1: Upgrade Alert: V8 of Design-Ease software released! Version 8 of Design-Ease software ("DE8") is now available for trial use, outright purchase or upgrade. For those who only need factorial design and analysis for process screening and characterization studies, Design-Ease is a low-cost alternative to our fully-featured Design-Expert ("DX") program. V8 provides many enhancements, the biggest of which may be the extension of half-normal plots for all factorial designs.  This simple and robust method for selecting important effects was formerly available only for two-level designs.

See DE8 detailed at this feature-list for Design-Ease. From there you can download a fully-functional free trial or make a purchase.

PS to Design-Expert users: If you haven't upgraded, see this feature-list for Design-Expert to learn what you're missing—many useful features for design and analysis, as well as nicer-looking and more functional graphics.  Why short yourself?

2: Software Alert: Version 8.0.4 of Design-Expert software released with new feature for confirmation runs (FREE UPDATE). The newly-released version 8.0.4 of Design-Expert software is posted at at this ftp site for free trial evaluation. This web site also provides free patches to update older licensed versions of 8.0. The release provides a valuable new feature—a confirmation node that now appears under the optimization branch of Design-Expert. Having searched out a desirable process setup or product formulation, you can enter in the sample size (n) of your confirmation runs and see the prediction interval for all measured responses—very handy!

View the ReadMe file for other features, installation tips, known 'bugs,' change history, and FAQs.

PS. Heads-up: If you want to receive notice when an update becomes available, go to Edit on the main menu, select Preferences and, within the default General tab, turn on the "Check for updates on program start" option.

3: FAQ: Two-factor interaction significant but the two main effects are not: Is this plausible? Original Question: From a compounder of pharmaceuticals: “I found an interaction term (AC) significant (p value=0.02271) but both main effects A & C were statistically insignificant (p values over 0.05). Could you explain how this could happen? Is it plausible?”

Answer: From Stat-Ease Consultant Shari Kraber: “It is a common misconception that when you have a significant interaction, the main effects must also be significant. This is not true at all. Look at the interaction graph for AC—it looks like an "X"—with both the top values in line horizontally and the bottom values also in line horizontally.

The main effects are the average effects at the left side of the graph versus the average at the right side of the graph. When both these averages are nearly identical, the difference will be zero and thus the main effect for that factor will be not statistically significant. However, the A and C terms are critical to the model because they are the parent terms to the interaction. The interaction coefficient is basically a correction to the individual parent term coefficient when the second parent is set at its different levels.

So, this is not a problem, and there is nothing you can or should do to avoid this situation. You should include the parent terms, A and C, in the model to maintain statistical hierarchy.”

Comment: My favorite example of this (an interaction being significant but not its parents) is a real-life case of a two-level factorial done on two suppliers providing two chemicals for a wafer-cleaning process—the goal being a speckless surface.

The factors were:
   A. Who supplies chemical X: P or Q
   B. Who supplies chemical Y: P or Q

Whenever two chemicals came from competing vendors the materials interacted antagonistically and the cleaning process failed. In other words, they could not have P supply X and Q supply Y (or vice-versa). The process succeeded only if both chemicals came from one vendor or the other. Thus, neither of the two main effects (supplier of X or supplier of Y) made any difference, but the two-factor interaction proved to be very significant!

—Mark

(Learn more about interactions by attending the two-day computer-intensive workshop "Experiment Design Made Easy (EDME)." See this link for EDME for a description of this class and go from this page to the course outline and schedule. Then, if you like, enroll online.)

4: Expert FAQ: Too many center points! Original Question: From a DOE Consultant/Trainer: “I have a question relating to center points, and I will pose it using a hypothetical example. Suppose that I wish to enter center points into a two-level factorial design (to test for curvature), and I have decided that 4 center points will be sufficient. It so happens that my design includes two categorical factors. I initially enter 1 center point when setting up the design. Design-Expert warns me that 1 center point won't be sufficient to test curvature, and I know this to be true. I decide to continue anyway (risking the ire of the software!!!). I am then warned that center points will be duplicated at each level of each categoric factor, and I will now have 4 center points.

My question: Is it OK to enter the 4 center points in this manner? Obviously, if I enter 4 center points initially, I will eventually end up with 16 center points, which appears excessive. On the other hand, do I really need the 4 center points at each level of the categoric factors in order to adequately test for curvature?”

Answer: From Stat-Ease Consultant Brooks Henderson: “The short answer is: Yes you can do it, but no I wouldn't recommend it. I would recommend that you choose at least 3 center points for each categoric combination. This will give you a better estimate of what's going on in the center of the design space. If you use only 1, then you are at risk of the curvature test being completely wrong, if that data point turns out to be off from what the true response at the center should be. Using 3 center points to get an average response in the center of your design space is a much better bet (even 2 would be better).

I understand you will get a large number of runs if you specify 3 center points for each of the categoric combinations, but there is another alternative. If you suspect there might be curvature, instead of testing for curvature, consider running an RSM optimal design for a quadratic model. For example, if you have a 2^4 factorial with 2 categoric factors and you only specify 1 center point (which I wouldn't recommend), you will get a design with 20 runs. If you instead choose an RSM optimal design for four factors and a quadratic model with 3 lack-of-fit and 3 replicates (which are the minimum you should choose, 5 is better) you have a 19-run design. This RSM design can actually model the quadratic terms and check for lack-of-fit in your model. So, you get more information for one less run.”

Stat-Ease Consultant Pat Whitcomb adds: “The reason Design-Expert replicates the center points 4 times is there are 4 continuous surfaces, one at each combination of the categoric factors. Curvature can differ depending on the categoric combination.

Therefore 4 estimates of curvature are required, 1 for each surface.

With less than 4 center points on a given surface its estimate curvature has little power. These estimates are pooled so, if you can assume curvature is the same on all surface (i.e., independent of the categoric factor combination), you can reduce the number of center points. (This is a big assumption!)”

(Learn more about RSM by attending the two-day computer-intensive workshop "Response Surface Methods for Process Optimization (RSM)." See this link for RSM for a description of this class and go from this page to the course outline and schedule. Then, if you like, enroll online.)

5: Info Alert: Pharma QbD site features RSM with Design-Expert, Design Product News picks up Whirley-Pop case study on DOE. See how response surface methods (RSM) led pharmaceutical researchers to new, robust process conditions which provided an 11.6% improvement in yield in this post by Pharma QbD. Design Product News re-published the Stat-Teaser article by Brooks Henderson detailing his DOE on Whirley-Pop popcorn—it is posted at this DPN blog. Check out comments and weigh in with your ideas on making this snack even tastier.

6: Events Alert: Talks by DOE experts and software showings. (Last Notice) Stat-Ease Consultant Pat Whitcomb will present "Practical Aspects of Algorithmic Design of Physical Experiments From an Engineer’s Perspective" for the 54th Annual Fall Technical Conference in Birmingham, AL, on October 7-8. Details for this meeting of quality experts and statisticians are posted at this Fall Technical Conference official website.

(2nd Notice) Pat will detail "Optimal Mixture Design of Experiments for QbD," for the American Institute of Chemical Engineers (AIChE) Meeting in Salt Lake City, UT, November 7-12. For more information, see this home page for the 2010 AIChE Annual Meeting.

(Last Notice) Also, we invite attendees of the Minneapolis MD&M (Medical Device & Manufacturing) on October 13-14 to visit us at booth #729. If you have not seen Design-Expert version 8 yet, ask for a demonstration.

Click on www.statease.com/events.html for a list of upcoming appearances by Stat-Ease professionals. We hope to see you sometime in the near future!

PS.  Do you need a speaker on DOE for a learning session within your company or technical society at regional, national, or even international levels? If so, contact me. It may not cost you anything if Stat-Ease has a consultant close by, or if a web conference will be suitable. However, for presentations involving travel, we appreciate reimbursement for travel expenses. In any case, it never hurts to ask Stat-Ease for a speaker on this topic.

7: Workshop Alert: See when and where to learn about DOE Seats are filling fast for the following DOE classes. If possible, enroll at least 4 weeks prior to the date so your place can be assured. However, do not hesitate to ask whether seats remain on classes that are fast approaching! Also, take advantage of up to a $395 discount when you take two complementary workshops that are offered on consecutive days.

All classes listed below will be held at the Stat-Ease training center in Minneapolis unless otherwise noted.

• Basic Statistics for DOE (SDOE)
  • November 2*
• Experiment Design Made Easy (EDME)
  • November 3-4*
  • December 7-8**
• Response Surface Methods for Process Optimization (RSM)
  • December 9-10**
• Mixture Design for Optimal Formulations (MIX)
  • October 26-27***
• Advanced Formulations: Combining Mixture & Process Variables (MIX2)
  • October 28-29***

* Attend both SDOE and EDME to save $295 in overall cost!
** Take both EDME and RSM in December to cut $395 off the combined tuition!
*** Take both MIX and MIX2 to save $395 in overall cost!

See this link to listing of upcoming classes for complete schedule and site information on all Stat-Ease workshops open to the public. To enroll, click the "register online" link on our website or call Elicia at 612-746-2038. 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.****

**** Once you achieve a critical mass of about 6 students, it becomes very economical to sponsor a private workshop, which is most convenient and effective for your staff. For a quote, e-mail [email protected].

8: Heads-up on DOE FAQ Alert format: HTML version in the works I am putting the final touches to an HTML version of DOE FAQ Alert, which for over a decade now (first issue April 2000) has gone out in text only. My strategy, common in computer circles, is to do a release in parallel with this October issue going out in text and also being posted to the website in HTML format. The November issue will be emailed in HTML. I suppose some folks like me who resist change may struggle a bit with the new look. If so, let me know. However, I'm not likely to look back—it's full steam ahead!