What I have is a generalized linear mixed model of the log OR of a rater (random effect) giving a response above a certain level on an ordinal scale, given a specification of what the rater was presented with (acoustic parameters of the stimulus). Now, the residuals look fine, but I cat wrap my head around the residuals v.s. fitted values plot. Presumably, heteroscedasticity should to should not be a huge problem for in a logit model, and the results are more or less equidistant across the range of predicted values, but why the odd shape?

Residuals vs. fitted values plot for logit GLMM

Is this what you usually see (this is my first logit GLMM)? Is this due to the link function,or what could it be?

You can download the model here (R save format) https://www.dropbox.com/s/rkygt5a8eat8qpc/gv.glmer02.rda?dl=0

if you want to play with it.


This is what you usually see in a logit model.

In general, the strange shape is because the outcome is an integer. As a result, you see curves corresponding to the continuous input (smooth) minus the discrete output (not smooth), with one curve for each possible output (this top curve for the positive examples, bottom for the negative ones).

Interpreting these plots is difficult and I'm pretty sure has many posts that I can't search because I'm on my phone, sorry.

I'm not sure why the residuals have the scale they do (-2, 2) but I assume that's some party of your model that I can't see.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.