What does a young cheater look like? An innovative approach

Personality traits matter. We could summarize with these simple words a vast literature in social sciences dedicated to explaining the behavior of individuals, acting alone or within societies. Examples abound in several economic contexts spanning from the intention to become a social entrepreneur (Nga and Shamuganathan, 2010), to management of household finances (Brown and Taylor, 2014), to labor market outcomes (Fletcher, 2013). Similarly vast is the literature on cheating, justified by the profound economic and social consequences of such behavior. In this chapter, we link these two growing strands of literature by studying the relation between experimental measures of cheating behavior among adolescents and personality measures obtained through a questionnaire. The first consistent finding from the experimental literature on cheating is that some (not all) individuals are dishonest, i.e., when facing the opportunity to lie in order to extract a gain (with the lie typically concerning the result of a dice roll, or a fair coin toss), a sizable share of individual do so: that is, the proportion of individuals reporting a win usually exceeds the objective probability of a win, while still being smaller (often considerably smaller) than one (Abeler et al., 2018). This general result overshadows, however, a huge heterogeneity in the observed individual cheating behavior. Most individuals are willing to cheat only a little (Shalvi et al., 2011), some entirely refrain from lying, while others lie to the maximum possible extent (Fischbacher and Föllmi-Heusi, 2013). This observed heterogeneity, coupled with the fact that individuals who cheat in the lab tend to cheat also in the field (Cohn and Maréchal, 2016), raises the question of which characteristics of an individual’s personality influence her decision to lie. As a precondition for any discussion, we all know that people care about their self-image and struggle to preserve it (Mazar et al., 2008). This struggle imposes a cost, of psychological nature, to the cheater, which changes according to the context. As a matter of example, we know that the decision to lie implies a different psychological cost when people have to report their immoral intentions before acting (Jiang, 2013), when acting dishonestly hurts (or benefit) others (e.g., Fischbacher and Föllmi-Heusi, 2013), when temporally distancing the decision task from the payment of the reward (Ruffle and Tobol, 2014), when individuals are under scrutiny (Ostermaier and Uhl, 2017; Pierce et al., 2015), when they act alone or in groups (Kocher et al., 2018), and when they have a potential accomplice (Barr and Michailidou, 2017). Only recently some papers have considered the importance of personality traits in cheating behavior. In a recent contribution, Pfattheicher et al. (2018) use economically incentivized cheating paradigms (a dice-rolling paradigm and a coin-toss paradigm) to show that, in line with previous literature (Hilbig and Zettler, 2015; Kleinlogel et al., 2018), the basic personality trait of HonestyHumility from the HEXACO personality model (Ashton and Lee, 2007) is negatively related to cheating behavior. That is, they identify a relation between cheating and personality which goes beyond the dark personality traits of narcissism, Machiavellianism, psychopathy, and sadism already studied by Jones and Paulhus (2017)—no effect is found instead when a third-party scrutiny is simulated by presenting the subjects with stylized watching eyes. Interestingly, personality traits have different effects on different types of lies. Jonason et al. (2014) show that while Machiavellianism is related to white lies, narcissism is related to lying for self-gain, whereas psychopathy is related to telling lies for no reason. In a companion paper, Baughman et al. (2014) show that psychopathy predicts scholastic cheating. In another recently published paper, Heck et al. (2018) address the question of power and sample size by exploiting the richness of 16 studies (N= 5002), assessing dishonest behavior in an incentivized, oneshot, cheating paradigm. While confirming the negative correlation between cheating and Honesty-Humility, which was independent of other personality, situational, or demographic variables, they found that one other trait only from the “Big Five” (Agreeableness) was (negatively) associated with unethical decision-making, although the strength of the relation is much lower than with the Honesty-Humility trait. Although cheating, lying, and deception are diffused behaviors both in the adult and in the young population, there is relatively limited evidence of the determinants of cheating among children and adolescents in the economic literature. Some relevant exceptions are Bucciol and Piovesan (2011), GlätzleRützler and Lergetporer (2015), Maggian and Villeval (2016), Korbel (2017), Battiston et al. (2018), Cadsby et al. (2019); see Heyman et al. (2019) for a recent review of this stream of research. In this chapter we use data gathered from an experiment conducted with scout groups from Trentino-Alto Adige, a region in Northern Italy, during their summer camps in August 2017. The experiment, employing a revised version of the fair coin toss paradigm proposed by Bucciol and Piovesan (2011), was followed by a rich questionnaire including, beyond standard demographic questions, a detailed self-assessment of risk aversion, “Big Five” personality traits, level of trust in other people and propensity to break the rules. In our empirical analysis, we employ a principal component analysis (PCA, henceforth), a dimensionality reduction technique aimed at capturing common moments in the data, to gauge the extent to which different personality traits might influence the propensity to cheat. We then reanalyze the decision to cheat by using decision tree classifiers, a very popular technique in the machine learning literature, which also achieve the aim of dimensionality reduction, but focusing on the interaction between variables rather than on (linear) common moments across them. Our results suggest that, while risk propensity is not a strong overall predictor of cheating behavior, self-confidence is irrespectively of the beneficiary of the payment being the individual or the patrol. The use of decision tree classifiers confirms these results, supports the validity of the PCA approach, and further suggests that, among less self-confident subjects, risk propensity does explain a larger propensity to cheat. The remainder of this chapter is organized as follows: Section 2 presents the data gathered from the experiment, Section 3 presents the empirical analysis, while Section 4 concludes.