How to normalise HB scores in MBC?

Question

What is the recommended method of normalising scores derived from the MBC software? I presume that for a top level model it is exp(Utility), re-based to the total, with linear terms converted back to their range, or should one zero centre the utility first within each attribute? For sub-models should this can only be done for respondents who 'qualify' for the submodel (i.e. have an RLH), then optionally multiply the sub-model ZCDs by the 'top model' qualifying values. I'm not sure there is an easy way to approach this but want to know if I'm missing a trick as I get my head around the outputs.

Answer 1

I suppose the big question is WHY one would want to do this?  We designed MBC with the thought in mind that beta weights shouldn't be shown to clients for MBC.  The focus would be on market simulators and sensitivity curves drawn using market simulators.  

The answer will probably be that the client asked for the weights.

The trouble is that there are often multiple models built in MBC: different models for different Dependent Variables on the menu.  So, one could think about somehow normalizing the betas within each model so that the range of potential weights (when considering the range of X values) was constant across people.  But, that would only hold within each model.  Such normalization would lead to inability to compare weights across different models.  Anyway, it seems like a messy area to me.

Comments

As to the why it was to explore possible segmentation, however, I'm quite happy with the answer as it confirms my own conclusions.