Determining if Random Effects are Significant in Proc Glimmix

ksn999 — Mon, 04 Feb 2019 15:20:22 GMT

Hi all,

I am used to working in proc mixed and using the option covtest in the first line to get Z and p-values of covariance parameter estimates (because I have a random effect that I only leave in if it is significant). For one of my variables, I had to move to proc glimmix because I needed to do a median split on the data and assess it using a binary distribution model. However, I can't seem to figure out how to get the same Z and p-value outputs for the covariance parameter estimates table. Is there a way to do that? I've read the literature on covtest in glimmix, but none of the options give output on individual random effects (only all random effects as a whole). This is the sas code I am using currently:

proc glimmix data=BL_SRms ;
class condition time Subject_ID;
model il6Median_Split(event='1')=condition time condition*time / dist=binary;
random Subject_ID Condition(Subject_ID);
random condition / subject=Subject_ID*time residual;
covtest diagg;
run;

What I want is something like this:

Right now, proc Glimmix is only giving me a table with the estimate and standard error. I want the z value and p-test too.

Other information that may or may not be relevant: I am using SAS University Edition.

Re: Determining if Random Effects are Significant in Proc Glimmix

StatsMan — Thu, 07 Feb 2019 14:35:38 GMT

The WALD option on the COVTEST statement might do what you want. Try

covtest / wald;

That will match the results you would see from MIXED (using COVTEST option) if your model had normal errors.

topic Determining if Random Effects are Significant in Proc Glimmix in Statistical Procedures

Determining if Random Effects are Significant in Proc Glimmix

Re: Determining if Random Effects are Significant in Proc Glimmix