multicollinearity in proc glimmix model

tunayhy — Fri, 20 May 2016 02:41:14 GMT

Hi,

Now i try to analyze a data using proc glimmix

dependent variable is binary ( 0, 1)

level 1 independent variable is categorial

level 2 independent variable is continous

befor making a multilevel model,

i wanna check multicollinearity between independent variables

especially, i doubt level2 variables would cause the multicollineariy problem because accroding to corrleation anyalsis i confirmed there is higly correlated among some level2 variables ( r>.8)

in case of level1 i've already done mulicollinearity diagnosis using below syntax

 proc reg data=fitness;
      model Oxygen=RunTime Age Weight RunPulse MaxPulse RestPulse 
            / tol vif collin;
   run;

but i don't know how to check the multicollinearity between level 2 variables

could i just put the level 2 variables in below regression model statement with other level1 variables?

or is there other syntax for logistic multilevel anaylsis?

if you have an answer, please let me know.

thanks !

Re: multicollinearity in proc glimmix model

Rick_SAS — Mon, 23 May 2016 15:12:02 GMT

I'm not an expert, but since this has gone unanswered for a few days, I'll make a suggestion.

In practice, I would try running PROC LOGISTIC and see what happens. If your data are degenerate, PROC LOGISTIC will probably issue some WARNING (maybe about "non-convergence due to quasi-separation"). If you don't get a WARNING, then your data are probably okay.

If you are paranoid, then see the Knowledge Base article on "Testing assumptions in logit models", which is for continuous variables. There is an example of using GENMOD to look at the information matrix (Hessian) and then passing information over to PROC REG to do the collinearity analysis. To handle the categorical variable, create dummy variables for the design matrix, and then follow the previous procedure for the dummy variables.

topic Re: multicollinearity in proc glimmix model in SAS Procedures

multicollinearity in proc glimmix model

Re: multicollinearity in proc glimmix model