The Effect of Tax Privacy on Tax Compliance – An Experimental Investigation

Abstract

In this paper, we use a tax compliance game with a public good to investigate the impact of public disclosure on tax evasion behavior experimentally. Three different types of tax privacy are tested, ranging from complete privacy to full disclosure. We expect two different effects: first, a contagion effect, arising when an individual observes non-compliance of other individuals and therefore reduces her own tax compliance; second, a shame effect of increased tax compliance due to the anticipated shame of being declared a tax evader. Both these effects are supported by the experimental results. However, the shame effect reduces tax evasion only in the short run. The influence of shame diminishes over the course of the experiment with subjects observing the non-compliance of other participants. Thus, our results indicate that when the contagion and the shame effect are present the latter is not strong enough to override the former in the long run. Furthermore, disclosing tax information anonymously increases tax evasion compared to providing no information on tax evasion behavior. These observations are of particular importance for tax policy because public disclosure may lead to more evasion instead of less when supporting a crowding-out of the tax morale.

Acknowledgements

We thank Richard Sansing (editor), an anonymous reviewer as well as Friedel Bolle, René Fahr, Martin Fochmann, Axel Möhlmann, and seminar participants at the Freie Universität Berlin, University of Göttingen, University of Hannover, and University of Paderborn for their helpful comments and suggestions.

Supplemental Data and Research Materials

Supplemental data for this article can be accessed on the Taylor & Francis website, doi: 10.1080/09638180.2016.1258319.

ORCID

Kay Blaufus http://orcid.org/0000-0001-9010-3772

Notes

1 One important economic extension is the interactive theory developed by Graetz, Reinganum, and Wilde (1986) and Reinganum and Wilde (1986) which models a tax compliance game in which the audit probability is determined endogenously. Prior accounting research used this game to study, for example, the effects of tax uncertainty (e.g., Beck & Jung, 1989) and of signals regarding the individual's taxable income (e.g., Sansing 1993; Mills & Sansing, 2000) on tax compliance. Still, in our experiment we implement an exogenous audit probability as this simplifies the analysis and allows us to focus on the proposed shame and contagion effects.

2 In contrast to Laury and Wallace (2005) and Coricelli et al. (2010), Bosco and Mittone (1997) redistribute taxes in their experiment. However, due to their one-shot design it is impossible to examine a potential contagion effect.

3 Following the recommendation of Alm (1991, 2010), we describe the game in neutral language to avoid subjects using individual scripts when interpreting loaded terms (i.e., instead of the term ‘tax,’ we use the term ‘fee’).

4 These pictures were taken before the experiment inside the laboratory. After the experiment was finished, all photos were deleted in the presence of the participants.

5 All appendices are provided in the Online Supplemental Material.

6 Note that we run a second experiment right after this experiment. However, as both experiments are completely independent of each other, we only present the relevant instructions and screenshots.

7 We provide an overview on the audit probabilities and respective evasion rates for each period in Table 4 in Appendix E.

8 We run pairwise Mann–Whitney U tests to test the difference's significance. The results including the p-values of all tests in this section are presented in Table 1.

9 In addition, we have used a dependent variable which measures the average group evasion level in a random-effects linear regression. The main results here (not reported) are overall the same. However, using group averages makes us lose information that we are able to exploit with the random-effects panel regression and mixed effects model: First, the influence of socio-demographic variables is not included when analyzing on group level. Second, we are neither able to analyze individual feelings of shame nor social empathy due to the group level variable. Third, we are not able to analyze the behavior of Full Evaders and Non-Evaders on subject level. Fourth, we are not able to analyze the influence of the other group members' contributions on individual tax evasion decisions separately. Therefore, we decided against the usage of group evasion as dependent variable.

10 We cluster on group level and do not also cluster on individual level as for nested two-way clustering one only clusters at the highest level of aggregation, see Cameron, Gelbach, and Miller (2012). Although both tests might appear equivalent, we present both models as there are some cases in the robustness analysis where the mixed-effects modeling's iterations do not receive concavity. As we are not able to present results for these cases, the random-effect regression's results serve as analyses' basis.

11 We run all of the presented regressions again using different control variables, such as period, income or whether the subject studies in a bachelor degree's program. We also used another measure as variable for risk aversion. Instead of considering the number of risk averse decisions, we determined the first crossover point according to Holt and Laury (2002). The results remain unchanged. We also analyze to what extent the contribution of other group members influences personal tax evasion decisions. Hence, we calculate the mean contribution of the other four group members per period and run all regressions again separately for each treatment including this variable. The regressions' results reveal a negative correlation between the group members' contributions and personal tax evasion behavior if tax information is disclosed.

12 We have also analyzed how the feeling of shame develops over time. After subjects have received all their relevant and available information on the period's decisions and outcome, they are asked to state on a 7-point scale (with 1 = ‘Does not apply at all’ and 7 = ‘Fully applies’) whether they were a bit ashamed by themselves in front of their group members (measuring SHAME). We run a random-effects panel regression for the Full Information treatment for those subjects who evade taxes with SHAME as the dependent variable and TAX EVASION, PERIOD as well as the control variables presented in Section 4.2 as independent variables. In line with the observed behavior, we find that SHAME significantly decreases over time. The feeling of shame is greatest in early periods and decreases when social norms are adjusted if group behavior suggests that tax evasion is acceptable and common. Therefore, the shame effect only predominates in the first periods, but is outweighed by the contagion effect in later periods.

13 If we solely consider decisions, in which tax evasion amounts to 100% (Full Evaders), we do not observe any treatment effects. Thus, subjects who are predisposed to fully evade are not influenced by shame punishment.

14 The post-experimental questionnaire is displayed in Appendix D.

15 We also add EMPATHY as further individual control variable and rerun the random-effects panel regressions and multilevel mixed effects linear regressions as in Section 4.2. The obtained results remain unchanged.