Have an idea?

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

Anchored Boosted/ Bandit MaxDiff - HB model

I am aware that you suggest aggregate logit for bandit Maxdiff, but would HB model work well for limited size bandit maxdiff exercise?
I am to test 43 items (20 sets, with 4 items per set), calculate the anchor (assuming that in direct anchoring questions respondents face a subset of the items [using the Bandit MaxDiff constructed list command]) and will need to have individual level utilities. Would it be possible to apply Bandit Maxdiff and run HB to have individual level utilities e.g. assuming that the levels omitted by a given respondent have utility equal to the utility of the worst item that was evaluated by this particular respondent?
Best regards,

related to an answer for: Anchored Bandit MaxDiff
asked Nov 26, 2018 by RafalNeska (450 points)

1 Answer

0 votes
Best answer
The software wouldn't provide an automatic way for you to force the utilities of items not drawn (via Thompson Sampling) onto the constructed list to have lower utility than the worst item evaluated by this particular respondent.  Beyond the mechanics, there is a lot of heterogeneity in tastes, so to assume that items not drawn onto the list via Thompson Sampling (which is based on pooled means and standard errors) would have lower utilities for an individual compared to those that had been drawn onto the list seems like too much of a stretch to me.
answered Nov 26, 2018 by Bryan Orme Platinum Sawtooth Software, Inc. (172,690 points)
selected Nov 27, 2018 by RafalNeska
Bryan, many thanks for your explanation, best regards, Rafał