Related Q: Interpreting plot of residuals vs. fitted values from Poisson regression

I have had "stripey" residuals from a Generalized Mixed Model (with gamma distribution) but am still confused from the linked discussion whether they are acceptable or not. I know why I am seeing them-- my outcome variable has only 10 possible values-- and the distribution of values is non-normal, but am not sure what can & should be done to better model this data.

enter image description here

My model:

glmer(y ~ x + sex + age + source + x:sex + x:source + x:sex:source + (1 | ID), 
      data = d,
      family = Gamma(link=identity),
               control = glmerControl(optimizer = "bobyqa" , 
                                      optCtrl = list(maxfun= 10000)),
      nAGQ = 1) 

...where y is score on a questionnaire (min-max: 1-11), ID are participants, source is a binary within-person repeated measure and x is a continuous variable. I chose to use a Gamma distribution due to positive skew in the outcome & an identity link function in order to facilitate interpretation of estimates in terms of the original outcome units.

Must I declare the outcome as ordinal? And if so, does that mean I must switch packages e.g. MCMCglmm (ordinal GLMM)?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.