The effect of gain and loss framing on cheating: Evidence for the null

We followed up on previous results showing increased cheating under the threat of potential losses compared to the promise of equivalent gains, as well as inconsistent findings in this literature. Our studies used diverse paradigms, including random number reporting, binary number reporting, performance-level reporting, and reliance on illicit resources. In seven studies of online workers (n = 3,803), we found that participants tended to cheat, though the effect size of cheating (Cohen’s d) varied from 0.14 to 1.18 in different settings. However, in all studied paradigms, we observed no significant effect of gain and loss framing, with an overall effect size of d = 0.004, and with the variance in different studies accounted for by sampling error. Examining the moderating effect of stake size did not yield significant findings. At the individual level, higher cheating was predicted by loss aversion, but, on average, participants did not exhibit loss aversion for the obtained incentives. Thus, we cannot overrule the possibility that the inconsistencies in the literature might simply be due to sampling noise around an extremely small (or zero) effect.