search

strengejacke / sjstats

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

R2 MLM #65

Closed DominiqueMakowski closed 5 years ago

DominiqueMakowski commented 5 years ago

Dear Daniel, thanks a lot for your great suite of packages and their documentation, which helped me a lot in my own learning of stats and programming 😉

I recently stumbled upon this blogpost using a newly (again!) developed R2 measure for mixed models (Rights & Sterba, 2018; Rights & Cole, 2018). However, the function is a bit clunky to say the least, and relatively difficult to use due to the need of manual extraction of relevant indices.

I have the feeling that it would be possible to re-implement it in a more digest way (i.e., that would automatically extract the things it needs from models). Unfortunately, my knowledge and understanding of the internals of mixed models appears as unsufficient to be sure of that. What is your opinion? Do you think it would be worth it trying to re-implement this R2 framework?

Again, thanks a lot 😊

strengejacke commented 5 years ago

Hi Dominique, I'm not quite sure what r2mlm() exactly does. It looks like extracting random effect variances. Is it similar to sjstats::re_var()?

DominiqueMakowski commented 5 years ago

Well, from the table 2 of Rights & Cole 2018 it seems as if the authors propose a more general (and extended) framework than the decomposition of Aguinis, or Nakagawa or such. From what I understood their indices (extracted with the r2mlm function) include and cover all the previously described indices of R², which sole report is deemed by the authors as insufficient.

They seem to suggest that their R2 framework would be a unifying answer to the "R2 for mixed models" controversy...

This blogpost uses it with a brms model.

strengejacke commented 5 years ago

Closing in favour of https://github.com/easystats/performance/issues/2