can fitrm handle unbalanced data (without decimating it)? if not, what alternative will still allow me to see (or even set) the response covariances?

it looks like fitrm (and ranova?) takes a multivariate (MANOVA?) approach to handling repeated measures much like i've heard other programs do (e.g. SPS's repeated measures GLM procedure), but our biostats guy has told me that MANOVA won't work for unbalanced data (maybe that's only for data as unbalanced as mine! i don't know...).

i wanted to get some verification here that fitrm can (or should) not handle unbalanced data. if it can't, then what else could i use that would still allow me to see (and ideally to set or constrain) the response covariances (NOT covariance pattern, but the covariances between, in my case, intracluster and intercluster members)?

thanks for any input,

ben

Answer 1

[Aditya]

Hi Ben,

For highly unbalanced data, I recommend exploring linear mixed-effects models using fitlme in MATLAB. They provide a flexible framework for modeling repeated measures data with complex covariance structures. If you need more specific control over the covariance, consider Bayesian methods or software that supports GEE.

filtime: https://www.mathworks.com/help/stats/fitlme.html