06-20-2013 10:56 AM
Anyone have any insights into proc logistic and using the R2 value out of it?
I recently suggested it and had a consultant tell me it wasn't valid, but SAS produces it and I've read articles that suggest its a valid measure, though it doesn't have the same meaning as in linear regression.
Thoughts are appreciated.
06-20-2013 11:08 AM
The current fashion with binay outcomes in the logistic model is to use the C statistic, based on ROC curves. Available in SAS and is interpretable in terms of PPVs. Rather than trying to explain away variation in the outcome variable, you are attempting to determine how well your model predicts that outcome, so to speak.