Have an idea?

Visit Sawtooth Software Feedback to share your ideas on how we can improve our products.

Approach for segmentation for CBC/HB


What is the best approach for segmentation based on results of CBC /HB?
Should we use:
Individual partworth of respondents?
Importance score of respondent?
Or based on the result of the market simulator in a competitive setting?

Best regards
asked Aug 9, 2017 by Robin59 Bronze (545 points)

1 Answer

+1 vote

I've seen it done all the ways you describe, but by far the most often on the utilities.  You don't want mere scale differences to pollute your results so use the ZCD utilities rather than raw.  You might also consider doing latent class MNL to create segments.
answered Aug 9, 2017 by Keith Chrzan Platinum Sawtooth Software, Inc. (116,175 points)
Hi Keith,

Thanks for your answer. It raised three other questions:

1. Would you please elaborate what you mean by second sentence? Why ZCD utilities are better for segmentation?
2. What is the formula to calculate ZCD? I know that the first step is to have a zero centered raw utility ( the sum of all levels utilities within an attribute is zero) What are other steps? I have read Bryan following post but i have problem understanding step 2 and 3.
3 In general, would you please make a comparison between application of  zero centered raw utility and zero centered diffs.

Many Thanks
If you are familiar with the logit scale factor, it has to do with the fact that utilities from respondents who answer the CBC questions more reliably will tend to be larger across the board than utilities from less reliably-answering respondents.  The scale factor acts as a multiplicative factor that makes some respondents' utilities and others smaller.  You don't want this mere difference in reliability to impact your segmentation and the ZCD transformation is an attempt to partially remedy this problem.

I'm not sure I can do better than Bryan's explanation of how to calculate ZCD.