WTP far above the reasonable price

Question

Hi,
I have a question regarding WTP calculation in my research. I've calculated the WTP of the levels of each attribute that I used in the CBC survey. I included 7 fresh fruit attributes in the survey (including price) and each attribute consists of 2-3 levels. I have also segmented respondents using Latent Class analysis to find out the differences in preferences and WTP of each group, where the results of this LC analysis resulted in 7 segments of respondents. I've calculated the WTP using the market simulator in lighthouse studio version 9.11.0, and run it all by default with the simulation method is Share of Preference, and the Behavior Range is Willingness to Pay. However, what bothered me, I found that there were several levels in some segments where the WTP value was very high, exceeding the reasonable price. These levels, indeed have a high utility value far compared to levels on other attributes. Is it possible in the calculation of WTP? Or maybe I did something wrong when I did the calculation using the software? Because I think consumers may not be willing to pay a kilogram of fresh fruit for more than 34 USD, with a relatively low income level in my research location.

Thank you..

Answer 1

Things to consider are:

1.  Did you carefully recruit people who are in the market to buy?
2.  Did you do a good job cleaning out random responders (to conjoint questions) and speeders?  (sometimes 30-40% of respondents fail basic "randomness" and "speeding" checks, and these respondents can vastly  inflate WTP estimates)
3.  Did you constrain price to be monotonically descending in utility as price increases?

Comments

Hi Brian, thank you for your response.

Yes, in my survey, I only recruit the right respondents, i.e. respondents who like, enjoy consuming and have recently purchased the product I am analysing. I also have conducted the cleanout process by removing respondents who answer the questions in a very short time compared to reasonable time, and also respondents who only answer the questions monotonously (e.g. only answer choice number 1 in all choice sets).
However, as you have explained in point 2, maybe I should check again whether I have done the correct cleaning out to make sure the respondents included in the analysis are the correct respondents.
I'm really sorry, I don't really understand your explanation in point 3. Is this related to the design of the price attribute and its levels?

---

In HB analysis within our software, you can click the "gear icon" (settings button) and find the setting that is "Constraints" and select that the Price attribute should be constrained.  This is not a design thing, but a utility estimation thing that will make sure that the price attribute is constrained to have lower utilities for higher prices for each respondent.

It seems that you did not look to clean respondents who answer near randomly.  You can see an explanation of how to identify respondents who essentially are answering randomly at:  https://sawtoothsoftware.com/resources/technical-papers/consistency-cutoffs-to-identify-bad-respondents-in-cbc-acbc-and-maxdiff