You have several of options. Many JMP DOE options do not define your response data type. That data type is defined by what you enter after you run your experimental conditions. I would recommend you try Custom Design as a first option because it very flexible for both input variable types and number of runs for your design.
Designs more specific to categorical/ordinal responses include Choice design and Covering Arrays. Covering Arrays are JMP 12 Pro only.
Hope this helps.
To pile on a bit to what my colleague Bill Worley shared above...it's the analysis of your experimental results that is more influenced by whether or not a response is continuous, categorical, or ordinal. The DOE planning process is largely agnostic with respect to the type of response data (continuous, categorical, ordinal). JMP and more so JMP Pro (I'm thinking some of the modeling personalities in the Generalized Regression platform) support a wide range of modeling and exploratory data analysis methods for categorical and ordinal response data.
Thanks for your suggestions. Since I don't have a background in statistics, I really appreciate your help in this topic.
As I understand, the 'normal' optimal design assumes that the error is normally distributed and standard or generalized linear regression is used. When the response is nominal or ordinal, the assumption is not true anymore and we will need logistic regression or cumulative logistic model.
On the practice side, when we design the experiment, is there a way to specify the response type? I could change the type in the column properties but that doesn't seem to help the analysis since I still cannot change the personality of the response to 'nominal logistic' or 'ordinal logistic'. Previous experience seems to suggest that I need to repeat the runs multiple times to get a probability of fail/pass in order to deal with say binomial responses but I cannot find anywhere to address that in JMP DOE.
If you can, is it possible for you to provide an example on how to deal with nominal and ordinal responses?
uniwander: It's not the design per se that requires your thinking around "... the 'normal' optimal design assumes that the error is normally distributed and standard or generalized linear regression is used...". It's the ordinary least squares regression modeling technique (which is the modeling technique the naïve gravitate towards) that is sensitive to these assumptions. The design is just a means to create a coherent set of treatment combinations to support (in the Custom Design platform) a user specified model containing main effects, interactions, or other higher order terms as you require.
So right away if someone tells me they have a categorical or ordinal response...I'm NEVER using OLS to analyze the results...but gravitate to some other technique. Two you suggest are nominal logistic or ordinal logistic regression. I do recommend you change the modeling type and data type in the column properties window as you suggest. This way, even if you aren't modeling...the fact that the variable is categorical or ordinal will always travel along with that variable no matter WHICH JMP analysis platform you are working with. In the Fit Model platform, once you cast responses into the Y window...JMP eavesdrops on the data and modeling type for the response and will toggle to what it thinks is the most appropriate specific modeling personality.
You do not need to necessarily have replication built into your design to make the magic of nominal or ordinal logistic regression work...there are lots of reasons to replicate a design...but it's not required from a modeling perspective.
Also I always start my empirical analysis, BEFORE modeling, by plotting responses in both a univariate and bivariate view of the world using the Distribution and Fit Y by X platforms. Graph Builder can also do some of this heavy lifting too. I shout from the rooftops whenever I can, before beginning modeling, "plot the data, Plot The Data, PLOT THE DATA" to help you find outliers, suspicious observations, or observations that don't make sense...and does the results 'fit' with prior process knowledge.
A good place to start with looking at examples is the JMP documentation itself...here is the landing page that can serve as an entry portal for you in the Online JMP Documentation:
I like the suggestion about plotting the data. It is a great place to start getting familiar with the data, and to screen data for further data cleaning.Thank you for sharing!