I was fitting a logistic generalized liner mixed model using glmer from the lme4 package when I stumbled upon the fact that the results change drastically when the order of the data was changed. So I tried running the same analysis with glmmTMB. glmmTMB actually produced more robust results. It also generated fewer warning messages. While glmer produced a scaling error for non-scaled predictors, glmmTMB did not. In addition, glmer produced a singular fit for some models, while glmmTMB did not.
That order matters for glmer is not a new (see: Different results of glmer in R when the order of data is shuffled, glmer in R: Significance estimates are not robust to order of data frame, github: inconsistent SE estimates by order of data #262 ), but still unsolved issue. However I could not find any information about the differences of glmer and glmmTMB.
I'm not a statistician, so I don't have the know-how to dig deeper into why the results of these two functions differ and which one I should prefer. Perhaps someone can explain to me a bit about the differences between glmer and glmmTMB and suggest how to use them.
I'm also happy to provide the output of my analyses if requested (unfortunately, I can't share the data frame for privacy reasons).
