PRIZM codes for marketing vs. LOGIT predictive models - Looking for data

Factual Questions
1

OK I’m an analyst that has built LOGIT predictive models to give individuals a probability to say yes or no to a marketing campaign. Models have always shown lift over random and are validated through testing before a general roll out. What I’m running into with some of my clients is that they are looking at PRIZM codes as a way of doing this same thing. While I agree that segmentation has it’s uses, when it comes to predicting who will respond, a group level segmentation that is indexed is just not as effective as an individual score.

So, I’ve been trying to find some kind of independent evidence that this is the case. But when I Google, I’m not finding anything except PRIZM code testimonials, no comparision of methodologies.

Any ideas?

2

I’ll give this one bump.

3

One last time.

4

I’m pretty sure that individual probabilities will be much better than segmentation. It has been shown repeatedly that Hierarchical Bayes will generally give better results than the latent class model you’re describing-- assuming that you have a decent amount of past data. In Rossi and Allenby’s (1993) paper about this subject they say:

For more info, check papers by Greg Allenby on Google Scholar. Here’s a few to get you started. The one quoted above is the first one:

http://www.jstor.org/stable/3172826
http://www.jstor.org/stable/3152035?cookieSet=1
http://www.jstor.org/stable/193195
http://www.jstor.org/stable/4129743

5

Thanks, Cagey, I will definately review.