topic GLM with Poisson in SAS GENMOD vs R (very different chisq), why? in Statistical Procedures

GLM with Poisson in SAS GENMOD vs R (very different chisq), why?

td1345 — Wed, 20 Nov 2019 15:58:52 GMT

Hey,

I'm using sas and r to perform a glm with poisson distribution. I get the exact same estimates of the coeffs but very different degress och freedom and chisq.

ods output ParameterEstimates=pe;
proc genmod data=input;
logoffset=log(offset);
model count =  /dist = poisson 
				  link=log
				  offset=logoffset;
run;

R:
glm(count ~ offset(logoffset), family = poisson)

Cant find anything about yates correction in sas proc genmod. Ideas?

Re: GLM with Poisson in SAS GENMOD vs R (very different chisq), why?

StatDave — Wed, 20 Nov 2019 15:13:04 GMT

Your GENMOD code uses the LOGOFFSET variable twice in the model, once as a predictor and then again as an offset (which is also a predictor with parameter estimate restricted to 1). If your purpose is to fit an intercept-only model with an offset, then remove LOGOFFSET prior to the slash (/) to remove it as a redundant predictor from the model.

Re: GLM with Poisson in SAS GENMOD vs R (very different chisq), why?

td1345 — Wed, 20 Nov 2019 15:58:28 GMT

Oh, that is a typo in my copying, sorry for that.

Re: GLM with Poisson in SAS GENMOD vs R (very different chisq), why?

StatDave — Wed, 20 Nov 2019 20:59:31 GMT

If you are referring to the continuity correction attributed to Yates, see the several options for continuity correction discussed in the description of the TABLES statement in the PROC FREQ documentation. From your code, it seems likely that you are trying to test a binomial proportion, so the BINOMIAL(CORRECT) option might be what you are looking for.