Have an idea?

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

How to re-weight CBC importance scores

If I want to get the importance scores from a CBC exercise, the number of levels across attributes should be the same. If not, the one with the more levels would end up becoming more important. Is this correct? If yes, is there a way to re-weight the attributes to deflate the impact of the attribute with the higher number of levels outside of conjoint?
asked Apr 26, 2013 by JKincaid Bronze (800 points)

1 Answer

0 votes
Ideally the number of levels for all attributes will be the same but clearly this will not always be the case:  some attributes are inherently binary (Feature A:  yes/no) while others are quantitative, like price, and we may want to let the data tell us whether some linear or curved or piece-wise function best captures respondents' utilities.    

It is not the case that attributes with more levels automatically and always have larger importnaces than those with fewer levels but it is true that an attribute with two levels will tend to have more importance when that SAME attribute, with the same endpoints, has more levels.  So there is a bias for attributes with more levels to have more importance.

Unfortunately I am not aware of a simple way of weighting this bias away.
answered Apr 26, 2013 by Keith Chrzan Platinum Sawtooth Software, Inc. (55,525 points)
Keith is right on this.

And, I'd add, in the real world, some attributes may be represented as more levels in the available products than others.  So, in the real world, there might be some true "Number of Levels" effects that we would want to somehow mimic in our conjoint design.

Thus, in a quest to somehow "weight this away" in conjoint (if that were even possible), we might end up losing some important pschological "Number of Levels" effect that could help us predicting real-world purchases.