topic Re: Covariance Estimates when Fitting Generalized Linear Mixed Model in Statistical Procedures

Covariance Estimates when Fitting Generalized Linear Mixed Model

binfang — Mon, 20 Aug 2018 18:43:50 GMT

I noticed there are something wrong with covariance estimates when fitting Generalized Linear Mixed Model, as it is shown below:

The model is a multinomial generalized mixed model, where the outcome is a 4-level categorical variable and all the predictors are continuous.

The SAS codes are as attached.

Can you suggest any way to fix the problem?

Thank you!

Re: Covariance Estimates when Fitting Generalized Linear Mixed Model

PaigeMiller — Mon, 20 Aug 2018 19:11:58 GMT

What is wrong with the results?

Re: Covariance Estimates when Fitting Generalized Linear Mixed Model

binfang — Mon, 20 Aug 2018 23:40:46 GMT

One estimate is 0, the other one is extremely small. And their standard error are both missing.

Re: Covariance Estimates when Fitting Generalized Linear Mixed Model

PaigeMiller — Tue, 21 Aug 2018 00:02:03 GMT

What makes a zero wrong? Those could be (and probably are) the correct calculations.

Re: Covariance Estimates when Fitting Generalized Linear Mixed Model

binfang — Tue, 28 Aug 2018 19:07:24 GMT

Hi Paige,

The model I am trying to fit is a categorical mixed model, where the outcome is a 4-level categorical variable. This means there should only be 3 rows of estimate in the covariance output, but there are 4 rows instead, which is not what I expected since one level should be used as reference.

I have attached the modified syntax if you are interested to run. The data is attached in the next post.

Best,

Bin

Re: Covariance Estimates when Fitting Generalized Linear Mixed Model

PaigeMiller — Tue, 28 Aug 2018 19:22:06 GMT

@binfang wrote:

This means there should only be 3 rows of estimate in the covariance output, but there are 4 rows instead, which is not what I expected since one level should be used as reference.

I don't think this is correct, I think there are covariance estimates for every level of your 4-level categorical variable. But honestly, it has been a very long time since I did PROC GLIMMIX with DIST=MULTINOMIAL. Maybe someone knows.