Logits, Weights, and R-Squares OH MY

[louisehadden]

I am running a series of logistic models for an analysis using 4 different survey sample data bases. Each has a stratified sample and sample weights (and psu/cluster information). I am running SAS 9.2 Maintenance Release 1. I ran a sample model using PROC LOGISTIC with a weight, PROC SURVEYLOGISTIC with a weight and stratum variable, PROC LOGISTIC unweighted, and then in desperation SAS-Callable SUDAAN 10 PROC RLOGIST (Logistic.) The R-Squares using PROC LOGISTIC with a weight are improbably high, and so are the R-Squares from SURVEYLOGISTIC with a weight and stratum. The R-Squares for unweighted LOGISTIC are more reasonable although the adjusted one is high. The R-Square for the SUDAAN logistic seem reasonable, a little higher than the unadjusted unweighted logistic. What do I believe here? Is there something weird about the SAS logistics and r-squares and weights??? Is there something I can specify to "fix" this? Any help appreciated!

Code for various procs

proc surveylogistic data=anal2002;
model bmi30p (descending) = &t2ivar1 / rsquare corrb covb;
strata stratum;
weight weight;
title2 'WEIGHTED SURVEYLOG';
run;

proc logistic data=anal2002 descending;
model bmi30p = &t2ivar1 / rsq corrb covb;
weight weight;
title2 'WEIGHTED LOG';
run;

proc logistic data=anal2002 descending;
model bmi30p = &t2ivar1 / rsq corrb covb;
title2 'UNWEIGHTED LOG';
run;

proc rlogist data=anal2005 design=wr notsorted;
nest stratum nfsu;
weight finalwt;
class &t2ivar1;
model bmi30p = &t2ivar1;
TITLE2 'PROC RLOGIST (sas-callable SUDAAN)';
run;

THANKS IN ADVANCE!!

[Doc_Duke]

The one contribution that I can add to your dilemma is that LOGISTIC is inappropriate. See this comment from the documentaiton:

"Caution:PROC LOGISTIC does not compute the proper variance estimators if you are analyzing survey data and specifying the sampling weights through the WEIGHT statement. The SURVEYLOGISTIC procedure is designed to perform the necessary, and correct, computations.."