Telling an Impossible Lie: Detecting Individual Cheating in a Die-under-the-Cup Task

Dishonesty is abundant in modern economic life. Over the past two decades, behavioral economists and social psychologists have been designing numerous lab and field experiments with the purpose of deriving insights on people?s tendency to cheat. A popular cheating experiment is the Fischbacher and Föllmi-Heusi (2013) die-under-the-cup paradigm, where subjects are asked to roll a six-sided fair die in private (under a cup or at a remote corner of the room) and are promised a payoff according to the outcome of the roll (e.g., 1, 2, 3, 4, 5 or 6 dollars for the corresponding die number rolled) which they report to the experimenter. While the die-under-the-cup task provides incentives for dishonest overreporting of the actual die outcome, there is no way to identify dishonesty on the individual level. It is only possible to elucidate the aggregate level of dishonesty among subjects as a group by comparing the average reported outcome to the expected outcome of 3.5 in a fair die roll or the percentage of higher reported outcomes, such as 5 and 6, to their expected outcome of 16.7 percent. The present paper reports the results of running a variant of the standard die-under-the-cup experiment which enables to detect dishonesty on the individual level. Individual dishonesty is detected when one reports an outcome which is practically impossible. Subjects are uncertain about the possible outcomes (other than the one they actually rolled) and may rationally refrain from telling a lie which, if impossible, might result in embarrassment and denial of any payoff. Would they be tempted to tell such a lie when faced with a proper incentive? Applying two payoff levels, we find that about 55 percent of subjects opted to tell an impossible lie even when the payoff was relatively low and about 70 percent did so when the payoff was doubled. Furthermore, about 30 percent and 50 percent lied to (what seemed to be) the maximum extent possible under the low and high payoff, respectively. These findings are in sharp contrast with Fischbacher and Föllmi-Heusi (2013) where the possible outcomes were known to subjects yet the reported outcomes remained stable even when the payoffs were tripled, or with Mazar et al (2008) who concluded, under a different riskless paradigm, that subjects cheated just a little bit and that the modest level of cheating was insensitive to the reward from cheating.