Wednesday, August 3, 2022

Widely cited psychology study, suggesting that winners of competitions are more likely to cheat subsequently, fails to replicate

Does competitive winning increase subsequent cheating? Andrew M. Colman, Briony D. Pulford, Caren A. Frosch, Marta Mangiarulo and Jeremy N. V. Miles. Royal Society Open Science, Volume 9, Issue 8, August 3 2022.

Abstract: In this preregistered study, we attempted to replicate and substantially extend a frequently cited experiment by Schurr and Ritov, published in 2016, suggesting that winners of pairwise competitions are more likely than others to steal money in subsequent games of chance against different opponents, possibly because of an enhanced sense of entitlement among competition winners. A replication seemed desirable because of the relevance of the effect to dishonesty in everyday life, the apparent counterintuitivity of the effect, possible problems and anomalies in the original study, and above all the fact that the researchers investigated only one potential explanation for the effect. Our results failed to replicate Schurr and Ritov's basic finding: we found no evidence to support the hypotheses that either winning or losing is associated with subsequent cheating. A second online study also failed to replicate Schurr and Ritov's basic finding. We used structural equation modelling to test four possible explanations for cheating—sense of entitlement, self-confidence, feeling lucky and inequality aversion. Only inequality aversion turned out to be significantly associated with cheating.

3. Discussion

Schurr & Ritov's [1] experiments were severely underpowered and vitiated by other design and methodological problems. In particular, their basic finding that competitive winning is associated with subsequent cheating was based on a study in which participants were not assigned randomly to experimental and control treatment conditions. Our study 1 replicated Schurr and Ritov's study as closely as possible with adequate power and random assignment to experimental and control conditions. We observed significant levels of cheating in both experimental and control conditions but failed to replicate Schurr and Ritov's basic finding of higher cheating by winners, although the experimental manipulation of winning or losing in both of our experiments was identical to Schurr and Ritov's. We also found no evidence for any significant effect of competitive losing on cheating in the subsequent game of chance.

In study 2, we tested the hypotheses that competitive winning or losing is associated with subsequent cheating in an even larger experiment, conducted online, with participants assigned randomly to winning, losing, paired control, and unpaired control treatment conditions. Once again, we observed significant levels of cheating in all treatment conditions but found no evidence to support the hypotheses that either winning or losing is associated with subsequent cheating. There was no significant difference in cheating between our paired and unpaired control conditions—whether cheating was associated with money being taken from another participant or from the experimenter.

This study also included an investigation, using SEM, to test the hypotheses that winning is associated with a latent variable that we labelled 'pride', indicated by self-confidence, a feeling of luckiness, and a sense of entitlement, and that pride is associated with subsequent cheating, or that losing is associated with a latent variable of 'shame', indicated by a sense of entitlement and inequality aversion, and that shame is associated with subsequent cheating. We measured all the indicator variables with psychometric scales that showed high reliability in our study, and the only significant association that emerged was between inequality aversion and cheating. This suggests that participants who were least inequality-averse were most likely to cheat in the coin-flip game, whether they had won or lost the previous competitive perceptual task. The association of inequality aversion with cheating was not strong, but it is worth investigating experimentally. It may reflect a more general sense of fairness among participants who are inequality averse. If those who value fairness strongly tend to be inequality averse and also construe cheating as a form of unfairness, the association would be explained, but that explanation requires further experimental evidence.

One key question that needs to be addressed is why the results of both of our studies failed to replicate Schurr & Ritov's [1] basic finding that competitive winning is associated with subsequent cheating in a game of chance. One possibility is that Schurr and Ritov's finding, based as it was on a severely underpowered study without proper random assignment to experimental and control groups, cannot validly be inferred from their results. A second possibility relates to their unusual methodology, in which half the participants in every testing session were randomly assigned as passive participants, whose only role was to receive the money that the other half—the active participants—left behind after taking what they claimed was owing to them after the dice game. In our studies, the participants were told that the money that they left behind would go to ‘the other person you are paired with’, with the implication that it was one of the other participants, and in that sense, from the participants' point of view, it was similar to what Schurr and Ritov's participants believed. However, the active participants in Schurr and Ritov's experiment believed that the money they left would go to others who had done absolutely nothing in the experiment, whereas the participants in our replications could have believed that the money would go to others who were fully participating. All of Schurr and Ritov's participants who rolled the dice were told that ‘The rest of the money will go to one of the participants sitting in the lab who did not play the two-dice-under-a-cup game’ [1, p. 1757]. This might possibly explain the failure of our experiments to replicate Schurr and Ritov's basic finding if their winners, in contrast to ours, believed that the recipients of money left behind were more deserving of being cheated, but that would suggest that the basic finding applies only in the artificial context of their experimental setup or in very limited and unusual circumstances. In everyday life, people who cheat rarely, if ever, know that their victims have done nothing to earn the money out of which they are being cheated. A third possibility is that the discrepancy between Schurr and Ritov's findings and ours arises from a cross-cultural difference between students in Israel and the UK; but we are unaware of any evidence that might support that interpretation, it is very unlikely given that Israel and the UK are both WEIRD (western, educated, industrialized, rich, democratic) cultures [36], and if correct it would severely limit the generality of Schurr and Ritov's basic finding (and also, by symmetry, our own basic finding).

Given the published evidence that more cheating tends to occur in online than laboratory studies [26,27], it is worth noting that we found no such difference. In study 1, incentives were lower than in study 2 and participants cheated, on average, by 31p out of a possible maximum of £3.00 (10.3%), while in study 2 they cheated by £3.25 out of a maximum of £50.00 (6.5%). The incentives were much greater in study 2, therefore participants' cheating translated into greater monetary terms in study 2. In our laboratory-based study 1, using Cohen's [11] index of effect size, the overall effect size of cheating was d = 0.46 and in our online study 2 it was d = 0.42. The finding of such a negligible difference can perhaps be explained by the fact that the dice-under-a-cup game that we used in the laboratory in study 1 provides an opportunity for cheating that seems almost entirely ‘safe’, in the sense that it would be impossible to detect a particular instance of cheating. If that interpretation is right, then our online experiment, in which the corresponding task was a coin-flip task, may not have provided a significantly greater sense of security, and participants may have felt equally disinhibited from cheating in both experiments. Thus the general cheating that we found across all conditions would be in line with recent evidence that cheating tends to occur particularly when it is unobservable by the experimenters [37].

Our studies have not provided much enlightenment as to what leads some people to cheat. In both studies, cheating occurred at a low but significant level in all treatment conditions, and the only psychometric variable that correlated significantly with cheating was inequality aversion. Our SEM revealed only one path that reached statistical significance, from shame to number of heads claimed and hence cheating. One of the indicators of shame was inequality aversion. Further research is clearly required to determine whether inequality aversion is indeed causally related to cheating and if so why. One possibility is that inequality aversion is associated with a more general concern for fairness and that people who value fairness are less likely to cheat because they perceive cheating as a form of unfairness, but without further evidence this interpretation remains speculative.

The aim of study 2 was to discover variables, possibly but not necessarily including competitive winning or losing, that might explain cheating in a subsequent game of chance. The SEM should reveal whether, and if so how, winning or losing is implicated. We hypothesized that sense of entitlement, self-confidence, personal luckiness and inequality aversion might help to explain cheating. For example, if Schurr & Ritov [1] were right, then winning should be associated with sense of entitlement and sense of entitlement should be associated with cheating. Sense of entitlement is interpreted by the authors of the scale that we used to measure it [31] as a personality trait, and we should perhaps expect a personality trait to be largely unaffected by an experimental manipulation such as winning or losing. However, the SEM does not require winning or losing to play any part in the potential relationship of any of the other variables to cheating. For example, we might have found that trait sense of entitlement is associated with cheating, irrespective of any association with winning or losing, just as we did, in fact, find that inequality aversion is associated with cheating without any significant association with winning or losing.


No comments:

Post a Comment