topic Need some help in Mathematical Optimization, Discrete-Event Simulation, and OR

Need some help

deleted_user — Wed, 14 Nov 2007 20:01:53 GMT

I have some data with historical binary target and it's risk driver. How can I in sas segment risk driver onto groups with (almost) homogeneous values of my target. (The mean value of the target within a segment should be around it's actual values, but segments should be differ from each other)

Re: Need some help

Doc_Duke — Fri, 16 Nov 2007 15:28:36 GMT

It seems like you could use something like the Hosmer & Lemeshow test for goodness of fit. It works with predicted risk under the logistic regression model. H&L is not a very powerful test, so I prefer to use a visual analogue to it. Sort the predicted scores, divide into deciles (quintiles if the N is small), and then plot the midpoint of each decile against the proportion of positive outcomes for the observations in that decile.

If your risk score fits well the line through those 10 points will be reasonably straight. It's basically testing the reliability of your risk driver(s).

Doc Muhlbaier
Duke

Re: Need some help

deleted_user — Tue, 20 Nov 2007 20:17:19 GMT

Thanks, that is what I am doing so far. Just thought that somebody maybe automated this procedure in sas. 🙂