Card or dice? An improved experimental approach to measure dishonesty

Abstract

We introduce a modified die-roll experiment carried out in Germany and Cambodia to measure precise dishonesty rates, while the individual lie is not observable to the experimenter. Participants draw an envelope from a box containing many envelopes. Each envelope contains a card depicting a die number, which participants view in private and then deposit into a different box filled with many envelopes. The payoff of participants depends on the reported number, thereby creating an incentive to dishonestly report numbers with higher payoffs. Although the individual lie remains hidden from the experimenter, the drawn distribution of cards by a group of participants is known. Results of the modified experiment are compared to the classical die-roll task, in which individual dishonesty is private information and the outcome distribution is assumed, based on a probability function. The comparison reveals that the modified card method shows comparable levels of lying to the classical die-roll task among students, but not among smallholders in rural Cambodia. Considering the farmers, the number of liars is lower in the card task compared to the die-roll task. Although the individual lie is not observable, we find partially different dishonesty proportions between numbers comparing the two tasks. This suggests that the observability of the drawn distribution affects the costs of lying.