We propose a class of mechanisms designed for eliciting honest responses to a multiple choice question (MCQ) when the truthfulness of these responses cannot be directly verified. Choice-matching mechanisms assign each respondent a score consisting of two terms: his score for predicting the answers of other respondents and the average prediction score of those respondents who give the same response to the MCQ.
We show that these mechanisms are truth-inducing when beliefs of respondents with the same truthful answers are sufficiently similar. We discuss this assumption extensively and compare our proposal to existing methods, which we argue are either less suitable for practical implementation or require additional assumptions on the setting.