Fit a Multinomial Mixed Model by Proc GLIMMIX

[binfang]

I am trying to fit a multinomial mixed model. The outcome is a categorical variables with 4 levels and the predictors are continuous. The random intercept effect come from one variable "case" in which subjects are pairs or unmatched.

The code, log, output and data set are all attached in this post. 

A very serious problem in the output comes from the dimension of estimated G matrix: in the output 4 parameters are estimated in the G matrix, whereas there are supposed to be only 3. As comparison, when a logistic mixed model is fitted, in which the outcome is 2-level categorical, only one parameter in the G matrix is estimated.

I guess there is something wrong with the model fitting.

[binfang]

           

Above is what the G matrix dimension and G matrix estimate is. But there is something wrong as I mentioned in the first post.