Linear Mixed Models - Satterthwaite Approximate Degrees of Freedom in SAS/R

[henriettebjrn]

Under certain circumstances, the same simulated datasets give differing conclusions in hypothesis tests between lmer in R and proc mixed in SAS due to differences in model fits and Satterthwaite degrees of freedom method. Would documenting this be of value?

[DrLynTaylor]

Yes we definitely would love this to be written up and put onto the repository. Which category do you see this best fitting into, or would you like an additional method category to be added? I assume it could go under MMRM? We currently have a lot of work to do on our MMRM section (the R and SAS examples, are a little disjointed to the R vs SAS page). Is this something you'd be interested in helping with updating and then adding in your Satterhwaite DF issue to it?

[henriettebjrn]

At this stage, we prefer not to contribute to updating the MMRM section, but rather to add a section on the Satterthwaite story. We suggest to add a line in the section “Repeated Measures”, perhaps just under the “Linear Mixed Model (MMRM)”. It could be called “Linear Mixed Model (degrees of freedom)” and we would then make a document called “R versus SAS”.

Maybe at a later stage, we could contribute further to the page.

[DrLynTaylor]

That's fine with me. I've just updated the table to add a row for "Linear Mixed Model (degrees of freedom)". Once the repo refreshes you should see the new row (in around 15 mins). Please store your comparison file under the comp folder and I suggest calling it: r-sas_mmrm_df.qmd. Many Thanks. Let me know if any questions