topic ROC in PROC LOGISTIC using weights in Statistical Procedures

ROC in PROC LOGISTIC using weights

deb193 — Mon, 15 Aug 2011 21:04:59 GMT

Hi all -

I have run a PROC LOGISTIC with and without my weight variable. The regression seems fine and agrees SPSS and Stata where I specified the same model. More of hte predictors are signifigant for the weighted data, and the overall R2 is higher. (I did have to tell SAS to model event=1 to get the same outputs.) However, the area under the ROC curve gets smaller in SAS weighted model, where it should be increasing. It increases when I produce ROC after weighted logistic in Stata, and when ROC curves are obtained in SPSS.

It almost looks like the ROC plot is not using the weights - except the area under the curve is even smaller than the ROC area form theunweighted analysis.

I am going to try generating a score dataset, and computing the sensitivity and susceptibility to do my own ROC calculations - but I though I would ask if anyone else has seen this.

Thanks

ROC in PROC LOGISTIC using weights

Rick_SAS — Mon, 15 Aug 2011 23:37:40 GMT

By default, weights are used to fit the model, but not to compute the ROC. See if the ROCOPTIONS(WEIGHTED) option on the PROC LOGISTIC statement addresses your needs.

Rick

ROC in PROC LOGISTIC using weights

deb193 — Tue, 16 Aug 2011 14:01:14 GMT

This worked perfectly. However, it begs the question, "Why is this the default?". In other words, what is the value of ROC and AUC without weights when weights have been used? IS there some downside or bias to the weighted ROC?