Is it lie aversion, risk-aversion or tax audit aversion? Modeling deception under risk and no risk Tei Laine (tlaine@msh-alpes.fr) Maison des Sciences de l’Homme-Alpes, Université Pierre Mendès-France 1221 avenue centrale, Domaine universitaire, St Martin d’Hères, France Tomi Silander (tomi.silander@xrce.xerox.com) Xerox Research Centre Europe, 6 chemin de Maupertuis 38240 Meylan, France Kayo Sakamoto (sakamotok@ihpc.a-star.edu.sg) Ilya Farber (farberi@ihpc.a-star.edu.sg) Institute of High Performance Computing, A*STAR, 1 Fusionopolis Way, #16-16 Connexis North, Singapore 138632 Abstract We studied deceptive decision making in hypothetical scenar- ios that involved risk of being caught of deceiving, or a penalty after being caught of deceiving, or both. We found that the de- ception rate was the lowest in the scenarios involving both the risk and the penalty. Our hierarchical model for deception sug- gests that in balancing the possible benefits from deception, the personal discomfort of getting caught is as large or larger than the inherent aversion to deception. Keywords: Decision making; risk attitudes; deception; incen- tives; MTurk. Introduction In our recent study that asked participants’ choices between risky and certain options (in the context of tax return), which either involved deception (deception condition) or did not in- volve deception (gamble condition), we observed a high rate of deception aversion 1 in the condition in which deception re- sulted in the better outcome than being honest, but involved both risk, i.e., non-zero probability of detection, and a poten- tial loss in the form of a tax penalty, if the deception was de- tected (Laine, Sakamoto, & Silander, 2013). The participants who were particularly deception averse in the deception con- dition were also more risk averse than others in the gamble condition, which had equivalent risky and certain outcomes, but involved no deception. 2 In other words, the participants who took more risk in gamble condition, also deceived more in the deception condition. We speculated if the reason for such a high level of tax compliance in the risky deception condition was exception- ally high level of risk aversion or exceptionally high level of deception aversion, or alternatively the task domain com- bined with the participant pool characteristics. We used Ama- zon MTurk workers from the US. This is a group of individ- uals who are willing to do simple tasks for little monetary compensation. Alternatively, based on their own prior expe- riences or knowledge of others’ encounters with the Internal 1 In this experiment 279 (42%) out of 672 participants did not choose the risky deceptive option a single time in the deception con- dition. 2 Only 25 out 672 participants never chose the risky option in the gamble condition. Revenue Service (IRS), our participants (most of them US tax payers) may have wanted to avoid any friction (even hy- pothetical) with the tax authorities, and indicated their will- ingness to pay due taxes, even in the presence of substantial financial incentives for evasion. To rule out the explanation pertaining to the task domain and the participant pool, we conducted another study with MTurk participants, and added conditions from which we ex- cluded either the risk or the tax penalty. Again we observed an exceptionally high rate of deception aversion in the con- dition that involved both a non-zero probability of detection and and a penalty after the deception was detected. Thus, it seems that the “IRS aversion”, in other words the aver- sion to a potential audit by the tax authorities, is not alone enough to explain the high level of tax compliance, since the participants demonstrated some willingness to evade taxes in conditions from which either the risk or the penalty for de- tected deception was absent. In this study we wanted to find out what distinguishes those participants who refused to de- ceive in their taxes no matter what from those who properly incentivized switched from complete tax compliance to some degree of tax evasion. Who are those who do X, where X ∈ {take risk, deceive, evade taxes} In general, people tend to be risk averse when facing gains and risk seeking when facing losses (Holt & Laury, 2002; Kahneman & Tversky, 1979). Many economic models of choice behavior are based on the concept of individual risk attitude, which can be measured experimentally and mod- elled with the shape and parameters of a utility function (Holt & Laury, 2002; Isaac & James, 2000; Weber, 1998). It has been considered a stable construct similar to personal- ity traits, which drives behavioral patterns across situations (Blais & Weber, 2006). However, this interpretation is problematic, since several studies have found that the risk attitude varies across task types (e.g., hypothetical vs. real outcomes) (Holt & Laury, 2002; Taylor, 2013), domains (e.g., financial or health related decision making), elicitation methods (e.g., choice between 204