A Culture of Cheating: The Role of Worldviews in Preferences for Honesty

We explore the interaction of cultural worldviews and cheating behavior using a coin-flipping task in an online experiment. Two treatments are conducted, one in which cheating has only private benefits and one in which cheating benefits the public. While we find no differences in behavior across treatments, we find significant differences in dishonest behavior between genders which are largely explained by cultural worldviews.


Introduction
A large literature within economics and social psychology over the past few decades has provided mounting evidence that preferences for honesty vary within the population.
Ongoing research seeks to identify which characteristics affect individual preferences for honesty, with an emerging line of work considering the role of culture.
When considering cultural differences, researchers often compare behavior across geographical locations, using location as a proxy for culture. Results from studies that compare across locations have found inconsistent evidence for differences. While some find that honesty levels differ (e.g., Cohn et al., 2019;Dieckmann et al., 2016), others fail to detect any significant difference (e.g., Mann et al., 2016;. In contrast to geographical location as a proxy, Ariely et al. (2019) uses family background from East Germany as a proxy for exposure for socialism and finds that people brought up under socialism are less honest.
This study provides new evidence by measuring culture at the individual level rather than using geographical location or family background as a proxy for culture. Cultural worldviews, defined as a socially constructed orientation that dictates how one interprets and interacts with society, has previously shown to be a strong predictor of individual preferences on various social issues such as gun control, climate change and free-riding Cherry et al., 2017a). The cultural cognition literature postulates that adherents of hierarchical and individualistic worldviews are less tolerant towards social deviance (Kahan, 2008), which suggests that these individuals will have higher preferences for honesty.
We extend this line of inquiry to consider dishonesty when it only benefits the actor (selfish) and when it benefits the actor and others (Pareto). From the literature, people appear to be more dishonest when the benefits are shared with others (e.g., Wiltermuth, 2011;Ploner and Regner, 2013). Given that people with hierarchical and individualistic worldviews tend to be more self-serving (Cherry et al., 2017b), we consider whether any difference between selfish and Pareto dishonesty is worldview-specific.
This study draws from two literatures to combine a variant of a coin-flip task commonly used in the dishonesty literature (e.g. Bucciol and Piovesan, 2011) and an established instrument used to measure individual cultural worldviews (e.g., Kahan et al., 2011). This allows us to investigate the possible interplay between cultural worldviews and preferences for honesty.

Experimental Design and Hypotheses
This paper utilizes a variation of the commonly used coin-flip task from the literature, where subjects are asked to report the outcome of ten coin-flips each paying $0.25 per "heads". 1 Actual outcomes were not observed, which allowed subjects to over-report with impunity. 2 We employ a between-subject design that randomly assigned subjects to one of two decision environment treatments. A private-return treatment follows the literature by paying subjects $0.25 for each self-reported head. A public-return treatment directs subject earnings to a group account akin to a public good game. Specifically, subjects are randomly assigned to groups of four and each subject's earnings from self-reported heads ($0.25 per head) is placed in a public good account. The sum of the group members' earnings is multiplied by 1.6 and shared equally among the four members. Therefore, the marginal per capital return is 0.40. 3 Following the coin-flip task, subjects completed a set of worldview questions drawn from the cultural cognition literature (Kahan et al., 2008). Following the literature (e.g., Cherry et al., 2017a), responses to the questions were used to organize an individual's worldviews along two dimensions: hierarchy-egalitarianism and individualism-communitarian. For each dimension, worldview scores range from 4 to 20, with a higher score indicating a more hierarchical (less egalitarian) worldview and a more individualistic (less communitarian) worldview. Figure 1 presents a scatterplot of the two worldview measures. 4 We assign subjects to worldview categories based on their scores with participants scoring above the median defined as hierarchists and individualists and those scoring at or below the median defined as egalitarians and communitarians. Subjects concluded by providing basic nonidentifiable characteristics, which we use to explore possible heterogeneous treatment effects.

Figure 1. Scatterplot of Hierarchy and Individualism Worldview Scores
Note: plots are slightly jittered to better illustrate the distribution of data We recruited 702 subjects through Amazon's online workforce, Mechanical Turk (MTurk). We targeted experienced MTurk participants residing in the United States that had over 95% approval rankings. We employed "rIP" (Kennedy et al., 2020) to remove duplicate and suspicious addresses, which left 590 observations for the analysis. 5 Including a $0.75 participation payment, the average hourly pay rate equaled $10.50.
We consider three research questions. First, we investigate the potential influence of cultural worldviews on preferences for honesty by testing the null that reported heads is equivalent across worldviews. From the literature (Kahan, 2008), we expect to find that individuals holding hierarchical and individualistic worldviews will self-report fewer heads.
Second, we investigate possible differences in self-reporting behavior across the private and public benefit treatments by testing the null across the two decision environments. Previous work shows that people are more likely to engage in dishonest behavior when the benefits are shared with others (Wiltermuth, 2011), so we anticipate higher self-reporting of heads in the public decision environment, relative to the private setting. We also consider the potential interplay between worldviews and decision environments (private or public). Third, given the existing literature on gender differences in dishonest behavior (e.g., Abeler et al., 2014;Conrads et al., 2014), we take advantage of soliciting personal characteristics to consider 5 Findings were similar with the full sample. possible gender-specific effects of cultural worldviews on preferences for dishonesty.
Previous work reports greater dishonesty among men, so we consider whether this finding is independent of individual cultural worldviews by testing the null that self-reporting is equivalent across gender by worldview. 6 Figure 1 provides the predicted and self-reported frequency distributions for the outcomes of 10 coin flips. It shows that self-reported heads skew higher than predicted with a spike at the maximum of 10, which indicates some subjects partially lied and some maximally lied.

Results
This finding is consistent with prior studies that report many subjects lying, but not to extremes (e.g., Birkelund and Cherry, 2020;Abeler et al., 2019).  Overall, subjects reported an average of 6.64 heads out of 10 coin-flips. We consider the first research question that worldviews may influence preferences for honesty by comparing selfreports across the spectrum of each worldview dimension. For the hierarchy dimension, the average number of self-reported heads was similar across subjects that hold hierarchical and 6 Evidence on gender differences in honesty when subjects do not interact is mixed, see Muehlheusser et al. egalitarian views (6.54 vs 6.74; p=0.186). 7 A null result also emerges in the individualism dimension, with self-reports being similar across subjects with individualist and communitarian views (6.57 vs 6.69; p=0.306). Worldviews often carry considerable sway in preferences, but research shows the influence can be nuanced (e.g., Kahan et al., 2011;Cherry et al., 2017a). We therefore examine heterogeneous effects from worldviews.
We first extend the analysis by exploring whether any influence of worldviews may be specific to the private versus public return treatments. Generally, tests indicate no difference in the number of self-reported heads between the private and public treatments (6.60 vs. 6.69; p=0.649). Worldview-specific comparisons across private and public treatments also reveal no significant differences in self-reporting behavior. Results therefore find no nuance in the interplay between worldviews and whether dishonesty yields individual or shared benefits.
This finding may be due to the experimental design creating a setting in which subjects are anonymous, have no group-identity and do not know others' decisions-all factors that have been shown to close the gap between self-and other-regarding dishonesty (e.g., Gino and Galinsky, 2012;. Next, motivated by the literature on gender differences in preferences for honesty, we utilize responses to an ex-post questionnaire that asked subjects their gender. Research suggests that men tend to cheat more than women, though many studies have uncovered potential factors to alter this conclusion. For our purposes, Erat and Gneezy (2012) report that women are more dishonest than men when others benefit. So, in addition to examining gender-specific effects from worldviews, we consider if gender differences in self-reporting exist across private and public treatments. Table 2 reports the average number of selfreported heads by gender and worldview (combined private and public treatments). The numbers follow previous studies that find a significant gender difference. In our study, men self-report more heads than women on average (p=0.009), and this result persists when segmenting by treatment (private: p=0.091; public: p=0.039).
We extend this line of inquiry by examining the interplay of gender and worldviews.
From Table 2, comparing the mean self-reported heads by gender reveals that gender differences may be worldview-specific. While tests find that no significant gender differences exist between communitarian and egalitarian types (p=0.365 and p=0.394), they do indicate significant gender differences among subjects that have individualist versus hierarchical views (p=0.004 and p=0.003). Further, when comparing men and women across worldviews, we find differences across worldview types for women but not men. Specifically, women with individualist views reported fewer heads than women with communitarian views (p=0.036), and women with hierarchical views reported significantly fewer heads than women with egalitarian worldviews (p=0.016). In contrast, men report statistically equivalent numbers of heads across each worldview dimension (individualism: p=0.853; hierarchy: p=0.913). These findings offer evidence that cultural worldviews may underlie observed gender differences, and that much of the influence from worldviews occurs among women. We corroborate the aggregate tests with a conditional analysis. Table 3 provides the estimates of regression models that define an individual's number of self-reported heads as a function of the treatment, worldview scores and gender. 8 We estimate two models using the full sample and four models that stratify the sample by worldview types. 9 Results follow the unconditional tests. In particular, estimates find a gender difference in the pooled models, and the worldview-specific estimates indicate the gender difference exists among people with hierarchical and individualist worldviews.

Conclusion
Previous research has investigated the role of culture on preferences for honesty. This literature generally relies on proxies such as location and background to measure culture.
This study draws from the cultural cognition literature to use an established instrument to measure cultural worldviews at the individual level. Combined with a variant of a coin-flip 8 Results were robust to alternative specifications, including an ordered Probit. 9 The controls include age, education, income, ethnicity, and regions in the U.S., which were collected in the expost questionnaire. task commonly used in the dishonesty literature, we investigate the possible interplay between cultural worldviews and preferences for honesty. Results offer no evidence that worldviews matter in preferences for honesty whether the benefits from lying are private or shared.
However, an investigation of the nuances of worldviews uncovers new insights on gender differences reported in the literature. Consistent with previous studies, we find that men, on average, are more dishonest than women. This finding holds whether dishonesty benefits an individual or the entire group. The key finding however is that gender-specific preferences for honesty depend on individual cultural worldviews. Results indicate that cultural worldviews play a meaningful role in gender differences in levels of dishonest behavior, with worldviews being particularly influential among women.

3A. Experimental Instructions
Common for both treatments: