How to deal with utility reversals?

Question

Hi,

In our last study in some countries (especially in china - maybe due to cultural reasons) respondents prefer higher prices over lower prices. I guess it's because of the used conditional pricing tables.

How should we deal with it? Is it better to remove respondents with utility reversals (between 2/5 and 3/5 of the country samples) or should we ignore it and try to work with subgroups?

Does anyone have a good advice for me?

Regards,
Alex

Comments

What do you mean, "I guess it's because of the used conditional pricing tables."

Conditional pricing tables shouldn't affect the number of reversals on price, if done properly.  I could envision, however, a situation in which price is made conditional on essentially all the other attributes and not enough variation in price from low to high for each conditionality leads to poor precision in the price part-worth estimates.

---

Sry Bryan, I have to be more specific. I mean, that I'm questioning the quality of our used conditional pricing tables. I suppose, that the utility levels are reversed, because we have used a too small price range from low to high (-20%/X/+20%) and/or too big or small price gaps between the starting points (priceniveaus).
(We have used three attributes (brand/feature/price) and the price is conditional on the attribute brand...)

---

Would you recommend the use of constraints on price in this case?

Answer 1

Well, I think it's a puzzle why the data are turning out like they are, and it makes me wonder whether isn't something just fundamentally going wrong with the way the respondents are understanding the questionnaire, who those respondents are, and whether or not they are able to put themselves in the mindset to make realistic answers similar to what they would do in the real world.  I hesitate to think about just applying the constraints band-aid in this case without coming to a deeper understanding of whether there is systematically something going wrong with the experiment.

Comments

Oh, I just thought of another idea...but it involves custom coding of the design matrix.  If the peculiarities of the conditional pricing design were causing problems, you could code up (user-specified coding) the price variable as a single continuous variable.  Then, you fit a log-linear or a linear term to the single price coefficient.  If the conditional pricing gaps have problems like you are suggesting, perhaps this is a way to unravel it.