topic Re: Proc MIANALYZE with Poisson regression with robust error variance in Statistical Procedures

Proc MIANALYZE with Poisson regression with robust error variance

llu — Fri, 17 Nov 2017 01:02:50 GMT

Hi,

I encountered a trouble when I wanted to use PROC MIANALYZE with poisson regression with robust error variance, which was performed by PROC GENMOD + repeated statement (see link ESTIMATE RELATIVE RISK IN SAS USING PROC GENMOD)

According to the example on SAS website (Reading Generalized Linear Model Results), I need three dataset to make MIANALYZE work. They are datasets for parameter estimates, parameter information and covariance matrix. Because my procedure is for poisson regression with repeated statement, I output three tables "GEEEmpPEst", "ParmInfo" and "CovB" for the three datasets needed.

The issue is, the approach works when I have only one predictor but does not work if I had multiple predictors. The error info says "CovB" was not created, which basically means it does not generate the CovB dataset. I tried to get table "GEERCov" and "GEENCov" and none of them exist.

My question is, how can I get a covariance matrix from a model with more than one predictor using poisson regression with robust error variance, to apply to the MIANALYZE procedure?

Thank you!

Re: Proc MIANALYZE with Poisson regression with robust error variance

SAS_Rob — Mon, 20 Nov 2017 18:34:53 GMT

In general, unless you are using the TEST statement or the MULT option, it is not necessary to include the COVB= and PARMINFO= data sets in Proc MIANALYZE. Thus if you provide just the PARMS(CLASSVAR=LEVEL)= data set then you should be good to go.

proc genmod data=outmi;
by _imputation_;
class a id;
model y=a x/d=bin link=log;
repeated subject=id;
ods output geeemppest=parms;
run;

proc mianalyze parms(classvar=level)=parms;
class a;
modeleffects intercept a x;
run;

Re: Proc MIANALYZE with Poisson regression with robust error variance

llu — Wed, 22 Nov 2017 15:00:53 GMT

Thank you Rob!

Actually, I also found that if I requested the GEE covariance matrix by adding "ecovb" after the repeated statement, then I am able to get the GEE empirical covariance matrix "GEERCov".

Thank you again for your advice!