# Willingness-to-pay (WTP) from CBC/HB estimation

I would like to calculate the standard marginal WTP for different attribute levels based on my CBC data. The products I investigated using CBC, consisted of 5 attributes (one of which was the attribute Price), each consisting of 4-5 levels.

As I do not use the market simulator, I would like to ask, whether the steps I took for the WTP analysis are correct:

1. I first designed a default product that I use to examine WTP increases and decreases in reference to this default product in order to facilitate interpretation of the results.
2. Using the CBC/HB estimation data file that I extracted from Sawtooth, I calculated a linear price coefficient that has a value between -1 and +1 based on the Individual Utilities (ZC Diffs).
3. I apply the formula ((default utility 1 - utility 2)/price coefficient) to get the WTP values.

My WTP values range from about -\$23 to +\$17. Based on the results, I don't have reason to believe there is a mistake in my analysis steps. However, comparing the results from the WTP with the results from the attribute importance scores, I notice that the attribute importance scores show that one specific attribute yielded higher importance here than it did in the WTP results. In other words, I find a slight discrepancy in the weighting and importance of one of the attributes between the attribute importance scores and the WTP results. Can anyone explain why this may be the case? Or did I indeed make a mistake calculating the WTP from my CBC/HB data?

I am happy to provide more detailed information if needed.

Looking forward to your reply. Thanks!
asked Dec 18, 2017

## 1 Answer

0 votes
Well, the two calculations are only similar and they're different in an important way.  Your importance metric is based on the difference between utilities for the very highest and the very lowest level on each attribute.  But your WTP calculation is based on the difference between a given level and the default level (the level used on your default profile.  While I'd expect these calculations to yield similar rank orderings of results, there's no real reason to expect them to be perfectly correlated, is there?
answered Dec 18, 2017 by Platinum (85,200 points)