Best Practices for Calculating Importance Scores

Last Updated: 23 Sep 2013Hits: 10121
In reading a Sawtooth Software article, The Basics of Interpreting Conjoint Utilities, it said, “When summarizing attribute importances for groups, it is best to compute importances for respondents individually and then average them, rather than computing importances using average utilities.” I would like to understand the implications from the two procedures. I tried to calculate in both ways and found that, at least in my example, the most important and the least important attributes are the same from both procedures; however, the remaining attributes in-between have different scores and rank orders. Mathematically, it seems to me both procedures are valid, but I am not sure how to interpret the differences in the results and why one would be better than the other.

Imagine that you do a study that has two attributes: Brand and Price. Imagine the two brands are Coke and Pepsi, and the two prices are $1.00 and $2.00. Imagine that brand is critical to the respondents (more so than price) and that half the respondents love Coke and the other half love Pepsi.

Now, if you compute the importance scores based on average utilities, the average utilities for Coke and Pepsi are tied, so the importance of price will be shown to be 100% and brand 0%. This is obviously wrong, as differences in opinion regarding brand across the sample have caused the signal for brand to wash out in the average.

If utilities are computed at the individual level and the importance scores are computed at the individual level, then it will clearly capture the fact for each respondent that brand is critical (despite the differences in opinion).