10-07-2011 02:28 PM
Any time I get asked to convert well behaved continous data into discrete classes/clusters/dichotomies, I begin to wonder--why isn't the more powerful continuous case good enough? What could be gained by lumping stuff together? I realize that you can't offer someone 0.82457 of a credit card, and I have been doing linear models with class variables since mammoths roamed, so I'm not totally against dichotomizing a response, or categorizing into classes, but I always wonder.
10-28-2011 10:46 AM