Binary and Ordered Categorical Responses Require a different kind of DOE?
May 24, 2019 9:06 AM
| Last Modified: May 24, 2019 9:14 AM(2081 views)
In the book “Optimal Design of Experiments: A Case Study Approach”, 1st Edition, by Peter Goos, and Bradley Jones Chap 9, the following text appears (Note that Brad Jones, @bradleyjones, is the Principal Research Fellow in the JMP division of SAS):
[Marc, looking at Brad] Why are you concerned about the response?
[Brad] Well, the type of response you have determines the kind of model you fit to the data. If your response were binary—for instance whether or not you have good adhesion—then a linear model would not be appropriate. Instead you'd need a logistic regression model, and the experimental design you need for such a model is different from the one that is required for the linear models we usually work with. If your response were an ordered categorical one—for instance, whether there is no, poor, good, or perfect adhesion—then you'd need an experimental design for yet another model, the cumulative logit model.”
This is just an aside in the book and there is no further detail.
So my question is, how do you do a DOE in JMP for Binary and Ordered Categorical Responses?