Using Constructed Lists in CBC -Analysis Considerations

Question

Hello all,
I am working on a CBC project and one of my attributes is Brand. I only want respondents to be exposed to the brands they indicated they have used in a prior question. It seems easy enough on the front end, per Bryan's note here (https://www.sawtoothsoftware.com/forum/7553/constructed-used-populate-levels-attribute-different-brands?show=7553#q7553).
However, I was hoping for some guidance on the back end adjustments I should be thinking about.  To complicate things further, I have 3 prohibitions between certain brand & attributes.

While switching to ACBC would be ideal, I'm already really far down the line with my CBC survey, about to go to field (and the client just sprung this new requirement on me!). Can someone give me a high-level run-down of the analytical adjustments I should be ready for?
Thanks!

Answer 1

One of the issues I'm sure you're already watching is that the number of brands used for each respondent must be the same for CBC.  Then, you have to decide what assumption to make about the dropped brands per respondent.

If you believe the dropped brands should be worse than included brands for a respondent, then I recommend doing some additional back-end massaging of your .CHO or .CSV file for CBC data.  You add a new level of brand (the "threshold" brand) for each respondent (that you end up ignoring in the end after utility estimation) and you compare for each respondent the included brands (one-at-a-time) to the threshold brand and you say for each of those paired comparisons that the respondent "picked" the included brand.  You also compare each of the dropped brands to the threshold brand and you say for each of those pairs that the threshold brand was picked.  For these augmented paired comparisons, you set the other attributes to 0.

Actually, using these augmented tasks to estimate brand preferences actually helps you out if you have prohibitions involving brands, since they'll be unaffected by the prohibitions and will actually lend additional stability to the brand estimation.

Comments

Thanks, Bryan! I did have a plan to ensure the same number of brands for each respondent...they have to be aware of at least 3, or they skip the exercise and terminate. Then everyone gets 3 (randomized, of course). The threshold brand approach was the piece I was not sure about; thanks so much for clarifying it!