Bayesian Truth Serum Fused Conjoint

Conjoint methodology has been widely used in marketing research to answer a variety of business questions. However, it is usually performed in hypothetical settings, which raised questions by both practitioners as well as academicians on its validity. Recent studies investigated whether implementing incentive-compatible schemes would improve conjoint analysis predictive power. The results have been positive so far, but there are certain disadvantages associated with each approach. In this paper I will explore a relatively new incentive-compatible mechanism, the Bayesian truth serum. The method assigns high scores to respondents when they have surprisingly common answers, e.g. when the frequency of their answer is higher than commonly predicted by the rest of the respondents. The Bayesian truth serum model outperformed the control group on several key metrics providing first plausible results for the adoption of the method into the preference measurement methodologies.