Logistic Regression > Differencce Between Proc LOGISTIC and GENMOD

muzlowe — Tue, 15 Mar 2016 14:06:56 GMT

I've been trying to replicate the results of a logistic regression programmed using Proc LOGISTIC with GENMOD. I believe the options and model specification are the same but some of my parameter estimates are differing around the 4th decimal. The code using Proc LOGISTIC is:

proc logistic data = mydata descending;

class treat (param = ref ref = "A") country / param = glm;

model resp = treat country var1;

run;

Code for Proc GENMOD is:

proc genmod data = mydata descending;

class treat (param = ref ref = "A") country / param = glm;

model resp = treat country var1 / dist = bin link = logit;

run;

Is there any reason why the results would differ using the above code? I'm using SAS v9.4.

Thanks

Re: Logistic Regression > Differencce Between Proc LOGISTIC and GENMOD

Rick_SAS — Tue, 15 Mar 2016 14:28:11 GMT

They use different algorithms to maximize the likelihood, so you might not be able to get an exact match. BY default, PROC LOGISTIC uses a Fisher scoring algorithm, whereas GENOMD uses a ridge-stabilized Newton-Raphson algorithm. You could try setting TECHNIQUE=NEWTON in the MODEL statement for PROC LOGISTIC.

Another difference is that GENMOD estimates a scale parameter. You could try using the NOSCALE option and see if that helps.

Personally, I wouldn't worry much about differences in the fourth decimal place, since that is probably much smaller than the standard errors.

topic Logistic Regression > Differencce Between Proc LOGISTIC and GENMOD in Statistical Procedures

Logistic Regression > Differencce Between Proc LOGISTIC and GENMOD

Re: Logistic Regression > Differencce Between Proc LOGISTIC and GENMOD