Increasing tax compliance auditing, appeals to tax morale, or both? A lab experiment.

Transcription

1 Increasing tax compliance auditing, appeals to tax morale, or both? A lab experiment. Kristina M. Bott Norwegian School of Economics May 17, 2016 Abstract In a novel public good game design, this paper explores the impact of an interaction between audit probabilities and moral reminders on tax compliance. Some participants earn money and make decisions about tax payments, while others perform the same real effort task and spend the time in the lab, but are not asked to pay taxes. Instead of incomes, the latter group receives taxes paid by the former group. Moral reminders inform about a majority of taxpayers behaving in accordance with taxation rules. While audit probabilities and tax contributions are positively correlated, tax compliance is surprisingly high in a zero audit probability context. This is contrary to predictions as in Allingham and Sandmo (1972). Moral reminders play an important role in increasing tax compliance, and the more personal the information the more impactful it is. No evidence of an interaction effect of audit probabilities and moral reminders is found. I would like to thank Katie Baldiga, Dirk Engelmann, Uri Gneezy, Agne Kajackaite, Menusch Khadjavi, Henrik Kleven, Agnar Sandmo, Silvia Saccardo, Erik Ø. Sørensen, and participants at the Choice Lab Seminar April 2016, in Bergen, Norway, and at the Second Experimental Methods in Policy Conference 2015 for very useful comments. I am further very grateful for financial support from the Center of Ethics and Economics, and the Choice Lab at the Norwegian School of Economics. All errors are my own.

2 1 Introduction Standard economic theory predicts that a taxpayer will evade taxes if the marginal benefits are greater than the marginal costs of evading (Allingham and Sandmo, 1972). Raising audit probabilities increases the cost of evading, but is at the same time costly to the tax administration (Pomeranz, 2015; Kleven et al., 2011; Slemrod et al., 2001; Alm et al., 1992). Allingham and Sandmo (1972) apply the Becker (1968) model of crime to tax evasion decisions, assuming that humans are primarily rational utility maximisers and as such motivated by audit probabilities, penalties and the marginal tax rate. Empirically however, tax compliance rates are significantly larger than what the model predicts, even in contexts with low audit probabilities (Frey and Feld (2002) provide an overview). Developments in behavioural economics have increasingly contributed to expanding the view of humans being solely economically motivated, to including moral motivations in various decision contexts (Fehr and Schmidt, 1999; Andreoni, 1990). Non-pecuniary incentives to pay taxes, such as moral motivations, are not taken into account in the Allingham and Sandmo (1972) model. However such moral motivations also increase the cost of evading and can be relatively inexpensive to the tax administration (Gordon, 1989). While appeals to tax morale have been recognised in the empirical literature to increase tax compliance (see Luttmer and Singhal (2014) for an overview), the incorporation of moral considerations into the theoretical analysis of tax evasion yet remains unclear (Andreoni et al., 1998). Moreover, an interaction of increasing enforcement parameters and moral appeals has not been tested yet. Assuming the effects of both components of such an interaction are additive or even multiplicative, the effect on compliance could be stronger than the combination of employing only either of the two. To know if and how an interaction can increase compliant behaviour not only provides a deeper understanding of the drivers of moral motivations, but may also aid in modelling them. This paper is, to my knowledge, the first to tests the effect of an interaction between enforcement parameters and moral reminders on tax compliance, relative to the effect either of the two has on tax compliance. This will be tested in a public good game of three periods. Luttmer and Singhal (2014) attempt to conceptualise tax morale as a set of underlying motivations to comply with taxation rules, acting through five broad mechanisms. Moral reminders take the form of positive compliance information of other taxpayers, which falls into category three of the authors list of mechanisms. This information reveals that at least more than half of the taxpayers are compliant. 1

3 Such information can increase tax compliance as the reader may assimilate to behaviour of his/her peers due to reasons of either efficiency or because if everyone is doing it, it must be a sensible thing to do (Cialdini et al., 1990). Alternatively, this type of information may work simply as a more general reminder to behave in compliance with tax regulations. Differences in tax payments between those that receive this compliance information and those who do not are then analysed in contexts with varying audit probabilities, holding penalties and tax rates fixed. Since tax evasion is observable in the lab, and since it is further possible to control for beliefs about audit probabilities by fixing them at objective probabilities, a lab experiment is well suited to test this research question. This experiment is further novel in its choice of public good. Participants were assigned to groups of five, and all of them spend the same time in the lab. Three group members earn incomes based on their performance in the real effort task, and pay taxes on their incomes. The other two group members receive incomes paid for by tax payments of the three taxpaying participants. The real effort task that all group members face to the same extent was chosen to reflect variation in ability levels in order to create a similar variation in feelings of entitlements and fairness that may influence an individual s tax compliance in the real world. Those that pay taxes do not know anything about the effort levels of the other two participants. It is possible for taxpayers to evade taxes, though if they are caught evading, they have to not only repay the taxes they owe but also pay a penalty. This payment structure is common knowledge. In public good games, the public good consists classically of multiplied and redistributed tax payments, and creates a conflict between self- and group interests (see Ledyard (1995) for a discussion). Here, the public good has no pecuniary value to the taxpayer, but is rather the fairness everyone enjoys that the other two participants receive publicly provided private goods. The experiment was conducted with 740 students at the Behavioural Lab at the Rady School of Management at UC, San Diego, during the month of February While results in this experiment confirm the general assumption of the Allingham and Sandmo (1972) of tax evasion that audit probabilities increase tax compliance, tax payments are considerable, even in a non-enforcement context. A more general moral reminder only has a weakly positive effect on tax compliance when there is no enforcement, but disappears in a context with positive audit probabilities. As the moral reminder hits closer to home, moral reminders increase tax compliance significantly. 2

4 While tax morale seems to play an important role in tax compliance, I find no interaction effect between audit probabilities and moral appeals. The rest of the paper is structured as follows: section 2 explains the structure of this experiment in depth, as well as the pilot session and the power analysis. Section 3 is subdivided into the comparisons of treatment group differences and heterogeneity analysis, while section 4 concludes. 2 Experimental design 2.1 Structure This experiment consists of three periods and an initial period in which participants are practising the real effort task. Only after this period do they receive the instructions for the experiment, followed by three open question to ensure understanding of these instructions (Appendix A5 provides a screenshot of the instructions). Participants were required to answer all three questions correctly in order to proceed to the main experiment. Each then following period starts with a real effort task allowing participants to first earn money. The real effort task is followed by a stage in which earnings and taxes at a 30% tax rate are displayed. Taxpayers are here asked to enter the amount of taxes they are going to pay (Appendix A.2.2 provides a screenshot of the instructions), while being subject to treatment specific audit probabilities. In case an evader gets caught, she needs to repay the taxes she owes, but also faces a fine constituting 50% of the taxes not paid. Audit probabilities and penalties are common knowledge. The experiment finishes with a short questionnaire about demographics and risk tolerance of participants, as well as a number of open questions about tax compliance (Appendix A.3). At the end of each session, one period is publicly and at random determined for payment. A volunteer drew a piece of paper with either zero, one, two or three written on it. The numbers correspond to the respective periods. Treatments consist in variations of audit probabilities within and between subjects, as well as gradually revealing more information about compliance of other participants. The entire experiment lasts approximately 30 minutes, and was programmed in ztree (Fischbacher, 2007). The task is adapted from Charness et al. (2014), and requires participants to assign 3

5 numbers to letters shown on the screen. Which numbers to indicate can be extracted from a list of letter-number pairs displayed shown at the same time. 1 The letter-number lists change each time the corresponding number has been filled in the blank next to the letter. Each successfully decoded letter-number pair translates into a monetary amount, in this experiment, to 35 Cents. Participants decode as many letter-number pairs as they can and want in 90 seconds. After the task, everybody receives information about the number of correctly decoded letter-number pairs, together with their possible total earnings in the respective period. All participants start to practise the real effort task for one period (period zero), but do not receive any information on the actual experiment. This practice period is incentivised, though participants know that only one period will be paid out at random at the end of the experiment. 2 The number of correctly decoded letter-number pairs in the incentivised practice tasks before any treatment information provides an individual control for ability in later regressions analyses. To let participants work for their money during the experiment instead of providing an endowment increases fairness considerations (Cappelen et al., 2010), but can also increase tax evasion (Bühren and Kundt, 2013). If participants are on the margin to evade, compliance information of other taxpayers may increase tax payments, while taxpayers without such information may be prone to more evasion. Within each group of five, three group members participate in earlier, and two group members in later sessions due to the different payment schedule. Group members are anonymous throughout the experiment. Participants entering the lab in the earlier sessions are randomly assigned to groups of three. These three participants will be referred to as taxpayers. In the later sessions, two participants are randomly allocated to the groups of three taxpayers. Those two participants earn money performing the same real effort task and spend as much time in the lab as the taxpayers. Apart from being timely separated in their participation in this experiment, they differ from the taxpayers in two more ways. First, they are not required to pay taxes on their earnings. Since they do not make any decisions to pay taxes, they will be referred to as passive participants. Second, while taxpayers receive incomes based on how many 1 Figure A4 in the Appendix provides a exemplary screenshot. 2 In case period zero was determined for payment, tax payments for the other two participants are calculated as if all taxpayers were fully compliant. The payment to taxpayers is interpreted as net income and is the base for taxation. 30% of this base are then paid as wages to the other two participants who do not pay taxes. 4

6 letter-number pairs they decode, passive participants are paid with tax payments made by the taxpayers. There is no interaction between the taxpayers and the passive participants other than that the passive participants receive tax payments from the taxpayers. If one or more of the three taxpayers in a group evade their tax payments, the total income of the two passive participants will be directly lowered by the evaded amount. To ensure that passive participants would be paid for their time in the lab even if all taxpayers evade their tax payments, passive participants receive a three dollar subsidy from the experimenter, comparable to a minimum wage. The income of passive participants is thereby a publicly provided private good. The public good that is by definition available to everyone, is the fairness everyone enjoys because passive participants get paid for their time spent in the lab. If taxpayers do comply with taxation rules, they do however not enjoy any monetary utility from the public good. In the Allingham and Sandmo (1972) model utility depends primarily on private consumption. As long as one s disposable income is in expectation unaffected, there is no reason to contribute to the public good. To the extent that taxpayers care about fairness however, they will pay their taxes. It is important to note that there is a strong incentive in this experiment to evade taxes. In the Allingham and Sandmo (1972) model, taxpayers that maximise their expected utility will always declare less than their actual incomes if pπ < φ. p is equal to the audit probability, π to the penalty rate, and φ to the tax rate. Treatment specific audit probabilities that will be discussed in further detail in the next section, are either zero, five or 25%, the penalty rate is 50% and the tax rate is 30%. Considering the parameters in this experiment, the Allingham and Sandmo (1972) model predicts that expected utility maximisers would always cheat. Given these strong incentives to cheat in the baseline, there is scope for the moral reminders to increase tax compliance. The experiments with taxpayers were run over the course of February 2015 at the Behavioural Research Laboratory at the Rady School of Management at UC, San Diego. With twelve computers in the computer lab, there were a total of 46 sessions with taxpayers, who provide the data analysed in the following sections. Sessions with passive participants were conducted after the sessions with decision-makers. Two passive participants were later randomly assigned to groups of three taxpayers. Four pilot sessions with a total of 48 participants were conducted in order to test the design and determine group sizes to test the treatments that will be discussed next. 5

7 2.2 Hypotheses and treatments Treatments in this study are along two dimensions, a gradual increase in revealing information on compliance, and changes in audit probabilities (Table 1 provides an illustration of the design). An attempt was made for all sessions to include all treatments. 3 The first hypothesis is based on a fundamental finding of the Allingham and Sandmo (1972) model that predicts that tax evasion increases with decreases in audit probability if all other variables are kept constant. If audit probability is zero, the model predicts that no tax payments will be made. In the first period after the practice period, period one, audit probabilities are zero and participants of all groups read that no audits will be conducted in this period. The first hypothesis to be tested can then be formulated as follows: Hypothesis 1 When audit probabilities are 0%, tax contributions are zero. In period one, taxpayers are further randomly assigned to two groups, namely a neutral and a moral group. The two groups are different from each other in two ways. First, the moral groups receive two additional questions following the control questions. These control questions ask about their perception of how compliant other taxpayers are. In particular, How many participants in this experiment do you think pay the taxes they owe, and What fraction of taxes owed is being paid by the participants. These two questions should elicit ex-ante beliefs, but are also intended as a moral framing since they challenge the reader to think about tax compliance more broadly. Second, the moral groups read an additional sentence about the level of compliance from the first two sessions of this experiment at the tax paying stage. The sentence reads as follows: In the first two sessions of this experiment, the majority of participants paid the taxes they owed. 4 The neutral group moves through the tax paying stage without seeing this information. The idea of providing this information is to increase tax compliance, and there are two potential channels through which the 3 Some sessions did not allow for all treatments. There were for example cases where not enough participants signed or showed up. Treatments were therefore alternated. In 20 sessions, all four treatments were conducted, in 15 sessions three treatments and in nine sessions two treatments. 4 In the first period of the first two sessions, 66.7% taxpayers paid the taxes they owed, in the second period 54.2% and in the third period 62.5%. 6

8 moral sentence can accomplish this. First, taxpayers are compliant conditional on other taxpayers behaving compliantly (Luttmer and Singhal, 2014). For those taxpayers that believe that the majority behaves in accordance with taxation rules, this sentence provides no new information. Those taxpayers are unlikely to revise their beliefs downwards based on this type of information. For those that believe that less than the majority are compliant, this sentence may come as a surprise and in turn can increase tax compliance. An alternative channel though which this sentence can avoid tax evasion is through emphasising thoughts about compliance. In combination with the two questions about compliance, compliance in general gets more mental attention, and may trigger more compliant behaviour in taxpayers at the margin of cheating than without framing and/or moral treatment. With constant audit probabilities and fines, one would expect a greater level of tax compliance in the moral compared to the neutral group in all periods. In other words, forcing participants to think about compliance and presenting them with a moral reminder about relatively high compliance levels of other taxpayers could reduce tax evasion. This then leads to the following null hypothesis: Hypothesis 2 Taxpayers in the moral group declare as much taxes paid as a share of taxes owed as taxpayers in the neutral group. Since the moral treatment is expected to increase tax compliance, one could consider two alternative hypotheses, resulting in either a one- or a two-sided test. This analysis will discuss both. Period two is characterised by randomly assigning individuals from the neutral and the moral group to further subgroups. These subgroups differ in audit probabilities taxpayers are facing in periods two and three. Participants in one neutral and one moral subgroup make decisions in a context of a five percent audit probability, and participants in the other neutral and moral subgroups in a context of 25% audit probability. Group one in Table 1 will be referred to as neutral 5%, group two as neutral 25%, group three as moral 5% and group four as moral 25%. As per Allingham and Sandmo (1972), and irrespective of belonging to a neutral or a moral group, tax payments should augment with audit probabilities: Hypothesis 3 Increasing audit probabilities does not increase tax payments in neither the neutral nor the moral groups. 7

9 This analysis will here as well consider two alternative hypotheses, one leading to a one-, the other to a two-sided test. Both hypotheses will be discussed below. The impact of increases in audit probabilities on tax contributions are only compared between-subjects, since individual contributions have a tendency to fall in repeated public good games even when there are no changes in incentives (Ledyard, 1995). However a moral reminder and audit probabilities may increase tax compliance, it still remains to be tested how an interaction of the two would effect tax compliance. Period two thereby provides a possibility for such a test, allowing to compare a moral group to a neutral group with the same audit probability. Assuming the effects of both components of such an interaction are additive or multiplicative, the effect on tax compliance could be stronger than the effect of employing only either of the two. If a taxpayer believes that there is a greater chance to get caught evading, she may be more susceptible to a moral reminder. The level of embarrassment that would arise in that case could be perceived to be greater due to having received a moral reminder and ignored it. In that case, tax payments could increase. At the same time, audit probabilities in combination with moral reminders could crowd out intrinsic motivations to pay taxes, thus increase tax evasion. Or an interaction could simply have no effect at all, because those taxpayers that are susceptible to either treatment are also the ones that are equally susceptible to an interaction of the two. If this is the case, then the scope to increase tax compliance has reached its limits with either one of the treatments. Since it is difficult to know ex ante how such an interaction between audit probabilities and moral reminders would affect the results, a one-sided alternative hypothesis will not be considered here. Hypothesis 4 Increasing the audit probability in period two for either of the moral groups does not increase the share of taxes paid in taxes owed by more than in the corresponding neutral group. Shortly before period three, all participants in both the neutral and the moral groups are presented with within-group compliance information (Appendix A.2.2 provides a screenshot). Participants then know the share of taxpayers within their group of three taxpayers who were fully compliant in the second period, as well as the share of taxes paid in taxes owed in period two. This may influence the share of taxes paid in taxes owed in two ways. First, compliance information from the pilots may be more abstract to the participants compared to compliance information from a group 8

10 one has a personal link to. Second, compared to previous compliance information, the share of taxes paid in taxes owed now also allows for an approximate assessment of wages of the passive participants. It is unclear however, in which direction tax contributions would move with such information. Taxpayers might want to assimilate to, or to counter the behaviour of others. This corresponds to testing the following null hypothesis: Hypothesis 5 Providing within-group compliance information from period two does not increase tax compliance to a greater extent than providing information about compliance behaviour from the first two sessions of the experiment. 2.3 Pilot sessions and power analysis The design was tested in four pilot sessions in computer labs with twelve participants each. Standard deviations collected during the pilots determined the sizes of the treatment groups necessary to test the above mentioned hypotheses at a sufficient level of power. 5 Standard deviations were pooled across all treatment groups. Table 2 provides on overview over the hypothesised mean treatment differences in taxes paid as a share of taxes owed, and their standard deviations from the pilots. As per hypothesis 1, contributions are assumed to be zero in a non-enforcement context. Technically, any contribution made by a single taxpayer in period one thus refutes this hypothesis. According to hypothesis 2, taxpayers in neutral and moral groups pay the same taxes as a share of taxes they owe. A natural way to test this hypothesis is in a context without enforcement strategy, such as in period one. Assuming then that taxpayers are susceptible to a moral reminder, taxes paid as a share of taxes owed are expected to be at least ten percentage points higher in the moral groups. For the between-group comparisons of neutral and moral groups in period one, Figure 1 shows three potential mean differences. The centre one corresponds to the hypothesised mean difference in Table 2, while the other two are included to show the sensitivity to variations in mean 5 Information on compliance behaviour of other participants did not yet exist in the pilot sessions, taxpayers in the moral group therefore received information about the latest tax compliance estimate of American taxpayers by the Internal Revenue Services (IRS) at the tax paying stage: According to the Internal Revenue Service, the large majority of American taxpayers pays the taxes they owe. This fact remains statistically unchanged compared to previous years (Internal Revenue Service, 2014). 9

11 differences. If each of the four groups contain at least 96 participants, as indicated by the red vertical line, a treatment difference of ten percentage points can be determined at roughly 70% power. In order to test hypothesis 3, namely, if audit probabilities always increase contributions, the first row in Figure 2 provide between-group comparisons in period two. As indicated again by the red vertical lines, group sizes that can detect the hypothesised mean treatment differences here at relatively high power levels are similar to group sizes from previous comparisons. The bottom row in Figure 2 compares contributions between neutral 5% and moral 5%, as well as neutral 25% and moral 25% in period two. If the moral groups were different from the neutral groups in the context of audit probabilities, then one could speak of an interaction effect between the moral reminder and an audit probability, in other words, testing hypothesis 4. If the differences in means for the groups with five percent audit probability range between 20 and 24 percentage points, sample sizes should vary from 120 to 83 observations per group. For the groups with a 25% audit probability, the assumed mean differences of roughly 17 percentage points can be detected at a relatively high power. 3 Results With at least 96 observations in each of the four groups, the average power to analyse each of the above mentioned hypotheses is around 80%. The budget ultimately allowed for final group sizes to be 114, 111, 111, and 108 for the single groups. 3.1 Descriptives on the taxpayers A taxpayer earned on average 8.19 USD (SD 2.04 USD), while a passive participant received slightly less, namely 6.81 USD (SD 1.49 USD). Participants were on average 21 years old, and the sample is well gender-balanced. About one third of participants was American, who are slightly outnumbered by Asians or Asian-Americans. Study backgrounds are divided almost evenly between social and natural sciences. Risk tolerance is a self-reported score elicited in the survey at the end of the experiment. Participants were asked to rank their willingness to take risk on a scale from zero to ten, with zero representing the willingness to take the least risk. According to Dohmen et al. (2011), such a general risk question performs very similar compared 10

12 to more specific risk questions, as well as to eliciting risk in an experimental context. Figure 3 shows the distribution of self-reported risk tolerance in this sample. With a mean of 5.9 and a standard deviation of 2.2, participants here are self-declared more risk-taking compared to those in Dohmen et al. (2011), where the mean answer was 4.8, with a standard deviation of 2.5. Ability is a measure from period zero, in which individuals did not yet receive any treatment information. It is the number of correctly decoded letter-number pairs in this period. Most participants solved on average 21 pairs in 90 seconds. The decoding task was chosen in order to reflect different levels of ability among participants, and as per Figure 4, the number of decoded letter-number pairs in period zero is almost normally distributed over the taxpayers. Lastly, roughly half of the taxpayers in the moral groups believed that only up to 50% of the taxpayers were fully compliant. The additional sentence at the tax paying stage visible to the moral groups should thus be strong enough to come as a surprise to many taxpayers. 3.2 Treatment group differences This section focusses on between-group differences that follow directly from the randomisation. The order in which the results are discussed here are aligned with the hypotheses mentioned in section 2.2. Result 1 When audit probabilities are 0%, tax contributions are not zero. While Allingham and Sandmo (1972) predict no tax payments in a zero audit probability context, Figure 5 shows that contributions are surprisingly far from zero. In fact, 83% of taxpayers contributed at least some of the taxes they owed, whereof 63% of taxpayers paid all, and 20% paid some of taxes they owed in period one. Only 17% of taxpayers evaded all their taxes payable and paid nothing. This pattern in tax contributions is similar compared to the all-or-nothing behaviour observed in Alm et al. (1992). The figure clearly refute the hypothesis that all taxpayers are solely motivated by audit probabilities, penalties and/or the tax rate. It appears to be the case that taxpayers exhibit a relatively high level of tax moral the Allingham and Sandmo (1972) model seems to miss. 11

13 Result 2 Taxpayers in the moral group declare slightly more taxes paid as a share of taxes owed as taxpayers in the neutral group. Moving on to testing the role of the moral reminder, Figure 6 compares contributions of the moral and neutral groups in period one. The p-values reported are from a one-sided t-test. Taxpayers in the moral group pay on average 76.7% of the taxes they owe, while taxpayers in the neutral group pay on average only 71.2%. This difference in contributions is statistically significant at the 10% level. In a two-sided test however, the difference is no longer statistically significant at conventional levels. Result 3 Increasing audit probabilities increases tax payments in both the neutral and the moral groups. Comparing Figure 6 and Figure 7, the introduction of audit probabilities as low as 5% in period two decreases contributions significantly for both the neutral and the moral groups. When testing hypothesis 3, the analysis focusses solely on comparing between-group contributions in period two due to the tendency of diminished individual contributions in repeated public good games (Ledyard (1995) provides a discussion). Figure 7 confirms the prediction of the Allingham and Sandmo model of a positive relationship between tax contributions and audit probabilities. The first panel compares neutral groups with five and 25% audit probabilities, while the second panel compares moral groups with five and 25% audit probabilities. The differences in tax contributions between five and 25% audit probability are statistically significant at the 5% level for the neutral, and at the 1% level for the moral groups. 6 Hypothesis three can therefore be rejected. Taxpayers contribute more in groups with higher audit probabilities, irrespective of being part of the neutral or the moral groups. Result 4 There is no interaction effect between increasing audit probabilities and presenting a moral reminder. Moral groups with audit probabilities greater than zero do not pay more taxes than their neutral counterparts. Comparing the groups with the same audit probabilities across the panels in Figure 7 shows that the moral groups are not statistically significantly different from the 6 Results are from a one-sided test. When testing the differences in a two-sided test, they are both significant at the 5% level. 12

14 neutral groups in neither the five (p-value from a two-sided t-test: 0.64) nor the 25% (p-value from a two-sided t-test: 0.58) audit probability context. Taxpayers in moral groups with audit probabilities greater than zero do not pay more taxes than taxpayers in neutral groups with the same audit probability. It may be that taxpayers that are susceptible to moral reminders were also susceptible to an introduction of audit probabilities. In other words, an introduction of audit probabilities limits the scope to increase tax compliance through a moral reminder. Alternatively, intrinsic motivations to pay taxes could have been crowded out by the introduction of audit probabilities. In that case, the risk of getting caught in either audit probability scenario is the price a taxpayer may be willing to pay to justify the evasion of his taxes. It is however sufficient to increase audit probabilities only, a moral reminder does not increase tax compliance significantly. 3.3 Heterogeneity in treatment effects Ability and risk tolerance could may be suggestive of two potential mechanisms through which tax compliance is influenced. Ability could trigger entitlement feelings in those that are more able and consequently have greater earnings. Guided by a meritocratic argument, the more able could consider it to be unfair to give away a share of their income to passive participants, because taxpayers believe they themselves deserve the money given the effort they put in to earn it (Cappelen et al., 2010). Another fairness issue concerns the fact that passive participants perform the same real effort task and spend the same time in the lab as taxpayers. If taxpayers agreed that these participants deserved to be paid for their time in the lab, this is an argument for paying taxes. High levels of risk tolerance may also pose a threat to tax compliance. In Allingham and Sandmo (1972), tax evasion is modelled as a gamble between a risky investment, tax evasion, and a non risky investment, behaving in accordance with taxation rules. The more risk tolerant a person is the less susceptible s/he may be to increases in audit probabilities, and the more likely it is that s/he will choose the risky investment and evade her/his tax payments. Period two is best suited to explore heterogeneity in ability and risk tolerance, as there are two audit probability environments and individuals have not yet received any feedback. 13

15 Table 5 first explores heterogeneity in ability. The dependent variable in all specifications is again the share of taxes paid in taxes owed, regressed on a dummy for being part of a moral group, a dummy variable for being subjected to a 25% audit probability, risk tolerance and ability. A comparison of column one and two reveals that the results do not change much after the inclusion of demographic control variables. 7 Demographics do not seem to play a great role in the context of this experiment. It is however important to control for these variables and specifications including demographic control variables will be preferred over those excluding them. Not surprisingly, the self-reported measure for risk tolerance is negatively correlated with the dependent variable. The more risk-taking a person is, the more likely she is to evade her tax payments. If the willingness to take risk increases by one standard deviation, the share of taxes paid in taxes owed reduces by ten percentage points. This result is statistically significant at the 1% level, and in line with the seminal model of tax evasion. Ability is also negatively correlated, thus supporting the idea that the more able taxpayer may potentially feel more entitled to his/her earnings than wanting to give it away to the passive participants. If ability increases by one more decoded letter-number pair, tax payments decrease by 0.77 percentage points. This result is statistically significant at the 10% level. While the magnitude of the coefficient on ability may be small compared to the risk measure, Figure 4 also suggests that there is some more variation in ability than in risk tolerance. Having controlled for demographics and heterogeneity in ability and risk tolerance, column two results confirm several results as discussed in section 3.2. Taxpayers in moral groups are not different from taxpayers in neutral groups, as per the twosided test. Being in the 25% audit probability environment significantly increases tax contributions, namely, by almost 14 percentage points. The moral group with five percent audit probability is not statistically significantly different from the omitted neutral group with the corresponding audit probability, and the insignificant coefficient of the interaction moral x high audit probability concludes the same for the neutral and the moral groups with 25% audit probability. The observation that there is no difference between the neutral and the moral groups in any of the audit probability environments greater than zero also confirms result 4. In other words, there is no interaction effect between a moral reminder and neither the five nor the 25% audit probability environment. 7 Demographics were collected in the questionnaire at the end of the experiment and are gender, age, nationality, study type and risk-taking behaviour. 14

16 Columns three, five and seven show regression results excluding this interaction. In neither case do the results change much and in the interest of simplification, I will look for heterogeneous treatment effects excluding the higher order interactions that could be constructed based on the interaction effect between audit probability and the moral reminder. Columns four to seven analyse heterogeneity in tax payments based on ability, adding the interactions moral x ability as well as high audit probability x ability. While the interactions with ability are insignificant, the main effect of ability is relatively stable in magnitude across all columns. Result 5 None of the interactions with ability are statistically significant. There are no treatment-specific differences in the effect of ability. Columns one through three in Table 6 are the same table as in Table 5, but the interactions added in columns four through seven are now with the self-reported measure for risk tolerance instead of ability. Interestingly, the interaction of Moral x risk in columns three and four is negative and statistically significant at the 10% level. In other words, the more risk tolerant stand out more in the moral groups. A one standard deviation change in risk tolerance decreases tax payments by six percentage points in the moral groups compared to the neutral groups. Controlling for those that are more risk tolerant, tax payments in neutral and moral groups are now statistically different. 8 Result 2 : Once controlling for risk tolerance in the moral groups, the moral reminder has a significant effect on the share of taxes paid in taxes owed among those that are less risk-tolerant. This result is statistically significant at the 5% level, and does not change greatly when excluding the interaction Moral x high audit probability. Comparing the results to columns two and three shows, that the positive effect of the moral reminder disappears once the more risk-taking are mixed with the ones that are less risk-taking. Tax payments by those that love to gamble thus outweigh tax payments made by those that ordinarily rather choose to the less risky alternative. This result is in line with the predictions of Allingham and Sandmo (1972). Shortly before starting period three, taxpayers receive compliance information from period two. This information includes the share of taxpayers within their own 8 This result is very similar as in period one. 15

17 group who were fully compliant, as well as the share of taxes paid in taxes owed within their group. Table 7 shows the results from regressions that now also incorporates compliance information from period two and interactions with that variable. 9 The regressions in Table 7 are similar to the the Table 5 and Table 6 in that it explores heterogeneity in compliance information across the four treatment groups instead of ability and risk. Column one in Table 7 shows that if compliance of the other two group members increases to 100% in period two, the share of taxes paid in taxes owed of the third group member increases on average by 17.4 percentage points in period three. 10 This result is statistically significant at the 1% level. Result 2 : Compliance information of direct peers (group members) increases tax compliance to a greater extent than providing information on compliance from the first two sessions of this experiment. Even though the coefficient on compliance information is estimated based on within-treatment variation compliance behaviour of others, it is not subject to endogeneity bias but identified by random assignment. Decisions in the first two rounds are made individually and without any feedback to participants. All within-treatment variation based on compliance of the other two taxpayers within a group in the second round arises from random assignment to a group of three taxpayers. Since this randomisation happened at the session level, regressions in table 7 are clustered at the session level. Neither interactions of compliance information with the moral group dummy nor the dummy for the high audit probability context are statistically significant. There is no reason to assume differential effects of this information based on treatment groups. While compliance information from the first two sessions of the experiment only has a weakly positive impact on tax compliance in a one-sided but no significance in a two-sided test, information about the other two group members increase tax compliance to a greater extent and is now also statistically significant at the 1% level. A moral reminder in the form of compliance information clearly matters in tax compliance decisions, as does the subject of compliance information. 9 While the information on screen includes the observing person s contributions, the measure in the analysis excludes them. 10 Due to an error, six observations needed to be dropped in period three. 16

18 4 Concluding remarks Taxpayers may be motivated by more than audit probabilities, penalties and the marginal tax rate as the Allingham and Sandmo (1972) predicts. While audit probabilities and penalties have a positive effect on tax compliance (Pomeranz, 2015; Kleven et al., 2011; Slemrod et al., 2001), a number of field experiments in taxation also found non-pecuniary motivations to increase tax compliance (Hallsworth et al., 2015; Bott et al., 2014; Blumenthal et al., 2001; McGraw and Scholz, 1991; Schwartz and Orleans, 1967). However an interaction between audit probabilities and non-pecuniary motivations has to my knowledge not been tested yet. This paper contributes to the existing literature on lab experiments in tax evasion by exploring an interaction of audit probabilities and a moral reminder in the form providing information about a majority of taxpayers that behaves in accordance with taxation rules. In a novel variation of a public good game, the public good in this experiment constitutes the fairness everyone can enjoy that some workers are paid through tax payments of some other workers and thus receive publicly provided private goods. In a first round without audit probabilities, contributions are surprisingly far from zero. Taxpayers contribute an average of 75% of taxes paid as a share of taxes owed. This result is in stark contrast to the zero contribution predictions of the Allingham and Sandmo (1972) model of tax evasion, and also in contrast to previous lab and field experiments in tax evasion. Dwenger et al. (2015) mention that in the zero audit probability context 23% of taxpayers are fully compliant, whereas in Alm et al. (1992) compliance rates start at about 40% and end with 50% compliant taxpayers in the last period. In this present experiment, 61% of the taxpayers are compliant in the first round. A potential explanation for this difference may be the choice of public good and a sign that participants care in fact about fairness. In Alm et al. (1992), group tax payments were multiplied and then redistributed among the group members. However, since this observation is for one period only it cannot be answered it this level of tax compliance could be sustained over a number of periods. A second finding of this paper concerns the effect of audit probabilities. The greater the audit probabilities, the greater the tax contributions. While this conclusion is in line with most of the tax evasion literature (see Ledyard (1995) for an overview, or more recently also Pomeranz (2015); Kleven et al. (2011); Slemrod et al. (2001)), contributions here do not lead to believe that participants overestimated small probabilities as in Alm et al. (1992). At five percent audit probability in round two 47% of participants 17

19 were fully compliant, and at 25% audit probability compliance is only 10 percentage points higher compared to the lower audit probability context. Audit probabilities in Alm et al. (1992) were zero, two and ten percent, while compliance rates were similar. Moral reminders in the form of information of high levels of compliance of other taxpayers appear to play an important role in increasing tax compliance. Information from the first two sessions of this experiment did not have the same impact on tax compliance as providing compliance information of immediate group members. Compliance increased by only roughly five percentage points in the former case. If the two members of a group however become fully compliant, the third group member s contributions are likely to increase by 17.4 percentage points. This result is highly statistically significant. Taxpayers are evidently conditionally compliant, which is in line with findings from the field. If Norwegian taxpayers are informed that a majority of taxpayers declare incomes from abroad correctly and completely, Bott et al. (2014) find income declarations to increase significantly. An interaction effect between audit probabilities and moral reminders cannot be found in this experiment. It is possible that audit probabilities limit the scope for moral reminders to increase tax contributions. Alternatively, and similar as in Dwenger et al. (2015), intrinsic motivation to pay taxes may have been crowded out by the audit probability. Ability and risk are two potential mechanisms that increase tax compliance. Though there is no treatment specific effect of ability on tax compliance, more able individuals tend to contribute less taxes. The reason behind may be a meritocratic argument, that a taxpayer deserves remuneration for her/his ability instead of giving it away to other workers lowers tax compliance. In line with Allingham and Sandmo (1972), risk has a negative effect on tax compliance. The more risk tolerant a taxpayer is, the more likely s/he is to evade her/his taxes. In the moral treatments, the more risk tolerant naturally stand out more. Once the more risk tolerant are accounted for, the moral groups are significantly different from the neutral groups. Compliance information of other taxpayers is a strong moral motivator for an individual s tax compliance. 18

20 References Allingham, Michael G. and Agnar Sandmo (1972). Income tax evasion: a theoretical analysis, Journal of Public Economics, 1(3 4): Alm, James, Gary H. McClelland, and William D. Schulze (1992). Why do people pay taxes? Journal of Public Economics, 48(1): Andreoni, James (1990). Impure Altruism and Donations to Public Goods: A Theory of Warm-Glow Giving? Economic Journal, 100(401): Andreoni, James, Brian Erard, and Jonathan S. Feinstein (1998). Tax compliance, Journal of Economic Literature, 36(2): Becker, Gary S. (1968). Crime and punishment: An economic approach, Journal of Political Economy, 76(2): Blumenthal, Masha, Charles Christian, and Joel Slemrod (2001). Do normative appeals affect tax compliance? Evidence from a controlled experiment in Minnesota, National Tax Journal, 54(1): Bott, Kristina, Alexander W. Cappelen, Erik Ø. Sørensen, and Bertil Tungodden (2014). You ve got mail: a randomised field experiment on tax evasion, Discussion Paper at the Department of Economics, Norwegian School of Economics, 26/2014. Bühren, Christoph and Thorben Kundt (2013). Worker or shirker Who evades more taxes? A real effort experiment, Magks papers on economics, Philipps-Universität Marburg, Faculty of Business Administration and Economics, Department of Economics (Volkswirtschaftliche Abteilung). Cappelen, Alexander W., Erik Ø. Sørensen, and Bertil Tungodden (2010). Responsibility for what? Fairness and individual responsibility, European Economic Review, 54(3): Charness, Gary, David Masclet, and Marie Claire Villeval (2014). The dark side of competition for status, Management Science, 60(1): Cialdini, Robert B., Raymond R. Reno, and Carl A. Kallgren (1990). A focus theory of normative conduct: Recycling the concept of norms to reduce littering in public places, Journal of Personality and Social Psychology, 58(6):

21 Dohmen, Thomas, Armin Falk, David Huffman, Uwe Sunde, Jürgen Schupp, and Gert G. Wagner (2011). Individual risk attitudes: measurement, determinants, and behavioral consequences, Journal of the European Economic Association, 9(3): Dwenger, Nadja, Henrik Kleven, Imran Rasul, and Johannes Rincke (2015). Extrinsic and intrinsic motivations for tax compliance: evidence from a field experiment in germany, American Economic Journal: Economic Policy. Fehr, Ernst and Klaus M. Schmidt (1999). A theory of fairness, competition and cooperation, The Quarterly Journal of Economics, 114(3): Fischbacher, Urs (2007). z-tree: Zurich toolbox for ready-made economic experiments, Experimental Economics, 10(2): Frey, Bruno S. and Lars P. Feld (2002). Deterrence and Morale in Taxation: An Empirical Analysis, CESifo Working Paper Series 760, CESifo Group Munich. Gordon, James P. P. (1989). Individual morality and reputation costs as deterrents to tax evasion, European Economic Review, 33(4): Hallsworth, Michael, John A. List, Robert Metcalf, and Ivo Vlaev (2015). The Behavioralist as tax collector: Using natural field experiments to enhance tax compliance. Journal of Public Economics. Internal Revenue Service (2014). IRS releases new tax gap estimates: Compliance rates remain statistically unchanged from previous study, Remain-Statistically-Unchanged-From-Previous-Study: last checked: Kleven, Henrik Jacobsen, Martin B. Knudsen, Claus Thustrup Kreiner, Søren Pedersen, and Emmanuel Saez (2011). Unwilling or unable to cheat? Evidence from a tax audit experiment in Denmark, Econometrica, 79(3): Ledyard, John O. (1995). Public goods: A survey of experimental research, In: Kagel, J., Roth, A. (Eds.), Handbook of Experimental Economics. Princeton University Press. Luttmer, Erzo F. P. and Monica Singhal (2014). Tax morale, Journal of Economic Perspectives, 28(4):

22 McGraw, Kathleen M. and John T. Scholz (1991). Appeals to civic virtue versus attention to self-interest: Effects on tax compliance, Law and Society Review, published by: Wiley, 25(3): Pomeranz, Dina (2015). No taxation without information: Deterrence and selfenforcement in the value added tax, American Economic Review, 105(8): Schwartz, Richard D. and Sonya Orleans (1967). On legal sanctions, The University of Chicago Law Review, 34(2): Slemrod, Joel, Marsha Blumenthal, and Charles Christian (2001). Taxpayer response to an increased probability of audit: evidence from a controlled experiment in Minnesota, Journal of Public Economics, 79(3):

23 Power for a two-sample means test H0: t test µ2 = assuming µ1 versus σ1 Ha: = σ2 µ2 = σµ1 5 Figures.9 Power (1-β) Group size Parameters: α =.05, µ 1 =.1, σ =.4 Experimental-group mean (µ2) Neutral 0% vs. moral 0% (P1) Figure 1: Power analysis hypothesis 2: between-groups comparisons in periods one and two (two-sample means test) 22

24 Group size Neutral 5% vs. Neutral 25% in period 2 Parameters: α =.05, µ 1 =.3, σ =.44 Power for a two-sample means test H0: t test µ2 = assuming µ1 versus σ1 Ha: = σ2 µ2 = σ µ1 Experimental-group mean (µ2) Power (1-β) Power (1-β) Group size Moral 5% vs. Moral 25% in period 2 Parameters: α =.05, µ 1 =.52, σ = Group size Neutral 5% vs. moral 5% (P2) Parameters: α =.05, µ 1 =.3, σ =.5 Power for a two-sample means test H0: t test µ2 = assuming µ1 versus σ1 Ha: = σ2 µ2 = σ µ1 Experimental-group mean (µ2) Power for a two-sample means test H0: t test µ2 = assuming µ1 versus σ1 Ha: = σ2 µ2 = σ µ1 Experimental-group mean (µ2) Power (1-β) Power (1-β) Group size Neutral 25% vs. moral 25% (P2) Parameters: α =.05, µ 1 =.5, σ =.35 Figure 2: Power analysis hypothesis 3 and 4: between-groups comparisons in period two (two-sample means test) Power for a two-sample means test H0: t test µ2 = assuming µ1 versus σ1 Ha: = σ2 µ2 = σ µ1 Experimental-group mean (µ2) 23

25 Percent participants Self-reported risk score Figure 3: Distribution of the self-reported risk score Percent participants Average number decoded pairs (round 0) Figure 4: Distribution of the number of decoded letter-number pairs per taxpayer in period zero 24

26 Percent participants Period 1 Percent participants Period 2 Figure 5: Share of taxes paid in taxes owed (all groups, periods one and two) 25

27 Share taxes paid in taxes owed and SE 50% 60% 70% 80% Neutral, 0% p = 0.09 Moral, 0% Period 1 Figure 6: Comparing shares of taxes paid in taxes owed between neutral and moral groups in period one. The p-value is from a one-sided t-test. 26

28 ground2 ground2 Share taxes paid in taxes owed and SE 50% 60% 70% 80% Neutral 5% p = 0.02 Neutral 25% Share taxes paid in taxes owed and SE 50% 60% 70% 80% Moral 5% p = 0.01 Moral 25% Figure 7: Between-group comparisons of shares of taxes paid in taxes owed, period two. The p-values are from a one-sided t-test. The p-values for a two sided t-test between Neutral 5% and Moral 5%, and Neutral 25% and Moral 25% are 0.64 and 0.58, respectively. 27

29 6 Tables Table 1: Illustration of experimental design Period 0 Instructions real effort task and practice Experimental instructions + control questions Group Neutral Randomisation Group Tax morale Period 1 T 1 0 T 1 0 Randomisation Randomisation Group 1 Group 2 Group 3 Group 4 Neutral 5% Neutral 25% Moral 5% Moral 25% Period 2 T 2 5 T 2 25 T 2 5 T 2 25 Within-group tax compliance information from period 2* Period 3 T 3 5 T 3 25 T 3 5 T 3 25 Questionnaire Payments T = treatment, subscripts refer to periods and audit probabilities, respectively. * 1) Taxes paid as share of taxes owed, and 2) percentage of compliant taxpayers. 28

30 Table 2: Hypothesised mean differences between groups for power calculations, and SD from pilots Period 1 Mean Difference SD Neutral and Moral groups 10 (40) Period 2 Neutral 5% and Moral 5% 22 (50) Neutral 25% and Moral 25% 17 (35) Neutral 5% and Neutral 25% 20 (44) Moral 5% and Moral 25% 15 (45) Mean differences in shares of taxes paid in taxes owed are hypothesised. Standard deviations from four pilot sessions, conducted on and Group one is Neutral 5%, group two is Neutral 10%, group three is Moral 5% and group four is Moral 25% Table 3: Descriptives from pilot sessions Mean SD Ability (4.90) Profit 8.51 (2.28) Share taxes paid in taxes owed (46.36) Evaded some (36.00) Evaded all (37.83) Complied fully (46.26) N 48 Standard deviations in parentheses. Data from four pilot sessions, conducted on and

31 Table 4: Descriptives demographics Mean SD American 0.33 Asian/Asian-American 0.44 Other / unknown 0.23 Social Sciences 0.53 Natural Sciences 0.47 Other / unknown 0.01 Fraction females 0.49 Age Risk tolerance Ability N 444 Standard deviations in parentheses. Ability is measured as number of correctly decoded letter-number pairs in period zero. 30

32 Table 5: Heterogeneity regressions: share of taxes paid in taxes owed, all groups, period two, ability (1) (2) (3) (4) (5) (6) (7) Period 2 Period 2 Period 2 Period 2 Period 2 Period 2 Period 2 Moral groups (5.782) (5.907) (4.101) (18.352) (17.805) (5.978) (4.109) High detection probability (5.760) (5.900) (4.029) (5.906) (4.035) (17.485) (17.433) Moral x high detection probability (7.915) (8.131) (8.139) (8.261) Risk tolerance (0.865) (0.918) (0.916) (0.918) (0.917) (0.921) (0.920) Ability (0.392) (0.413) (0.413) (0.578) (0.577) (0.613) (0.605) Moral x ability (0.805) (0.804) High detection probability x ability (0.802) (0.788) Demographics No Yes Yes Yes Yes Yes Yes R Observations Standard errors in parentheses Dependent variable in (1)-(5): share of taxes paid in taxes owed. Omitted group is Neutral 5%. SE clustered at the individual level. Ability is the number of decoded letter-number pairs in period 0. Demographic control variables are age, gender, nationality and field of study. p < 0.10, p < 0.05, p <

33 Table 6: Heterogeneity regressions: share of taxes paid in taxes owed, all groups, period two, risk (1) (2) (3) (4) (5) (6) (7) Period 2 Period 2 Period 2 Period 2 Period 2 Period 2 Period 2 Moral groups (5.782) (5.907) (4.101) (11.305) (10.651) (5.906) (4.111) High detection probability (5.760) (5.900) (4.029) (5.899) (4.026) (12.042) (10.841) Moral x high detection probability (7.915) (8.131) (8.107) (8.140) Risk tolerance (0.865) (0.918) (0.916) (1.383) (1.378) (1.140) (1.144) Ability (0.392) (0.413) (0.413) (0.411) (0.411) (0.413) (0.412) Moral x risk (1.752) (1.748) High detection probability x risk (1.798) (1.797) Demographics No Yes Yes Yes Yes Yes Yes R Observations Standard errors in parentheses Dependent variable in (1)-(5): share of taxes paid in taxes owed. Omitted group is Neutral 5%. SE clustered at the individual level. Ability is the number of decoded letter-number pairs in period 0. Demographic control variables are age, gender, nationality and field of study. p < 0.10, p < 0.05, p <

34 Table 7: Regressions: share of taxes paid in taxes owed, all groups, period three (1) (2) (3) (4) (5) (6) Period 3 Period 3 Period 3 Period 3 Period 3 Period 3 Moral groups (5.671) (3.719) (8.054) (7.139) (5.405) (3.802) High detection probability (6.135) (4.631) (5.548) (4.594) (10.869) (9.280) Moral x high detection probability (8.344) (8.115) (8.244) Risk tolerance (0.829) (0.831) (0.847) (0.846) (0.843) (0.843) Ability (0.379) (0.378) (0.377) (0.376) (0.378) (0.378) Percent fully compliant in group w/o self in p (0.061) (0.061) (0.078) (0.078) (0.092) (0.093) Moral x compliance info p (0.118) (0.116) High detection probability x compliance info p (0.139) (0.140) Demographics Yes Yes Yes Yes Yes Yes r N Standard errors in parentheses Standard errors are clustered on the session level. Dependent variable in (1)-(6): share of taxes paid in taxes owed. Omitted group is Neutral 5%. p < 0.10, p < 0.05, p <

35 A Appendix: Additional analysis This appendix provides complementary analyses, as referred to in the main text. A.1 Additional figures Percent participants Share of compliant taxpayers (in %) Figure A1: Beliefs of group three and four members, of overall compliance behaviour 34

36 A.2 Procedure A.2.1 Instructions when entering the lab Dear participants. Please enter your name, your student ID and your into the participant sheet. Please put your phone on silent and stow it away, it should not be on your desk nor on your body such that it cannot distract you. As soon as every participant has signed into the sheet, you can now enter the computer lab. You can choose any of the computers that are turned on, please find you seat quietly. On your desk, you find a consent form. Please read the consent form carefully and make sure you understand everything. If you have any questions, please ask. Once you read it, please write your name in capital letters, sign it and write today s date its February xx, 2015 today. As soon as you are ready, you can just hold up the sheet, I will go around and collect them. All instructions regarding this experiment will be shown on the screen. It is very important that you read them carefully and ask me, if you have any questions. Make sure you do not click on any Next, Ok, Continue or Start buttons unless you have read the instructions. You cannot go back once you have pressed these buttons. All decision you are going to make in this experiment are anonymous. After the experiment, your payment will be paid out according to the letter on the post- it above your screen. As soon as your letter is called, you can approach the experimenter and you will be handed an envelope with your payment. Before you leave, please make sure the amount in the envelope is the same as on the payment form. If it is, please sign the payment form, to confirm, that you have received the payment. Do you have any questions? Then the experiment will now start the screen changes automatically. Figure A2: Instructions that were read out loud when entering the lab 35

37 A.2.2 Screenshots of the experiment Figure A3: Instructions for real effort task, before treatment information (Screenshot ztree). 36

38 Figure A4: Screenshot of the real effort task (Screenshot ztree). Figure A5: Instructions after practising real effort task (Screenshot ztree). 37

39 Figure A6: Results after real effort task and taxation (Screenshot ztree). Figure A7: Results after real effort task, and taxation with moral sentence (Screenshot ztree). 38

40 Figure A8: Within-group compliance information from period two (Screenshot ztree). 39