08-22-2015 12:15 AM
I'm trying to estimate the Odds ratio of a treatment after adjusting confounding factors in a 1:2 matched data sets (ex file in the attachement)
Due to the dependence between these data, I'm using conditional logistic regression as statistical method.
PROC Logistic & PROC GENMOD all seems able to deal with matched data well but reported differently.
I'm wondering what is the difference between proc logistic & proc genmod for dealing with matched data
and which would be better or more accurate ? .
## ex data set:
Treatment option: Tx; Outcome: Outcome; Confounding factors: B1, C1, B1*Tx;
## The Code I used in PROC Logistic for conditional logistic regression;
PROC LOGISTIC DATA= EX ;
class Tx(REF='0') B1(REF='0') /param=ref;
MODEL OUTCOME(EVENT='0')= Tx B1 C1 B1*Tx / expb ;
## The Code I used in PROC GENMOD for 1:2 matched data ;
PROc GENMOD DATA=ex;
CLASS MATCH_ID Tx(REF='0') B1(REF='0') /param=ref;
MODEL OUTCOME =Tx B1 C1 B1*Tx /dist=BIN link=logit ;
Sincerely thanks for your help ~
09-08-2015 03:14 PM
The two procedures use entirely different estimation methods. Conditional logistic regression in PROC LOGISTIC maximizes a conditional likelihood, while PROC GENMOD uses the Generalized Estimating Equations (GEE) method which is not a likelihood-based method. These methods, and others, are compared in the book "Logistic Regression Using SAS: Theory and Application, Second Edition," (Allison, P., SAS Institute, 2012).