I have a set of data with an ordinal response ranging from 1-5 (worst to best) and a categorical predictor with five unordered levels. The experiment is a language experiment whereby subjects are asked to rate different sentence types. In the literature it seems that people fit lmer() most of the time by scaling the ordinal response. On the other hand, it seems a reasonable approach to use ordered linear mixed effects probit models or ordered linear mixed effects logit models as implemented in e.g. MCMCglmm() or clmm().

(1) I find myself asking what the best method would be. When is one method to be prefered over the other in an experiment as I have sketched above?

(2) Why do people prefer fitting an lmer() model with a scaled response? What is the advantage?

(3) If it is advisable to use an ordered probit or logit model, how do I decide between an ordered probit or ordered logit model?

Thanks for any help!

I have a set of data with an ordinal response ranging from 1-5 (worst to best) and a categorical predictor with five unordered levels. The experiment is a language experiment whereby subjects are asked to rate different sentence types. In the literature it seems that people fit lmer() most of the time by scaling the ordinal response per subject (i.e. taking the mean response and standard deviation of that subject and dividing the every single response by it.). On the other hand, it seems a reasonable approach to use ordered linear mixed effects probit models or ordered linear mixed effects logit models as implemented in e.g. MCMCglmm() or clmm().

(1) I find myself asking what the best method would be. When is one method to be prefered over the other in an experiment as I have sketched above?

(2) Why do people prefer fitting an lmer() model with a scaled response? What is the advantage?

(3) If it is advisable to use an ordered probit or logit model, how do I decide between an ordered probit or ordered logit model?

Thanks for any help!