How to switch from REML to ML for Model Comparison

[#1]

Hello!

I am trying to compare two mixed effect models with AIC (or AICc) values. When generating AIC values for model comparison of mixed effect models, I've seen (at least in R) that one must switch from Restricted Maximum Likelihood Estimation (REML) to Maximum Likelihood Estimation (ML). How do I do this in JMP?  I am using JMP 15 Pro provided through my work.

Thank you!

Info for this procedure in R:

https://r.qcbs.ca/workshop07/book-en/step-2.-code-potential-models-and-model-selection.html 

[#2]

Hello! 

I have also attached below some dummy data and screenshots of model formulations following the structure of the two models I am trying to compare below. I'm looking for a place to switch from Restricted Maximum Likelihood Estimation (REML) to Maximum Likelihood Estimation (ML) for AIC comparison

Thank you!

[#3 · Accepted solution]

Hi @DensityOfOkapi2 . I don’t think ML is an option…it certainly isn’t in Standard JMP.

[#4]

Thank you! Yeah I haven't had any luck in my search on JMP 15 Pro for ML

[#5]

Thank you @Mark_Bailey ! My rationale for wanting to switch from REML to ML is based in model selection protocol for R. Where one cannot compare the AIC values of two mixed effect models that differ in their fixed effect structures when using REML. The math underlying this goes over my head, but presumably the same comparison applies to comparisons of mixed effect models using REML in JMP? Or perhaps JMP automatically refits the model using ML to generate the AICc values to facilitate model comparisons. If such refitting is going on "under the hood" that'd be great to know!

Faraway (2006) Extending the linear model with R (p. 156):

"The reason is that REML estimates the random effects by considering linear combinations of the data that remove the fixed effects. If these fixed effects are changed, the likelihoods of the two models will not be directly comparable"

[#6]

I do not see why you need to use ML instead of REML to estimate the model parameters and compute AICc. I used another example from the JMP Sample Data Folder, Wheat.jmp, and set up a similar model:

Click the red triangle at the top and select Regression Reports > AICc:

This criterion is added to the Summary of Fit outline: