I have a question which is half theoretical and half practical.
I am working on longitudinal datasets with small sample sizes, which i analyze with mixed models. Now, i am aware of the possible corrections in such cases, Satterthwaite and Kenward-Rodger. I am also aware of the different covariance structures that you can impose in the level-1 variance.
And here comes my question: Can we combine these 2 ??
That is, when we are talking about longitudinal data it is very possible that measurements closer in time are more correlated than those who are further apart, and the AR(1) is a potential covariance structure. Hence, can i have a model where the residuals are autocorellated for instance and at the same time i apply the Satterthwaite correction for the df's and consequently for the inferences of the fixed effects ?
If the answer to that question is yes (as I assume it is), how can I do that in R ? I know that for the AR(1) structure i have to use the nlme package, and for the Satterthwaite correction the lmerTest package which actually uses the lmer package. Do i miss a package or an update to the previous that make the combination possible?
Do you have any hint for me ?
P.S. I found the following paper where they combines these 2 'techniques' in SAS...
https://www.jstor.org/stable/1400374?seq=1#page_scan_tab_contents
Thank you, John
