An experimental analysis of acquired impulse control among adult humans intolerant to

Jianxin Wanga, Yulei Raoa, and Daniel E. Houserb,c,1

aSchool of Business, Central South University, Changsha 410083, People’s Republic of China; bDepartment of Economics, George Mason University, Fairfax, VA 22030; and cInterdisciplinary Center for Economic Science, George Mason University, Fairfax, VA 22030

Edited by Roy Baumeister, Florida State University, Tallahassee, FL, and accepted by Editorial Board Member Michael S. Gazzaniga December 22, 2016 (received for review July 9, 2016) The ability to control tempting impulses impacts health, education, leads to systematic differences in the need to practice small acts and general socioeconomic outcomes among people at all ages. of self-control. This paper fills that gap. Consequently, whether and how impulse control develops in adult Our interest is to investigate a key prediction of the strength populations is a topic of enduring interest. Although past research model: repeatedly practicing small acts of self-control leaves a has shed important light on this question using controlled in- person more able to use willpower than another observationally tervention studies, here we take advantage of a natural experiment identical person without this practice. Many experimental tests of in China, where males but not females encounter substantial social this prediction have appeared (17, 18). A typical experiment assigns pressure to consume alcohol. One-third of our sample, all of whom people randomly to practice self-control (or not) for a relatively are Han Chinese, is intolerant to alcohol, whereas the remaining short period (usually days or weeks), and subsequently assesses control sample is observationally identical but alcohol tolerant. their ability to use self-control in a laboratory task. The meta- Consistent with previous literature, we find that intolerant males analysis of Hagger et al. (17) reveals that the strength model is are significantly more likely to exercise willpower to limit their generally supported by the literature, and in particular the ability to alcohol consumption than alcohol-tolerant males. In view of the practice self-control does seem to increase as a result of practice. strength model of self-control, we hypothesize that this enables [Others who have analyzed these data using different approaches

improved impulse control in other contexts as well. To investigate SCIENCES reach different conclusions (19).] Baumeister and Vohs (18) fur- ECONOMIC this hypothesis, we compare decisions in laboratory games of self- ther argue that, “exercising self-regulation of one or two sorts of control between the tolerant and intolerant groups. We find that behaviors causes discernible improvements on seemingly unrelated males intolerant to alcohol and who regularly encounter drinking tasks.” At the same time, as noted by Muraven (20), this literature environments control their selfish impulses significantly better than suffers from confounds that create difficulties in tracing the source their tolerant counterparts. On the other hand, we find that female of this effect to an increase in the effective stock of willpower. Han Chinese intolerant to alcohol do not use self-control to limit ’ alcohol consumption more than tolerant females, nor do the tolerant For example, one reason a person s self-control might improve and intolerant females exhibit differences in self-control behaviors. after an intervention is because they become more aware of the Our research indicates that impulse control can be developed in adult importance of self-control. In the same way that raising the sa- populations as a result of self-control behaviors in natural environ- lience of a norm tends to improve obedience to that norm (21), ments, and shows that this skill has generalizable benefits across emphasizing the importance of self-control by asking people to behavioral domains. practice self-control may leave them more likely to comply with tasks that require self-control. Muraven (20) suggests that another self-control | strength model | alcohol intolerance | cheating | confound is that a person, focused on and thus realizing they are natural experiment able to practice self-control, may gain additional confidence in this

umans are often tempted to pursue an immediate benefit, Significance Heven when they know it may result in substantial long-run cost. Controlling these tempting impulses requires willpower (1). Impulse We find that alcohol intolerance, which occurs randomly and control is important to a large number of social and economic en- leads affected people systematically to resist the impulse to vironments, including health behavior (2) and weight loss (through consume alcohol, enables improved impulse control in another both food control and exercise) (3), violence and crime (4), and behavioral domain. Our findings show that impulse control can develop in adult populations as a result of exercising routine spending and saving in household finances (5). A long multidisci- behaviors in naturally occurring environments. This finding is plinary empirical and theoretical literature discusses the role of important because the ability to control tempting impulses sig- impulse control in social and economic decision making (6, 7). nificantly impacts the health and well-being of people at all ages. The question of whether impulse control can be acquired or Our findings complement theory by providing rigorous empirical – improved has been long studied, both empirically (8 10) and the- evidence that simple lifestyle changes may have significant per- oretically. An important contribution to this literature is the so- sonal well-being benefits. In particular, small acts of self-control, called “strength model” of self-control (see, for example, refs. such as resisting drinking or always waking at the same time, can 11–13), which argues self-control is like a “muscle” that can be increase one’s overall ability to resist selfish temptations. depleted by short-term exertion, and yet that the same exertion can lead to it strengthening. [Although we focus on the strength model, Author contributions: J.W., Y.R., and D.E.H. designed research; J.W. performed research; J.W. and D.E.H. analyzed data; and J.W., Y.R., and D.E.H. wrote the paper. alternative models of self-control have been suggested (14, 15).] The authors declare no conflict of interest. The model predicts that practicing self-control on specific tasks ’ This article is a PNAS Direct Submission. R.B. is a Guest Editor invited by the Editorial increases one s ability to exercise self-control broadly across mul- Board. tiple domains (16). This prediction has been frequently tested using Freely available online through the PNAS open access option. designed experimental interventions (17, 18). To our knowledge, 1To whom correspondence should be addressed. Email: [email protected]. however, there have been no tests of the strength model based on This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10. “natural experiments,” where exogenous population variation 1073/pnas.1610902114/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1610902114 PNAS Early Edition | 1of6 Downloaded by guest on October 1, 2021 ability. This increased confidence may affect their ability to exer- display greater self-control over selfish cheating behavior than cise self-control in a subsequent task, particularly one in which their tolerant counterparts. Furthermore, females do not exhibit their performance is being monitored. differences between tolerant and intolerant groups in either We study impulse control among adult Han Chinese displaying propensity to use self-control in drinking environments or in intolerance to alcohol, in comparison with a control group of adult decisions to cheat. Han Chinese without this intolerance. We provide evidence that These results are consistent with the view that impulse control, male Han Chinese with alcohol intolerance practice small acts of like a muscle, can be developed in natural environments as a self-control over alcohol more regularly than males in our tolerant consequence of practicing small acts of self-control, and that this groups. We argue this is the only systematic difference in the ex- has generalizable benefits across multiple behavioral domains (16, ercise of self-control between the tolerant and intolerant groups. 18, 20). It is worth emphasizing that, whereas substantial experi- Consequently, the alcohol-intolerant male sample would be expec- mental research based on designed interventions has provided ted, by the strength model, to be better able to control tempting support for the strength model (e.g., refs. 13 and 37–39), to our impulses. Note that our approach has the advantage that confounds, knowledge ours is unique in discovering the positive spillover such as salience and confidence-building, are absent. effects associated with practicing self-control in a naturally Our sample (48.6% female) is drawn from the Han Chinese occurring context. undergraduate population at a leading Chinese university. Mem- bers of our sample are similar in age and background: homoge- Literature Review neity of the participant pool is the reason we use this population. Acquired Impulse Control. Impulse control can be acquired We are aware of no evidence that alcohol intolerance is associated through practice. Ainslie (8) used pigeons to study the possibility with personality or preferences, and in our sample we find no that animals can learn self-control, finding that some subjects evidence that this polymorphism systematically predicts person- could learn to control their behavior for a delayed reward. An ality type, attitudes toward risk, or prosociality. Other than alcohol early survey of learned self-control is found in Ainslie (9). intolerance and gender, both of which are expected to impact Among humans, self-control is usually developed during child- systematically use of self-control, we are aware of no other reason hood (4). There is evidence, however, indicating self-control can or evidence that practice of self-control would differ systematically also be strengthened among adolescents and adults. Muraven et al. among this population. (40) studied college students who spent 2 wk performing one of Alcohol intolerance arises because of the presence of the three self-control exercises (improving posture, regulating moods, ALDH2 Lys487 allele (see SI Appendix for details). Since there is an and maintaining a diary of eating). Compared with a no-exercise equal chance that either parental allele will be passed on to off- control group, these participants showed significant improvements spring, alcohol intolerance can be understood as occurring ran- in their self-control, as demonstrated by their ability to squeeze a domly in the population (22). This randomization across our sample handgrip for a longer time. The strength model of self-control of is crucial for the causal inferences we report below. Note that our Muraven and Baumeister (20) argues that self-control, like a research strategy connects to the Mendelian randomization litera- muscle, can be strengthened through practice. ture (23–26), which also takes advantage of natural experiments Substantial empirical support for the strength model has implied by random assignments of genes from parents to offspring. appeared in the literature. For example, Muraven (20) repor- It is important to note that exposure to drinking environments is ted data from 92 adult participants engaged in a “stop signal” task also randomized across our sample: we find no evidence that any to measure their self-control. The participants found that after variable we measure differs systematically among males who are 2 wk of practicing self-control by, for example, avoiding sweets more or less frequently exposed to environments that include al- or squeezing a handgrip, participants’ self-control performance cohol, and the frequency of exposure to drinking environments is improved significantly. Furthermore, Muraven (41) found that statistically identical between the tolerant and intolerant groups. practicing self-control helped one to quit smoking. Other ex- (We find that females measure higher on extroversion more fre- amples have been found in the practice of physical exercise quently in drinking environments, but no other trait differences (42), academic study (43), and financial monitoring (44). between those regularly and rarely in environments that include To our knowledge, all previous tests of the strength model have alcohol.) A key reason is that participants are randomly assigned been based on designed interventions. Our paper is unique in that to dorm rooms, and the frequency of exposure to alcohol has been our investigation takes advantage of naturally occurring variation shown to be dependent on peers and the social pressure they in the frequency with which people practice self-control. exercise (27–29). [Many have taken advantage of random dorm assignment to study peer effects. Early examples are Sacerdote Assessing Self-Control Using Laboratory Cheating Games. Self-control (30), Zimmerman (31), Stinebrickner and Stinebrickner (32), and in the laboratory is frequently studied through the use of cheating Boisjoly et al. (33). More recently, Jain and Kapoor (34) and Shue games. Cheating is a socially costly behavior and has been studied (35) used random dorm assignment to study peer effects on atti- heavily in economics. Early cheap-talk games explored humans’ tudes and behavior, as well as academic and career outcomes.] cheating behavior, where the subjects could choose cheating China does not have a , and college-aged strategies to increase their own payoff and decrease their oppo- Chinese males experience substantial social pressure to drink to nents’ payoff (45, 46). Later studies used self-reported payoffs, so promote business and social relationships. Despite its immediate that the experimenter could not detect individual participant’s relational benefits, previous research finds that alcohol-intolerant cheating behavior (47–49). males use self-control more than their alcohol-tolerant counter- Prior research has established a clear causal relationship be- parts to limit their drinking (36). In contrast, females face little tween willpower and cheating decisions. For example, Mead et al. social pressure to drink in China and, moreover, generally con- (50) studied the relation between self-control and cheating for sume very little alcohol. Consequently, we do not expect alcohol selfish gain. The authors found that participants with depleted intolerance to modulate Chinese females’ use of self-control in self-control misrepresented their performance for monetary gain drinking environments. to a greater extent than did nondepleted participants. The impli- We find alcohol-intolerant males in our sample are significantly cation is that cheating for profit increases when people’s capacity more likely to use self-control to limit their alcohol consumption to exert self-control is impaired. Gino et al. (51) found similarly in drinking environments than tolerant males. Moreover, consis- that depletion of self-control leads to heightened cheating because tent with the strength model, we find that males regularly exposed of a reduced attention to moral standards (see also ref. 47). In to drinking environments and who exhibit alcohol intolerance addition, Barnes et al. (52) found that people who consumed

2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1610902114 Wang et al. Downloaded by guest on October 1, 2021 Table 1. Payoffs in the self-control game ambiguity aversion. In all cases, we found no differences between Throws and payoff Amount tolerant and intolerant groups. The sequence of the four tasks was randomly determined in each session. Number thrown 123456Our goal with these initial decision tasks was to obtain evidence Resulting payoff (Yuan) 5 10 15 20 25 0 on whether attitudes toward risk or ambiguity, prosociality, or time-preference might systematically vary between tolerant and intolerant groups. If so, this evidence might suggest alternative self-control resources (in their case, by having too little sleep) were explanations for any differences discovered in cheating behavior. more likely to choose to cheat. As shown below, however, we found no such evidence and thus did not pursue this issue further. Hypotheses Detailed instructions for each of these tasks are provided in SI The above discussion motivates the following three related hy- Appendix. In brief, to elicit prosociality, participants made decisions potheses we tested using our Han Chinese sample. Detailed as Sender in a 20 Renminbi (RMB) dictator game (53, 54). Our explanation of these hypotheses is found in SI Appendix. risk preference elicitation was based on a 30 RMB risky investment game (55, 56). Time preferences were elicited by asking subjects to Hypothesis 1. Males with alcohol intolerance practice self-control in decide between two future payments: smaller sooner or larger later drinking environments more than tolerant males. Female self-con- (57). Ambiguity aversion was elicited using the task of Ellsberg trol over drinking alcohol is not modulated by alcohol intolerance. (58), which involves drawing a ball from an urn with either known or ambiguous color distribution. Hypothesis 2. Males intolerant to alcohol and regularly in drinking Following these decision tasks, we administered an patch environments will develop generalized improvement in self- test to detect alcohol intolerance (Materials and Methods)anddis- control in relation to otherwise identical tolerant males and in- tributed a questionnaire regarding alcohol consumption and toler- tolerant males not regularly exposed to drinking environments. In ance, as well as other allergies and gambling behaviors. Participants particular, in relation to alcohol-tolerant males and -intolerant also completed a demographic and personality survey. The survey males not regularly exposed to drinking environments, intolerant included questions about age, gender, and grade. The survey also males regularly in drinking environments will demonstrate an included several personality scales: Big Five (59), Machiavelli (60), SCIENCES improved ability to use willpower to resist selfish temptations. Self-Monitoring (61), Rosenberg’s Self-Esteem (62), and Consid- ECONOMIC eration of Future Consequences (63) (see SI Appendix for details). Hypothesis 3. Female self-control will be modulated by neither alcohol intolerance nor exposure to drinking environments. Results Experiment Design Descriptive Statistics. We obtained data from 480 participants. We excluded three participants from the analysis: one participant did not We measure self-control using a simple game where participants attend the patch test and two others were discovered after the ex- must control a selfish impulse to cheat for monetary gain. [Mead periment to be graduate students outside of our targeted recruitment et al. (50) and Gino et al. (51), among others, report data showing group. Our final sample for analysis included 477 observations: 152 that people with low self-control are more likely to misreport their in the intolerant group and 325 in the tolerant group. We obtained performance to achieve monetary gain. Further discussion of the data from 232 female and 245 male subjects, with numbers of al- relation between willpower and behavior in cheating games is cohol-tolerant and -intolerant subjects detailed in Table 2. found in Assessing Self-Control Using Laboratory Cheating Games, As is clear from SI Appendix,TableS1A, our tolerant and in- above.] We used this game because it is easy for participants to tolerant groups did not systematically vary in the dimensions we understand, rapidly conducted, and a reliable as a measure of self- measured. For males, among the 15 demographic, preference, and control (50, 51). personality measures taken, only one is marginally statistically sig- Following Fischbacher and Heusi (49), each participant was nificantly different in bivariate comparison (SI Appendix,TableS1A: individually given a six-sided die. Participants were told that they “ ” t = P = could roll the die as many times as they wished to verify it worked Altruism , 1.667, 0.096). The magnitude of the difference properly. However, only the first roll, which they were told to re- is, however, quite small, and that one measure is marginally signif- member during the rolling process, would decide their payoff. All icant is not surprising in view of the number of comparisons made. die rolls were private: only the participant knew the outcome of With respect to females, preference and attitude measures do their first die-roll. The payoff structure is described in Table 1, and not significantly differ between alcohol-tolerant and -intolerant groups. Extraversion is marginally significantly higher among tol- all participants were aware of this payoff structure when reporting = = to the experimenter the outcome of their first die-roll. erant females (t 1.944, P 0.053), although again the magnitude ’ of the difference is small. All of the other measures, including As in the literature cited above, because participants die-rolls ’ were private, we are not able to detect cheating at the individual Machiavelli, Self-Monitoring, Rosenberg s Self-Esteem, and Con- level. We draw inferences regarding cheating from the overall sideration of Future Consequence are statistically identical be- outcome distribution. If nobody cheats, each number should be tween female tolerant and intolerant groups. reported at a rate that does not statistically differ from a one-in-six SI Appendix, Table S1B provides details on the same variables frequency. If people systematically cheat to gain a higher payoff, as SI Appendix, Table S1A, but broken down further by gender then we would expect outcomes three, four, or five to be reported and level of exposure to environments that include alcohol (in “ ” “ ” at statistically greater than a one-in-six frequency. the table, Regular is defined as at least monthly and Rare We began each session by conducting four decision tasks. The means less frequently than monthly). It is clear from inspection first three were: the self-control (cheating) game, a risk-attitude elicitation, and an elicitation for prosociality. The fourth task was Table 2. Sample by alcohol intolerance and gender either a time-preference elicitation or a task that measured ambi- guity aversion. The length of our experiment precluded running all Tolerance Female Male Total preference elicitations in all sessions. We decided risk-attitudes and Intolerant 59 93 152 prosociality were important to be able to measure as accurately as Tolerant 173 152 325 possible, and so included them in all sessions. We measured time Total 232 245 477 preference in most sessions (nearly 90%) and otherwise measured

Wang et al. PNAS Early Edition | 3of6 Downloaded by guest on October 1, 2021 of SI Appendix, Table S1B that few P values for bivariate com- evidence is consistent with the view that female cheating behavior parisons fall below 0.05, and none of the differences are large in is not modulated by intolerance to alcohol. magnitude. Overall, with the exception that females with regular exposure to drinking environments measure as more extraverted, Self-Control by Exposure to Drinking Environments. We turn now to there is no evidence of systematic variation within males and hypotheses 2 and 3. Subjects were asked: “How often are you females according to their exposure to environments that include around people drinking alcohol (regardless whether you chose to drinking.* drink)?” The possible answers were “every day,”“weekly, but not These findings are evidence that our participants were ran- daily,”“monthly, but not weekly,”“occasionally, but less often domized effectively into alcohol exposure. As noted above, a key than monthly,” or “never.” We obtained a 92.4% response rate source of this randomization is peer effects stemming from ran- to this question. dom assignment to dorm rooms (see, for example, refs. 27 and 28). SI Appendix, Fig. S3 describes the frequency of high payoff re- ports by frequency of exposure to drinking environments, where the Test of Hypothesis 1: Drinking Behavior by Gender and Intolerance. frequencyofeverymonthorhigherisconsidered“regular expo- We use self-reported answers to survey questions to test the hy- sure”.(Weused“monthly” as the cutoff for “regular exposure” to pothesis that alcohol-intolerant males use self-control in drinking drinking environments. This is both intuitively reasonable and also environments more than tolerant males or females (regardless of provides a roughly balanced sample. The distribution of exposure to tolerance). People responded to the following question “Generally drinking environments, by gender and alcohol intolerance is de- speaking, when I am out and after I have had one ” tailed in SI Appendix,TableS3.) It is clear from SI Appendix,Fig.S3 the following three ways: (i) I give little thought to the pace at that females report higher payoff numbers about two-thirds of the which I’m drinking (or not drinking) alcohol; (ii) I use self-control time, insignificantly differently across all conditions. Similarly, males to reduce the pace at which I drink alcohol; (iii) I become more report high payoff outcomes about four-fifths of the time, in- eager to drink alcohol and drink faster. significantly differently across all conditions, with the sole-exception Responses are detailed in SI Appendix,Fig.S1andTableS2. of alcohol-intolerant males with regular exposure to drinking envi- Approximately four of five of participants responded to this ques- ronments. These males reported high payoff numbers 61.8% of the tion, with responses missing in roughly equal proportion by gender time: lowest among all male groups and 24.8 percentage points and intolerance. Our interest is in the frequency with which people lower than tolerant males exposed to alcohol rarely. This difference responded that they used self-control in drinking environments. We is significant (P = 0.005, two-sided t test). In addition, among males found 68.3% of tolerant males and 70.5% of females indicated that with regular exposure to alcohol, those who are alcohol-tolerant they use self-control, and 73.7% of intolerant females indicated report higher payoff numbers marginally significantly more than the that they use self-control. In contrast, 81.1% of intolerant males intolerant group (P = 0.079, two-sided t test). The rate of high indicated that they use self-control. The effect size is large and the payoff reports among intolerant males with regular exposure is difference between tolerant and intolerant males is (marginally) 14 percentage points lower than those with rare exposure (76%). statistically significant (P = 0.057, two-sided Mann–Whitney test). Fig. 1 describes the results of a direct test of the three-way in- Whether a female is intolerant does not impact her propensity to teraction hypothesis implied by the strength model: intolerant use self-control. These findings support our hypothesis 1. males regularly exposed to alcohol cheat less than tolerant males or intolerant males with infrequent alcohol exposure. The former Self-Control by Gender and Intolerance. SI Appendix, Fig. S2 dis- group reports higher payoff outcomes with frequency 61.8%, plays the frequency of high payoff outcomes reported by gender whereas the pooled sample of other males report high payoff and tolerance. In the tolerance group, 79.6% of males reported the outcomes with frequency 78.6%, and this difference is significant ‡ number 3, 4, or 5, corresponding to the higher payoff; only 60.7% of (t = 2.496, P = 0.013, two-sided t test). The analogous com- females reported the number of 3, 4, or 5. Thus, tolerant males parison for females does not find significant differences (t = 0.565, † cheated significantly more than females. This difference is consis- P = 0.573).§ tent with Houser et al. (48), who report decisions from European A logit regression analysis of cheating behavior is detailed in SI players of a similar cheating game showing that males cheat sig- Appendix,TableS4. The dependent variable equals 1 if a high nificantly more than females [alcohol intolerance occurs with very payoff outcome was reported, and 0 otherwise. The first column of low frequency in European populations (64)]. In contrast, the rate SI Appendix,TableS4details the explanatory variables, with at which intolerants reported a 3, 4, or 5 was insignificantly different “gender” taking value 1 if the participant is male, “regular” taking between males and females at 66.7% and 64.4%, respectively. value 1 if the participant reported regular exposure to alcohol, Comparing within gender, we see intolerant males report 3, 4, and the other variables as described in SI Appendix,TableS1. and 5 significantly less often than tolerant males (P = 0.023, two- The reported coefficients are marginal effects, and the bold sided t test). This evidence supports a link between alcohol in- numbers are each coefficient’s P value against the null that it is tolerance and an improved ability to exercise self-control among equal to zero. males. Females intolerant to alcohol, however, do not report The second column of SI Appendix, Table S4 reports a logit higher payoff outcomes less often. Indeed, as is clear from SI regression of main effects. The baseline is alcohol-tolerant women Appendix, Fig. S2, they reported high payoff outcomes more fre- rarely exposed to environments that include alcohol. As already quently on average, although the difference is not significant. This seen in SI Appendix,Fig.S3, this group reports the high payoff

‡ *One can compute a Bonferroni correction for P values assuming 15 comparisons, showing Although our comparison is implied by theory, one can perform the data-mining exercise that to achieve significance at the 5% level, one requires an uncorrected P < 0.004. and compute this same statistic for all four comparisons created by holding out one Accordingly, we find no significant differences among males, and females differ only group, and comparing to the pooled others. When doing so, only the hypothesized in that those with regular exposure to drinking environments measure higher comparison is significant (P < 0.05), whereas the others all have P values in excess of in extraversion. 0.15 (all Bonferroni corrected for four comparisons). †Our analysis is based on aggregating the three higher (high payoff) and three lower (low §One can posit a testable “difference-in-differences” extension of the three-way interac- payoff) numbers. The reason is that our design does not allow us to detect how people tion hypothesis. In particular, alcohol-intolerant males would be expected to show cheat at the individual level. Aggregating provides a statistically powerful approach to greater improvement in self-control because of regular exposure to drinking environ- detect cheating while remaining agnostic about how cheating occurs. SI Appendix, Ta- ments than would alcohol-tolerant males. We denote this hypothesis 2a, and find evi- bles S6 and S7 detail results of analyses at different levels of aggregation. The results are dence supporting it using a slope test (65, 66) (see SI Appendix, Table S8 and related qualitatively unchanged. discussion).

4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1610902114 Wang et al. Downloaded by guest on October 1, 2021 self-control. We find that alcohol-intolerant males who regularly stropeRffoyaPhgiHfoycneuqerF 1.0 Others encounter drinking environments are significantly better able to 0.9 Intolerant with Regular Exposure exercise self-control in cheating games than alcohol-tolerant males 0.8 or -intolerant males less frequently exposed to drinking envi- ronments. The reason is that alcohol-intolerant males exercise 0.7 more self-control than tolerant males to resist the impulse to 0.6 drink, and this regular practice improves intolerant males’ ability to control their selfish impulses. Furthermore, we find that 0.5 females behave statistically identically in cheating games regard- 0.4 less of their exposure to drinking environments and regardless of whether they are intolerant to alcohol. We argued the reason is 0.3 that alcohol-intolerant females do not exercise self-control 0.2 in drinking environments differently than do alcohol-tolerant females, and thus this result is also consistent with the strength 0.1 model. 0.0 Although we did not observe differences in preferences or N=168 N=55 N=202 N=16 personality between tolerant and intolerant males, this might be in part because of the narrow range of ages in our sample. Sys- tematic differences in the practice of self-control could lead to Fig. 1. Frequency of high payoff outcomes (numbers 3, 4, and 5) reported by such differences emerging over time. It would be profitable for intolerant males regularly exposed to drinking environments and others. Fre- future research to investigate whether, when, and which trait quency of high payoff number reported are significantly different between in- differences emerge, and how these impact social and economic tolerant males with regular exposure to alcohol and all of the other males = = outcomes. (t 2.496, P 0.013, two-sided t test) but do not differ significantly among Self-control is an important component of social and economic females (t = 0.565, P = 0.573). Error bars are 95% confidence intervals. environments. Our research is unique in that it shows that an adult’s ability to manage selfish impulses improves as a conse- SCIENCES ECONOMIC outcome about two-thirds of the time. Overall, men are found to quence of using self-control in naturally occurring environments. cheat significantly more often than women (an overall difference An implication is that small and simple day-to-day lifestyle of 14.9 percentage points). Although “Intolerant” and “Regular” changes can have significant personal health and well-being ben- both have the expected signs, neither is statistically significant. efits. In particular, in view of our findings, one might speculate The strength model of self-control predicts that alcohol- that regular acts of self-control, such as resisting drinking or per- intolerant males with regular exposure to drinking environments haps routinely exercising or waking at the same time each day, will cheat significantly less than otherwise identical males. To test might improve one’s overall ability to resist selfish temptations. this, column four of SI Appendix,TableS4reports an analysis that includes this three-way interaction as well as gender. The effect Materials and Methods size of the interaction is large and statistically significant, lending Human Subjects Protections. This study was reviewed and approved by George { direct support to the theory. Mason University’s Human Subjects Internal Review Board, protocol no. The fifth and sixth column of SI Appendix,TableS4reveals that, 575329–1. Informed consent was obtained from participants at the begin- after controlling for preferences and personality traits, alcohol- ning of each session. intolerant males with regular exposure to drinking environments are 20 percentage points less likely to report high payoff outcomes Detecting Alcohol Intolerance. Following the die-roll game, we tested each than other males in our sample, and this effect is statistically sig- subject for alcohol intolerance using a procedure detailed by Muramatsu et al. (67). Specifically, 0.1 mL of 70% (by volume) ethanol was pipetted onto nificant at the 5% level. Moreover, these males cheat at rates a15× 15-mm lint pad affixed to an adhesive tape. The same volume of statistically identical to females (the sum of the coefficients of “ ” distilled water was pipetted onto an identical pad and used as a control. the three-way interaction and Gender is near zero, and sta- Both patches were attached to the inner surface of each participant’s upper tistically insignificant with P = 0.584). Finally, the fifth and sixth arm for 7 min and then removed. A patch area that was judged to show columns of SI Appendix,TableS4reveals that people who are erythema 10–15 min after removal to a greater degree than the distilled more altruistic, as well as those who are more likely to consider water baseline indicated intolerance to alcohol (68). the future consequences of their decisions, are both significantly less likely to cheat. Note that these results are robust to the Procedures. Experiments were conducted in the laboratory of the Business inclusion of other variables from our survey. For example, the School in Central South University. We initially collected 226 observations results are unchanged by including controls for home province followed by a second set of sessions resulting in 254 additional observa- or self-reported drinking behavior in comparison with peers tions. We conducted 22 sessions in total. Recruiting was done by posting (“Peer Effect”). flyers around campus. Subjects were undergraduate students at Central South University, all of whom were from the Han ethnic group and at least Discussion 18 y old. We paid the participants with cash, in private, and immediately after the experiments (with the exception of the delayed payoffs from the We studied self-control among a population intolerant to alcohol, time preferences task, which was paid on the corresponding payment date). in comparison with an observationally identical but alcohol- Each participant received 5 Yuan for participating, and the average payoff tolerant control group. Our data support the strength model of was 82 RMB. The four decision tasks were implemented using paper and pencil. In- structions for each task were put into envelopes and distributed. Participants { The model described by column 4 of SI Appendix, Table S4 is supported by the theory as wrote their experiment ID on the instructions. Participants returned the well as a data-based stepwise regression procedure (SI Appendix, Table S5). SI Appendix, instructions to the envelope after decisions were made, and the experimenter Table S5 shows that the two-way interactions can be significant when entered individ- collected the envelopes. Each session lasted about 2 h. ually. The reason is their high correlation, as high as 0.86, with the three-way interaction. SI Appendix, Table S5 also shows gender and the three-way interaction are chosen for After the experiment participants completed a short questionnaire about the model first, and conditional on their inclusion no two-way interaction contributes behaviors related to alcohol. If the survey was not completed while in the significantly to the model. laboratory, we followed-up via text-message.

Wang et al. PNAS Early Edition | 5of6 Downloaded by guest on October 1, 2021 ACKNOWLEDGMENTS. We thank the editor and two anonymous reviewers research was provided by the National Natural Science Foundation of China for many insightful comments that improved this paper. Funding for this (Grants 71501193, 71372063, and 71631008).

1. Green L, Fry AF, Myerson J (1994) Discounting of delayed rewards: A life-span com- 34. Jain T, Kapoor M (2015) The impact of study groups and roommates on academic parison. Psychol Sci 5(1):33–36. performance. Rev Econ Stat 97(1):44–54. 2. Allom V, Mullan B, Hagger M (2016) Does inhibitory control training improve health 35. Shue K (2013) Executive networks and firm policies: Evidence from the random as- behaviour? A meta-analysis. Health Psychol Rev 10(2):168–186. signment of MBA peers. Rev Financ Stud 26(6):1401–1442. 3. Junger M, van Kampen M (2010) Cognitive ability and self-control in relation to di- 36. Hendershot CS, et al. (2011) Evaluating a cognitive model of ALDH2 and drinking etary habits, physical activity and bodyweight in adolescents. Int J Behav Nutr Phys behavior. Alcohol Clin Exp Res 35(1):91–98. Act 7(6):22. 37. Bucciol A, Houser D, Piovesan M (2010) Willpower in children and adults: A survey of 4. Gottfredson MR, Hirschi T (1990) A General Theory of Crime (Stanford Univ Press, results and economic implications. Int Rev Econ 57(3):259–267. Stanford, CA). 38. Bucciol A, Houser D, Piovesan M (2011) Temptation and productivity: A field exper- 5. Gathergood J, Weber J (2014) Self-control, financial literacy & the co-holding puzzle. iment with children. J Econ Behav Organ 78(1):126–136. J Econ Behav Organ 107(Part B):455–469. 39. Bucciol A, Houser D, Piovesan M (2013) Temptation at work. PLoS One 8(1):e53713. 6. Fudenberg D, Levine DK (2006) A dual-self model of impulse control. Am Econ Rev 40. Muraven M, Baumeister RF, Tice DM (1999) Longitudinal improvement of self-regu- 96(5):1449–1476. lation through practice: Building self-control strength through repeated exercise. 7. Fudenberg D, Levine DK (2012) Timing and self-control. Econometrica 80(1):1–42. J Soc Psychol 139(4):446–457. 8. Ainslie GW (1974) Impulse control in pigeons. J Exp Anal Behav 21(3):485–489. 41. Muraven M (2010b) Practicing self-control lowers the risk of smoking lapse. Psychol 9. Ainslie G (1975) Specious reward: A behavioral theory of impulsiveness and impulse Addict Behav 24(3):446–452. control. Psychol Bull 82(4):463–496. 42. Oaten M, Cheng K (2006a) Longitudinal gains in self-regulation from regular physical 10. Bechara A (2005) Decision making, impulse control and loss of willpower to resist exercise. Br J Health Psychol 11(Pt 4):717–733. drugs: A neurocognitive perspective. Nat Neurosci 8(11):1458–1463. 43. Oaten M, Cheng K (2006b) Improved self-control: The benefits of a regular program – 11. Baumeister RF, Heatherton TF (1996) Self-regulation failure: An overview. Psychol Inq of academic study. Basic Appl Soc Psych 28(1):1 16. 7(1):1–15. 44. Oaten M, Cheng K (2007) Improvements in self-control from financial monitoring. – 12. Baumeister RF, Bratslavsky E, Muraven M, Tice DM (1998) Ego depletion: Is the active J Econ Psychol 28(4):487 501. – self a limited resource? J Pers Soc Psychol 74(5):1252–1265. 45. Gneezy U (2005) Deception: The role of consequences. Am Econ Rev 95(1):384 394. 13. Vohs KD, Heatherton TF (2000) Self-regulatory failure: A resource-depletion ap- 46. Charness G, Dufwenberg M (2006) Promises and partnership. Econometrica 74(6): – proach. Psychol Sci 11(3):249–254. 1579 1601. 14. Inzlicht M, Schmeichel BJ (2012) What is ego depletion? Toward a mechanistic re- 47. Mazar N, Amir O, Ariely D (2008) The dishonesty of honest people: A theory of self- – vision of the resource model of self-control. Perspect Psychol Sci 7(5):450–463. concept maintenance. J Mark Res 45(6):633 644. 15. Fishbach A, Converse BA (2010) Identifying and battling temptation. Handbook of 48. Houser D, Vetter S, Winter J (2012) Fairness and cheating. Eur Econ Rev 56(8): – Self-Regulation: Research, Theory and Applications, 2nd Ed, eds Vohs KD, Baumeister 1645 1655. 49. Fischbacher U, Föllmi-Heusi F (2013) Lies in disguise—An experimental study on RF (Guilfird, New York), pp 244–260. cheating. J Eur Econ Assoc 11(3):525–547. 16. Muraven M, Baumeister RF (2000) Self-regulation and depletion of limited resources: 50. Mead NL, Baumeister RF, Gino F, Schweitzer ME, Ariely D (2009) Too tired to tell the Does self-control resemble a muscle? Psychol Bull 126(2):247–259. truth: Self-control resource depletion and dishonesty. J Exp Soc Psychol 45(3):594–597. 17. Hagger MS, Wood C, Stiff C, Chatzisarantis NL (2010) Ego depletion and the strength 51. Gino F, Schweitzer ME, Mead NL, Ariely D (2011) Unable to resist temptation: How model of self-control: A meta-analysis. Psychol Bull 136(4):495–525. self-control depletion promotes unethical behavior. Organ Behav Hum Decis Process 18. Baumeister RF, Vohs KD (2016) Strength model of self-regulation as limited resource: 115(2):191–203. Assessment, controversies, update. Adv Exp Soc Psychol 54:67–127. 52. Barnes CM, Schaubroeck J, Huth M, Ghumman S (2011) Lack of sleep and unethical 19. Carter EC, McCullough ME (2014) Publication bias and the limited strength model of conduct. Organ Behav Hum Decis Process 115(2):169–180. self-control: Has the evidence for ego depletion been overestimated? Front Psychol 53. Kahneman D, Knetsch JL, Thaler RH (1986) Fairness and the assumptions of eco- 5(823):823. nomics. J Bus 59(4):S285–S300. 20. Muraven M (2010) Building self-control strength: Practicing self-control leads to im- 54. Forsythe R, Horowitz JL, Savin NE, Sefton M (1994) Fairness in simple bargaining proved self-control performance. J Exp Soc Psychol 46(2):465–468. experiments. Games Econ Behav 6(3):347–369. 21. Cialdini RB, Reno RR, Kallgren CA (1990) A focus theory of normative conduct: Re- 55. Gneezy U, Leonard KL, List JA (2009) Gender differences in competition: Evidence cycling the concept of norms to reduce littering in public places. J Pers Soc Psychol from a matrilineal and a patriarchal society. Econometrica 77(5):1637–1664. 58(6):1015–1026. 56. Cameron L, Erkal N, Gangadharan L, Meng X (2013) Little emperors: Behavioral im- 22. Irons DE, McGue M, Iacono WG, Oetting WS (2007) Mendelian randomization: A pacts of China’s one-child policy. Science 339(6122):953–957. novel test of the gateway hypothesis and models of gene-environment interplay. Dev 57. Andreoni J, Kuhn MA, Sprenger C (2015) Measuring time preferences: A comparison – Psychopathol 19(4):1181 1195. of experimental methods. J Econ Behav Organ 116:451–464. ‘ ’ 23. Smith GD, Ebrahim S (2003) Mendelian randomization : Can genetic epidemiology 58. Ellsberg D (1961) Risk, ambiguity, and the Savage axioms. Q J Econ 75(4):643–669. contribute to understanding environmental determinants of disease? Int J Epidemiol 59. Costa PT, MacCrae RR (1992) Revised NEO Personality Inventory (NEO PI-R) and NEO – 32(1):1 22. Five-Factor Inventory (NEO FFI): Professional Manual (Psychological Assessment Re- 24. Davey Smith G, Ebrahim S (2005) What can Mendelian randomisation tell us about sources, Odessa, FL). – modifiable behavioural and environmental exposures? BMJ 330(7499):1076 1079. 60. Dahling JJ, Whitaker BG, Levy PE (2008) The development and validation of a new 25. Davey Smith G (2007) Capitalizing on Mendelian randomization to assess the effects Machiavellianism scale. J Manage 35(2):219–257. – of treatments. J R Soc Med 100(9):432 435. 61. Briggs SR, Cheek JM, Buss AH (1980) An analysis of the Self-Monitoring Scale. J Pers 26. von Hinke S, Davey Smith G, Lawlor DA, Propper C, Windmeijer F (2016) Genetic Soc Psychol 38(4):679–686. – markers as instrumental variables. J Health Econ 45:131 148. 62. Rosenberg M (1965) Society and the Adolescent Self-Image (Princeton Univ Press, 27. Baer JS (1994) Effects of college residence on perceived norms for alcohol con- Princeton, NJ). sumption: An examination of the first year in college. Psychol Addict Behav 8(1): 63. Strathman A, Gleicher F, Boninger DS, Edwards CS (1994) The consideration of future 43–50. consequences: Weighing immediate and distant outcomes of behavior. J Pers Soc 28. Borsari B, Carey KB (2001) Peer influences on college drinking: A review of the re- Psychol 66(4):742–752. search. J Subst Abuse 13(4):391–424. 64. Goedde HW, et al. (1992) Distribution of ADH2 and ALDH2 genotypes in different 29. Kremer M, Levy D (2008) Peer effects and alcohol use among college students. J Econ populations. Hum Genet 88(3):344–346. Perspect 22(3):189–206. 65. Dawson JF (2014) Moderation in management research: What, why, when, and how. 30. Sacerdote B (2001) Peer effects with random assignment: Results for Dartmouth J Bus Psychol 29(1):1–19. roommates. Q J Econ 116(2):681–704. 66. Dawson JF, Richter AW (2006) Probing three-way interactions in moderated multiple 31. Zimmerman DJ (2003) Peer effects in academic outcomes: Evidence from a natural regression: Development and application of a slope difference test. J Appl Psychol experiment. Rev Econ Stat 85(1):9–23. 91(4):917–926. 32. Stinebrickner R, Stinebrickner TR (2006) What can be learned about peer effects using 67. Muramatsu T, et al. (1989) Ethanol patch test—A simple and sensitive method for college roommates? Evidence from new survey data and students from disadvan- identifying ALDH phenotype. Alcohol Clin Exp Res 13(2):229–231. taged backgrounds. J Public Econ 90(8):1435–1454. 68. Brooks PJ, Enoch MA, Goldman D, Li TK, Yokoyama A (2009) The alcohol flushing 33. Boisjoly J, et al. (2006) Empathy or antipathy? The impact of diversity. Am Econ Rev response: An unrecognized risk factor for esophageal cancer from alcohol 96(5):1890–1905. consumption. PLoS Med 6(3):e50.

6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1610902114 Wang et al. Downloaded by guest on October 1, 2021