| Vol. 26, No. 2 - March/April 2026 |
|
|
|
|
|
| | IN THIS ISSUE: R-squared as a fit statistic, webinar for German speakers, and more |
|
| | FAQ Is it OK to ignore some of my data to increase R-squared?
Original question from an Academic Researcher:
“I have noticed in the literature that R-squared values typically exceed 0.85. I can only achieve such values on my two-level-factorial milling experiment if I ignore between 4 and 6 data points for each of my three responses. However, I am uncertain if removing so many points is excessive. In other words, I’m unsure whether the models developed by excluding these points are reliable.”
Answer: Pay little attention to the R-squared values—only worry if the predicted R-squared goes below zero. After resetting ignored data to normal (these not being statistical outliers), I refit your data. All three surface-roughness models are significant at p<0.05 and generate positive predicted R-squareds—the first two being above 0.5, which is not bad for industrial experimentation. The third surface-roughness measurement is not modeled nearly as well, as evidenced by a prediction R-squared of 0.18. However, another measure for goodness of fit that I like as a ‘bottom line’—adequate precision—exceeds (barely!) our guideline of 4. Be a bit more wary of this model than the other two, but do not immediately reject it as explained by me in Don’t Let R-Squared Rule You.
I will note that this Stat-Ease software user’s concern about R-squared is widely shared. I hear this very frequently. I credit DOE expert Doug Montgomery for allaying my worries about this fit statistic being too low. He assured us industrial researchers that a low R-squared does not mean the model is useless or that no meaningful relationships exist, particularly in experiments where the focus is on the significance of factors rather than predictive power (as is the case for factorials). However, Professor Montgomery could not stop his students from discussing troublesome R-squareds behind closed doors—he overheard them!
P.S. Never delete results to make R-squared better!!!
(Learn more about fit statistics by enrolling in the next Introduction to Design of Experiments (DOE) short course.) |
|
| | EVENTS
I will present a webinar on “Milestones to Modern DOE for Rapid Manufacturing Improvement" for the International Society of Six Sigma Professionals (ISSSP) on May 20 at 2 pm CDT. You need not be a member of ISSSP to attend. Register at this site. If you do not see my webinar posted, check back in early May.
Do you need a speaker on DOE for a learning session within your company or professional society at regional, national, or international levels? If so, please get back to me. |
|
| | ONLINE LEARNING Sharpen up your DOE skills with a mix of free and paid training: whatever fits your business needs.
Short courses to quickly level up on DOE essentials — online instructor-led learning
All courses are 3.5 hours, registration $295*. Classes will be held in US Central time. *Only $95 for students, faculty or researchers. $50 discount per person for enrolling in more than 1 course or by registering multiple people for any given course. Current software or Annual Support & Maintenance subscribers may also ask about an introductory rate. Contact us to qualify.
Comprehensive DOE courses — online instructor-led learning
**Only $149 for students, faculty, or researchers at an accredited academic institution. Contact us to qualify.
Don’t see the course you want, or the dates don’t work for you? Ask our team about taking a course asynchronously using recorded video sessions.
See this web page for the complete schedule of upcoming Stat-Ease courses. To enroll in the workshop that suits you best, click Register on that webpage, or click here to contact us.
If you lead a group of six or more colleagues, save money and customize content via a private workshop. For a quote, please contact us.
Free webinars — sign up to take advantage
On-Demand Videos By the way, our Statistics Made Easy By Stat-Ease YouTube channel provides a free library of highly educational recorded webinars covering a wide variety of DOE tools. It offers videos at all levels—from those new to DOE on up. Take advantage! |
|
| | INFO Check out my newly updated DOE It Yourself—a long list of fun and educational science projects for children and/or adults, all of which demonstrate the value of design of experiments. If you try any of my suggestions or come up with a new one, get back to me with the outcome and tips for others who enjoy DIY DOEs to do at home or in class.
Stat-Ease recently published a technical paper on Mixture Design for Optimal Food and Beverage Formulation—drawn from part of my one-hour webinar, Deploying DOE to Accelerate R&D on Food and Beverages. It is now available in the “Case Studies” section of the Stat-Ease website, where you will also find a substantial collection of papers to help make the case for DOE in your organization.
The Stat-Ease marketing team has launched a monthly LinkedIn newsletter collecting educational material on a focused DOE topic each time. If you’re active on that platform, consider subscribing at the above link—it is a convenient way to stay current on the tools and techniques that make the most from every experiment.
Also, see our latest publication roundup, featuring application of Stat-Ease software for exceptionally successful experiments. |
|
| | BLOGS StatsMadeEasy My wry look at all things statistical and/or scientific with an engineering perspective.
Stat-Ease Blog Great tips from the Stat-Ease team for making DOE easy; for example, these recent posts: |
|
| | Feel free to get back to me via [email protected] with further questions or comments: I would really appreciate hearing from you!
All the best, |
|
|
| | | | | “A consistently ‘lucky’ person must, almost certainly, be doing a number of things differently and doing them right.”
–“How to Get Lucky,” George Box, Quality Engineering, 1994, Vol. 5, No. 3, pp.517-524
|
|
|
| | Stat-Ease, Design-Expert and Statistics Made Easy are registered trademarks of Stat-Ease, Inc. |
|
| | | |
|
|
|
| |
|
