It’s World Statistics Day! To honor the theme of the day, the JMP User Community is having conversations about the importance of trust in statistics and data. And we want to hear from you! Tell us the steps you take to ensure that your data is trustworthy.
Community Manager Community Manager

Essentials of Designing Experiments Using JMP - Journal

Video from the live Mastering JMP webinar  Phil Kay presented on this topic and the attached journal are available for viewing and to practice what you learned.  See Phil's related blog.


Thanks @phil_kay for your contribution and for sharing that useful tip of using the LogitPct() (Logit Percent) a transformation on the Response = Yield in order to impose the constraint of proportion defective ≥0 and ≤1 when running the most "typical" designs and using Yield as your Response.



Thanks @PatrickGiuliano . I would also point out that for users of JMP Pro you can use the Generalized Regression Personality and select the Beta distribution for your response. In this case you would need to first convert percentages to proportions, e.g. 98% becomes 0.98.

Awesome Thanks @phil_kay. I do have one question here, if I run "Fit Definitive Screening" on your Chemical Experiment Example, and then I select "Make Model", JMP ends up giving me only specific model effects, namely: Catalyst, Temperature, Time, Catalyst*Temp and Time*Time.  At a high level, what is JMP doing in the "Fit DSD" model output, in order to determine which specific model terms are most important to optimally fit this 13 run design (model cannot include all factors so factors were removed but by what decision-making criteria generally)? 


Hi @PatrickGiuliano ,


Good question. Fit Definitive Screening is a specific model screening analysis for DSDs that utilises the unique structure of these designs. You can find more details in this Discovery presentation from @bradleyjones.


Basically it splits the model analysis process into 2 steps:

1. Main Effects and

2. Second Order Effects (2-ways and Quadratics) and the intercept.


In each step the algorithm is looking for those effects that are significantly larger than the error. 


The algorithm then combines the significant effects from each step to give you an overall model. 


There are options as to whether effects in the second step should be added depending on whether they obey strong heredity. That is, second order effects are only added if they have been found to be significant in step 2 AND the contained main effects have been found to be significant in step 1. For example, X1*X2 would only be added if X1 and X2 main effects were found to be significant. 


What this means for you is that you can quickly find the active effects from the response surface model. This method works particularly well when there are only a few (half or fewer) active factors.


I hope that helps.



This Discovery presentation might also be useful.

Article Labels

    There are no labels assigned to this post.

Article Tags