>>>>>>>>>>>>>>>>>>>>>>>>>>DOE FAQ Alert<<<<<<<<<<<<<<<<<<<<<<<<<< Issue: Volume 1, Number 1 Date: March 1, 2001 From: Mark J. Anderson, Stat-Ease, Inc. (http://www.statease.com) "Statistics Made Easy" (tm) Dear Experimenter, Here's our first 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 articles and free software updates. In this message, we discuss: 1. FAQ: How to Handle Multiple Factors with Varying Number of Levels (Factorial Design) 2. FAQ: Mission Impossible: Predictions That Go Out Of Bounds (Response Surface Methods) 3. Upgrade Patch Available for Design-Expert(R) Version 6 Software (link to free download) 4. Alert: Talk on semifold at Annual Quality Congress (link to free download) 5. Alert: "Stat-Teaser" Newsletter Featuring the Article "Mad Mark and the Meter Gnomes" (link to free download) You received this newsletter because of your expressed interest in DOE via an inquiry to Stat-Ease, or registration as a user of the company's software and/or status as a graduate of its DOE workshops. It's easy to unsubscribe (leave our list) - just follow the instructions at the bottom (footer) of this message. Feel free to forward the DOE FAQ Alert e-mail to any friends and colleagues who need tools for doing DOE. They can then subscribe (join our list) if they like by going to the Internet link shown at the end of this message. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 1 - FAQ: How to Handle Multiple Factors with Varying Levels -----Original Question----- From: South Africa "I have a mixed 2-level & 3-level design (2^3x3^3). It has to be a fractional factorial since the full factorial is too costly to run 216 combinations (= 2x2x2x3x3x3)." Answer: In version 6 of Design-Expert we now support design of experiments with up to 12 categorical factors at up to 20 specific levels, so long as the total number of combinations does not exceed 32,766 rows. To do the full factorial, go to the Factorial tab on Design-Expert and choose General Factorial. However, if you want to cut back on the number of runs, choose the D-optimal design. The default model estimates up to the two-factor interactions. You can edit the model if you like, thus increasing or decreasing the number of model points required in the d-optimal design. If all or most of your factors are continuous, such as time or temperature, but you only want a discrete number of levels for each, then you could try using the D-Optimal option on the Response Surface tab. However you will first need to create a proper candidate set by going to Design Tools, Create Factorial Candidates. For detailed instructions on this, see page 9-14 of section 9 (Advanced Design Features) in your User Guide (download from http://www.statease.com/dx6_man.html). When you analyze the results from mixed level response surface methods (RSM) designs, you must edit the model so it excludes squared terms on factors you restrict to only two levels. (Learn more about factorial design by attending the 3-day computer-intensive workshop "Experiment Design Made Easy". Go to http://www.statease.com/clas_edme.html for a description and links to the course outline and schedule.) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 2 - FAQ: Mission Impossible: Predictions that Go Out of Bounds -----Original Question----- From: Malaysia "We have been using your Design-Expert (DX) software both for response surface methodology (RSM) and mixture experiments. Our experiments involve a response that is always positive. After fitting data to a central composite design, we generated contour plots and 3D graphs with negative values in experimentally unexplored regions. This cannot happen in reality. Could you suggest: a. How these negative values can be interpreted. b. How we can avoid such unusual values in the response surface." Answer: First of all, be very careful about interpreting predictions in unexplored regions of your factor space, particularly if they fall outside of the "box" defined by the factorial levels (plus or minus one). You'd best interpret any negative values as zero (nil) for a response that can only be positive. To avoid these impossible predictions, consider re-analyzing the data after applying an appropriate response transformation. For example, in chemical processes the molecular yield must be between 0 and 100 percent. If a DOE to optimize such a process is successful, the results will push up toward the upper bound and predicted results may actually exceed the theoretical limit. In this case we recommend use of the "logit" (=log-e [(Y-LL)/(UL-Y)], Y is the response, LL is Lower Limit, and UL is upper limit). Here's what DX online Help says about this function (under Response Transformation): "The logit transformation is useful if the response has a finite range, such as 0% to 100%. Logit spreads out the values near the boundaries. The actual response data collected must be between the lower and upper limit, and not equal to either one. When using this transformation, it is very important to correctly set the lower and upper limits to the natural limits of the response. The software defaults to the response maximum and minimum values, but these are generally NOT correct." (Learn more about RSM, including use of transformations, by attending the 3-day computer-intensive workshop "Response Surface Methods for Process Optimization". See http://www.statease.com/clas_rsm.html for a description and links to the course outline and schedule.) ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 3 - Alert: Free Patch for Stat-Ease Version 6 Software If you own a permanently licensed copy (individual or networked) of Design-Ease(R) or Design-Expert version 6 (DE6 or DX6), go to www.statease.com/soft_ftp.html#dx6updt for downloads that will patch your software with the latest enhancements. If you do not currently use Stat-Ease software, download a fully-functional free trial of DX6 at http://www.statease.com/dx6descr.html which you can use at no cost for 30 days. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 4 - Alert: Talk on Semifold at 2001 Annual Quality Congress Via case studies, this talk reviews the strategy of foldover on low-resolution (III) two-level fractional factorials and demonstrates how to reduce experimental runs by making use of semifoldover methods to augment medium-resolution (IV) designs. Get a free download of the talk in Portable Document Format (PDF) (viewable with Adobe's Acrobat Reader shareware) by clicking http://www.statease.com/pubs/semifold.pdf. Come to the Congress if you can (details at http://aqc.asq.org/). ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 5 - Alert: "Stat-Teaser" Newsletter Featuring the Article "Mad Mark and the Meter Gnomes" Get a free download of the March 2001 Stat-Teaser in PDF by clicking http://www.statease.com/newsltr.html. In addition to the article noted above, this issue discusses new features offered in Design-Expert version 6, and how they've been rated by reviewers (see "Rave Reviews for DX6"). ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ I hope you learned something from this issue. How did you like it? Do you have any suggestions on how to make it better? Address your questions and comments* to me at: Mark@StatEase.com I hope to hear from you! Sincerely, Mark Mark J. Anderson, PE, CQE Principal, Stat-Ease, Inc. (http://www.statease.com) Minneapolis, Minnesota USA *Please do not send me requests to subscribe or unsubscribe - follow the instructions at the end of this message. 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) - Stat-Ease programmers, especially Tryg - Heidi Hansel, Stat-Ease communications specialist "It's frustrating when you know many answers if nobody bothers to ask you the questions." DOE FAQ Alert - Copyright 2001 Stat-Ease, Inc. All rights reserved. Subscribe (join our list) by going to: http://www.statease.com/doealert.