strengejacke / sjstats

Effect size measures and significance tests
https://strengejacke.github.io/sjstats
189 stars 21 forks source link

CV for mixed models is valid? #109

Closed SchmidtPaul closed 5 months ago

SchmidtPaul commented 3 years ago

In the R² vigniette of {performance} it is written that "As models become more complex, the computation of an R2 becomes increasingly less straightforward."

Shouldn't this whole topic also be true for the coefficient of variation? Don't get me wrong - I was happy to find a function that would get me a CV for merMod/lme class objects. At the same time I was skeptical whether it is valid to simply get the RMSE from any mixed model accounting for heteroscedasticity and for covariance among observations present in residual error.

Please correct me if I'm wrong - I'd simply like your thoughts on this.

strengejacke commented 5 months ago

Yes, it's probably not accurate for mixed models (depending on whether we get accurate residuals). Though there are other packages that also compute the RMSE for mixed models (merTools e.g.), and the CV is just the scaled RMSE

SchmidtPaul commented 5 months ago

Alright, thanks!