Mixed effects models with ordinal response variable

Hello everybody,

I have a dataset with an ordinal response of 5 levels which wish to 
model on 4 potential predictors. There's a lot of repeated measurements 
(of individual howler monkeys) so a mixed model is required.

I tried pretending the response was a continuous variable but received 
an extremely structured residuals pattern indicating the model was flawed.
I then tried clmm2 (package "ordinal") and MCMCglmm (package "MCMCglmm") 
which should work with ordinal variables. Both seem to be "over 
optimistic" and find significant results where I don't expect them to be 
(essentially in all predictors). The latter's under-conservatism with my 
data is also pronounced by the fact that it found significant or close 
to significant results in a completely random variable I generated and 
included in the model.

Does anyone have an advise or some good experience with mixed models of 
an ordinal response?

Best,
Omer