Have an idea?

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

Integrating Dual Response into ACBC response data

I started manipulating my response data from a ACBC excercise for the purpose of estimating alternative specific price parameters (to avoid collinearity between summed price and certain levels that impact the summed prices significantly).
Once I'm done, to my understanding I won't be able to apply the calibration function for the None-Threshold. So instead I decided to add a yes/no question to every tournament task ("Would you really buy your preferred option?") to calibrate my "none".
Now that I'm coding these dual response tasks into the response data I realize that those tasks are identical to the screening tasks for the respective concepts. In the screening I phrased the question "Would you generally consider this concept?" and in dual response task I ask whether the preferred concept would actually be bought in real life.
HB won't make a difference between the two question phrasings and, therefore, I wonder what the effect on the preference parameters will be if the same concept was chosen over "none" in the screening section but would not be bought according to the tournament dual response task. Does this contradicting response regarding the same concept "cancel out"? Does it add noise or does it possibly cause problems in the estimation?
Thanks for some thoughts on this,
asked Oct 30, 2014 by alex.wendland Bronze (2,430 points)

1 Answer

0 votes
Best answer
Respondents make errors in choice tasks all the time.  Logit-based estimation methods have good error theory to deal with inconsistencies.  If a concept is first rejected compared to the "None" and later accepted compared to the "None" (by the same respondent), then the information from those two tasks tends to cancel out and the information from those two tasks would want to nudge the utility of the concept very close to the utility of the None.  The more inconsistencies the respondent demonstrates, the more logit-based estimation approaches tend to reduce the magnitude of all the part-worth utilities for the respondent (reduce the Scale Factor).
answered Oct 30, 2014 by Bryan Orme Platinum Sawtooth Software, Inc. (184,140 points)
selected Oct 30, 2014 by alex.wendland
Thanks again!