How to interpret the effect of covariates on the extreme categories in ordinal data models