Preferences for Hedging in Multiple-Choice Tests

Abstract

This paper proposes two novel scoring rules which allow the test-takers to hedge when responding to multiple-choice questions. We randomly assign 2,986 participants in an IQ test into three conditions. In the control condition, participants choose one option, and receive one point for correct response. In the treatment conditions, they can explicitly hedge by choosing k options: if the correct option is among the k chosen options, they receive 1/k point in the outcome-mixing treatment, and one point with probability 1/k in the probability-mixing treatment. We find that participants in both treatments hedge pervasively and score lower compared to those in the control. The observed differences depend on risk preferences of the participants. While scores in three conditions exhibit similar psychometric quality, we observe a significant correlation between academic performance and IQ score measured in the probability-mixing condition, but not in the other two conditions. These observations contribute to the literature on preference for hedging, the design of multiple-choice tests, and the relationship between risk preferences and cognitive ability.

Keywords: convex preference, deliberate randomization, cognitive ability, multiple-choice test, experiment

JEL Classification: C91, D81