topic Re: model selection based on AICC in proc logistic in SAS Programming

model selection based on AICC in proc logistic

lillymaginta — Wed, 27 May 2020 10:31:07 GMT

I am trying to specify model selection based on AICC but I get error in the following statement:

proc logistic descending data=aim2bx;
class ut (ref= '1')  drug / param = ref;
model ut= drug age/ link = glogit SELECTION = FORWARD (select= AICC);
run;

Re: model selection based on AICC in proc logistic

ed_sas_member — Wed, 27 May 2020 11:42:23 GMT

Hi @lillymaginta

Maybe you could try to switch to PROC HPLOGISTIC, which is high-performance statistical procedure that fits logistic regression models for binary, binomial, and multinomial data on the SAS appliance.

The SELECTION statement allow the use of AICC as a selection criterion:

proc hplogistic data=aim2bx ;
	class ut (ref= '1')  drug / param = ref;
	model ut= drug age/ link = glogit;
	selection method= forward (select= AICC);
run;

Re: model selection based on AICC in proc logistic

lillymaginta — Wed, 27 May 2020 11:51:49 GMT

Thank you! The statement output parameter estimates only, would it be possible to output the odds ratio?

Re: model selection based on AICC in proc logistic

ed_sas_member — Wed, 27 May 2020 12:12:57 GMT

Hi @lillymaginta

The LOGISTIC procedure offers a wide variety of postfitting analyses, such as contrasts, estimates, tests of model effects, least squares means, and odds ratios. PROC HPLOGISTIC is limited in postfitting functionality, since with large data sets the focus is primarily on model fitting and scoring.

I would suggest that you use it for model selection and then run a PROC LOGISTIC on the select model to get OR.

best,

Re: model selection based on AICC in proc logistic

lillymaginta — Wed, 27 May 2020 12:21:11 GMT

Thank you this is very helpful!