 2fi's). To repair the design you must perform at least two additional runs: one to replace the one you botched, and the other to deal with the fact that this will be a second block. Design-Expert will set this up for you via a nifty repair feature called "Factorial D-optimal" that you access via Design Tools, Augment Design. You many want to call Stat-Ease and ask for statistical help before trying this. To avoid all this it would be tempting to just sneak back and re-do the botched run, but of course this would not be a good idea if anything may have changed in the interim. In future, when you want to experiment on 4 factors, consider doing a larger design - either the 16-run full factorial, or an irregular (3/4th) fraction of 12 runs (for an enlightening example see http://www.statease.com/news/news9707.pdf ). Then you can afford one or more botched runs and still achieve at least resolution IV for screening purposes. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2 - FAQs (two): Import/export of data to/from Design-Expert (DX), specifically for setting up a customized design -----Original Question (First on this topic)----- From: Design-Expert Help System's FAQ file "How can I import or export data with Design-Expert software?" Answer: >Data can be imported and exported by using the standard cut and paste operations from Windows. To export, highlight the group of cells (or Select All) and select copy from the View menu (or Ctrl-C). Go to the program you want to paste to and select paste (Ctrl-V). If you are trying to import a new design from a spreadsheet, first set up a similar dummy design in Design-Expert making sure that the factor levels are correctly specified. [Set your -1 coded value to the smallest number you will paste in and the +1 value to the largest number in your data set.] This will ensure the coded values of the independent factors are centered on zero.] Duplicate or Delete rows in the design layout screen to get the desired number of rows. Last, copy and paste the data from the other source into the design layout over the old (dummy) data.< (This FAQ answer came from the Design-Expert Help System authored by Shari Kraber, Consultant, Stat-Ease, Inc. [Bracketed point of clarification added by Pat Whitcomb.] If you own a copy of DX6, be sure to search Help first whenever you have a question, not only about the program itself, but also about any statistical aspect - it's a fabulous educational resource.) PS from Mark: When pasting data into Design-Expert or Design-Ease, you must highlight the identical number of columns and rows for your destination area as you've copied to the clipboard from Excel or the like. -----Original Question (Second on this topic)----- From: South Africa "My efforts to get the data into DX6 has frustrated me seriously! If I paste it in, four of the factors are not accepted in the data sheet with the message: "Values are outside the range for coded values ... " (which I understand). I am sure that you will laugh at this but I am just plain stuck!" Answer: >The problem you were having is that the actual lows and highs for all the factors were set to the default levels of -1 and 1. To import the data properly you have to change the -1 and +1 coded levels to the values of the actual lows and highs for each factor before you copy the data into the design. This can be done by right-clicking on the factor header on the design layout screen and choosing Edit Info to change the -1 and +1 values. You can also enter this information in the low and high fields for the factor when you build the dummy design that the information will be imported into.< (Answer from: Neal Vaughn, Programmer, Stat-Ease, Inc.) PS from Mark: We did NOT laugh about this problem, because it's happened to the best of our users. Hopefully, publicizing it via this FAQ Alert will keep the rest of you out of trouble. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3 - X-FAQ (Six Sigma): Using a Response Surface Method (RSM) optimization feature called propagation of error (POE) to minimize variability -----Original Message----- From: A user who participates in a Six-Sigma program "I have DOE's running that were setup using Design-Expert. During the analysis of these DOE's, an important optimization criterion will be the minimization of the variance of one of the responses. I do not see that Design-Expert will allow me to optimize my design based on the minimization of the variance of one of the responses. Is there a way DX can do this?" Answer: Yes, Design-Expert offers a feature called propagation of error (POE) that finds regions in your response surface that will be most robust to variations in input factors. To take advantage of POE you must do a response surface methods (RSM) design and get a non-linear predictive model. Then, of course, you must determine the variability of your input factors and enter their standard deviations into DX software. (You will find a tutorial showing how to apply POE in Section 6, page 34 of the Design-Expert version 6 User's Guide. If you don't possess a copy of the Guide, see: http://www.statease.com/x6ug/DX06-RSM.pdf .) Another approach to minimizing variability, called "dual response", involves the actual measurement of standard deviation (sigma) at each setup (experimental run). We show how this is done in our workshop via an example on printing ink. The experimenters varied speed, pressure and distance each at three levels and replicated each printing setup three times. Then they measured the mean response and standard deviation. After fitting models to each response (suggestion: take log of standard deviation first), Design-Expert can be utilized to optimize the mean (to hit a target in this case) and minimize standard deviation. *(See http://www.statease.com/clasrdrv.html for course content of the "Robust Design, DOE Tools for Reducing Variability" workshop. It requires proficiency in RSM which can be gained by attending Stat-Ease's three-day computer-intensive workshop called "Response Surface Methods for Process Optimization". See http://www.statease.com/clas_rsm.html for a description and links to the RSM course outline and schedule.) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 4 - Workshop alert: Coming soon to Philadelphia and Detroit The "Robust Design, DOE Tools for Reducing Variability" workshop noted above will be presented in Philadelphia on June 5th. The following month (July 12th), Stat-Ease returns to the city of brotherly love (we like it so much!) with "DOE Simplified" (DOES), a one-day presentation based on the book of the same name. It's a "concise, upbeat, jocular, real-life" class (actual quote from a recent participant) that provides an overview of the DOE tool set. See http://www.statease.com/does.html for class content. Although "DOE Simplified" is fun and informative, it's only intended get people started on the path to more effective experimentation. We hope that participants will then be motivated to take the next step by attending our "Experiment Design Made Easy" (EDME) workshop, which will be presented next (June 5) in Detroit (see http://www.statease.com/clasedme.html ). Refer to http://www.statease.com/clas_pub.html for a complete schedule and listing of sites (not just Philadelphia and Detroit) for all Stat-Ease workshops open to the public. To enroll, call Stat-Ease at 612-378-9449. Don't delay, seats sometimes fill in fast. If spots remain available, bring along several colleagues and take advantage of quantity discounts in tuition, or consider bringing in an expert from Stat-Ease to teach a private class at your site. Call us to get a quote. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 5 - Events alert: A heads-up on DOE talks and demos. Click on http://www.statease.com/events.html for a listing of where Stat-Ease consultants will be giving talks and doing DOE demos. We hope to see you sometime in the near future! ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 6 - Info Alert: Go to http://www.qualitymag.com/articles/2000/apr00/0400f2.bak (this link redirects to main page, archives only go back to 2001) for an article that originally appeared in Quality magazine entitled "Success with DOE." The content is taken from "DOE Simplified: Practical Tools for Effective Experimentation" by Mark J. Anderson and Patrick J. Whitcomb (Copyright 2000, Productivity Inc., Portland, OR). To details on this soft-cover primer on DOE go to http://www.statease.com/doe_simp.html . It can be ordered online from http://www.statease.com/prodbook.html . ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ I hope you learned something from this issue. Address your questions and comments to me at: Mark@StatEase.com PLEASE DO NOT SEND ME REQUESTS TO SUBSCRIBE OR UNSUBSCRIBE - FOLLOW THE INSTRUCTIONS AT THE END OF THIS MESSAGE. Sincerely, Mark Mark J. Anderson, PE, CQE Principal, Stat-Ease, Inc. (http://www.statease.com) Minneapolis, Minnesota USA "Logical consequences are the scarecrows of fools and the beacons of wise men." - Thomas Henry Huxley Trademarks: Design-Ease and Design-Expert are registered trade- marks of Stat-Ease, Inc. Acknowledgements to contributors: - Students of Stat-Ease training and users of Stat-Ease software - Fellow Stat-Ease consultants Pat Whitcomb and Shari Kraber (see http://www.statease.com/consult.html for resumes) - Statistical advisor to Stat-Ease: Dr. Gary Oehlert ( http://www.statease.com/garyoehl.html ) - Stat-Ease programmers, especially Tryg Helseth ( http://www.statease.com/pgmstaff.html ) - Heidi Hansel, Stat-Ease communications specialist, and all the remaining staff DOE FAQ Alert - Copyright 2001 Stat-Ease, Inc. All rights reserved. Subscribe (join our list) by going to: http://www.statease.com/doealert