Have an idea?

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

Incorporating "extreme response behavior" in WTP-estimates

Hello, everyone,

I would like to know how you proceed with respondents showing so called „extreme response behavior“, i.e. when a respondent always or never chooses the none-option?

A few words about my study first: My study is about electricity tariffs. Therefore, every survey participant uses/purchases this product. I used a pivotal design, i.e. a design where the none-option is displayed as the current product. To describe the current electricity tariff, I use the CBC attributes and levels. I asked for these levels at the beginning of the survey.

To avoid context effects, I separated the forced choice part, i.e. the part without a none-option, and the free choice part, where only the tariff chosen before and the current tariff were displayed. In total, the respondent had to complete 24 choice tasks (12 forced choice sets + 12 free choice sets).

Out of 838 respondents, more than 200 always chose the none-option, i.e. the current electricity tariff. Nearly 40 have never chose the none-option.
Now I am asking myself: What should I do with these responses? I think that I have the the following options:

a) Joint estimation with all data from the free choice tasks
b) Joint estimation, but discard data from the free choice tasks where extreme response behavior occurred
c) Separate models with all data from the free choice tasks
d) Separate models, but discard data from the free choice tasks where extreme response behavior occurred

I doubt that Option a) makes sense in my case. Furthermore, I assume that the cognitive process when deciding to change the electricity tariff (yes/no) is different compared to when someone is forced to choose between different options. Consequently, I have a slight tendency towards options D.

But what do you think? What are your experiences?

I am looking forward to your answers! =)

asked Nov 19, 2020 by Nico Bronze (1,060 points)

1 Answer

0 votes

In cases like this I first remove the folks who always choose the none option.  These are folks who really aren't in the market for the product and who arguably might even be left out of the simulator (AND the calculation of market size.    

Then I use the remaining respondents in the  joint analysis.   

For the simulator, if I include them at all, I have the 200+ who always choose none in the survey also always choose none in simulation.  I simulate the choices for the other respondents as a function of their utilities.
answered Nov 23, 2020 by Keith Chrzan Platinum Sawtooth Software, Inc. (111,275 points)

thank you very much for your answer.

You wrote that you remove "[...] folks who always choose the none option." Does it make sense to exclude only those tasks of the respondents where extreme response behavior occurred, i.e. only the tasks of the free choice are discarded and not the tasks of the forced choice?

I assume that respondents who would not buy the product still have valid preferences regarding the product characteristics if they are forced to choose between the alternatives.

What do you think?
Nico, I do not know what to assume about those respondents.  Perhaps they're people who do have valid preferences.  Perhaps they're people who really aren't "in" the market for that product category at all.  If you're trying to make predictions about behaviors in the market, I doubt a little that the opinions of these folks will help you predict anything about relative preferences for products among the folks who ARE in the market.
Ok, thanks Keith!
Hey Keith,

sorry to bother you again, I was wondering if it makes sense in a model estimation to capture preferences for the current product with an alternative-specific constant? In this case, however, I would not use a constant, but a random variable that is normally distributed or - to model extreme response behavior - lognormally distributed.

The advantage would be: (i) Respondents would not have to be screened out of the data set. (ii) The preferences (=beta coefficients) for the current product would be finite.

The reason I ask is that my goal is to estimate willingness to pay. It sounds implausible to me to say that wtp is infinite ( =out of market) for respondents with extreme response behavior.
That is an option, yes.  These folks end up with very large utilities for the None parameter.  Form a practical standpoint, their large None utilities work in the simulator to keep them from making product selections in the market, so the effect is the same.