The Social Norms of Tax Compliance: Evidence from Australia, Singapore, and the United States

Transcription

1 Journal of Business Ethics (2007) Ó Springer 2007 DOI /s x The Social Norms of Tax Compliance: Evidence from Australia, Singapore, and the United States Donna D. Bobek Robin W. Roberts John T. Sweeney Donna D. Bobek is an Associate Professor in the Kenneth G. Dixon School of Accounting at the University of Central Florida. Her research focuses on taxpayer and tax professional judgment and decision-making, with an emphasis on ethical decision-making. Donna has published in a number of academic journals including Accounting, Organizations & Society, Behavioral Research in Accounting, the Journal of the American Taxation Association, Advances in Taxation and Advances in Behavioral Accounting Research. John T. Sweeney is the Ted Saldin Distinguished Professor of Accounting and the Chair of the Department of Accounting at Washington State University. His research interests include accounting ethics and organizational justice. He has published in a number of accounting research journals, including Accounting, Organizations, & Society, The Accounting Review, Behavioral Research in Accounting, the Journal of Accounting & Public Policy, the Journal of Business Ethics, and Research on Accounting Ethics. Robin W. Roberts is the Al and Nancy Burnett Eminent Scholar and Director of the Kenneth G. Dixon School of Accounting at the University of Central Florida. His recent research focuses on ethics and regulation in the accounting profession and on corporate social responsibility. Robin has published in a number of academic journals including Accounting and the Public Interest, Accounting, Organizations & Society, Advances in Accounting, Auditing: A Journal of Practice & Theory, Critical Perspectives on Accounting, Journal of Accounting and Public Policy, Journal of Accounting Research, Journal of Business Ethics, Public Budgeting, Accounting & Financial Management, and Research in Governmental and Nonprofit Accounting. ABSTRACT. Tax compliance is a concern to governments around the world. Prior research (Alm, J. and I. Sanchez: 1995, KYKLOS 48, 3 19) has attributed unexplained inter-country differences in compliance rates to differences in social norms. Economics researchers studying tax compliance in the United States (U.S.) (see for example J. Andreoni et al.: 1998, Journal of Economic Literature 36, ) have called for more attention to social (as opposed to economic) influences on tax compliance. In this study, we extend this prior research by explicitly examining the role of social norms [Cialdini, R. and M. Trost: 1998, The Handbook of Social Psychology (Oxford University Press, New York)] on tax compliance in three different countries. We test our research hypotheses using a hypothetical compliance scenario, which was administered in Australia, Singapore, and the U.S. There were differences in compliance rates and social norms among the three countries. Factor analysis of the social norm questions identified three distinct social norm constructs. Two of these factors were significant in explaining tax compliance behavior. The first and most influential factor was taxpayers own personal moral beliefs, along with the beliefs of those close to them (e.g., friends and important others). The second significant factor represented societal views of proper behavior. We conclude that social norms help to explain tax compliance intentions and why tax compliance rates are higher than would be predicted by strictly economic models. KEY WORDS: social norms, tax compliance Introduction Tax compliance is a concern of governments around the world (see e.g., Alm and Sanchez, 1995; Feige, 1989; Frey and Weck-Hanneman, 1984). In the United States (U.S.) alone, non-compliance is estimated to cost the federal government over $300 billion annually ( However, traditional economic models of tax compliance, which primarily emphasize enforcement and detection variables (for a review see Andreoni et al., 1998), are unable to explain current levels of compliance. In fact, particularly in the U.S., compliance is much higher than these economic models would suggest.

2 Donna D. Bobek et al. For this reason, researchers have begun to incorporate non-economic variables into their models. Economists in particular have focussed on the concept of social norms (e.g., Alm et al., 1999; Pommerehne et al., 1994; Scholz and Pinney, 1995; Wenzel, 2004). However, most of these economic studies (Wenzel, 2004 is a notable exception) do not specify precisely what these social norms are; instead they model a variable that affects compliance in a manner consistent with a strong effect from some outside social influence. With tax return audit rates continually declining (Schnepper, 2004), a better understanding of this important influence on tax compliance is essential. In the present study, we directly test Alm and Sanchez s (1995) supposition that differences in compliance across countries are the result of differences in social norms. We draw on Cialdini and Trost s (1998) taxonomy of social norms in an effort to measure a comprehensive set of social influences on tax compliance. We tested our research hypotheses with a questionnaire administered in three separate countries: Australia, Singapore, and the U.S. The questionnaire included a hypothetical compliance scenario and items designed to measure each of the four types of social norms identified by Cialdini and Trost (1998). Factor analysis of the social norm questions revealed three distinct factors. The most influential factor on tax compliance was a factor that included both personal norms (self-based standards or expectations for behavior) and subjective norms (a person s perception of how most people important to him/her think he/she should behave). The second factor represented injunctive norms (general societal expectations of behavior) and was also significantly related to compliance. The third factor, which represented descriptive norms (standards that develop out of observation of others behavior), was not significantly related to subjects compliance intentions. We believe this research is important for several reasons. First, from a theoretical perspective we offer important extensions to both the tax compliance literature and accounting cross-cultural research. This study represents the first attempt to develop and test a comprehensive set of social norms for tax compliance. In addition to adding specificity to the inter-country tax compliance literature, we also address the recent call for research that investigates the effect of culture on ethical decision-making (e.g., Blodgett et al., 2001; Chanchani and MacGregor, 1999). Second, from a practical perspective, tax noncompliance is a widespread problem that costs governments, and thus its citizens, billions of dollars each year. There is evidence in the U.S. that public opinion toward cheating is becoming more lax. In a recent survey, the percentage of people who believed it was every American s civic duty to pay his fair share of taxes was only 68%, compared to 81% in 1999 (Dalrymple, 2003). Thus, to the extent that we can identify the antecedents of these attitudes, new strategies to address non-compliance can be developed. Third, with increasing economic globalization, there is a greater need to interact with people from other countries as well as understand different approaches to ethical decision-making (Blodgett et al., 2001; Cohen et al., 2002; Hsee and Weber, 1999). For example, differing social norms regarding tax compliance may lead to misunderstandings between business associates in different countries, or may hinder the effective international transfer of employees. Bolstering this concern is evidence that individuals are not good at predicting what their counterparts from other countries would do if faced with a similar situation (Hsee and Weber, 1999). Thus, a better understanding of the social influences on individuals ethical decision-making can be helpful to researchers as well as participants in the global economy. The remainder of this paper is organized as follows: in the next section, we synthesize prior tax research relating to social norms and cross-cultural tax compliance. Based on this discussion we propose research hypotheses for the present study, followed by a discussion of our method for testing the research hypotheses. Next, the results of the study are reported. In the final section, we draw conclusions and offer suggestions for future research. Prior research and hypothesis development The proposition that social norms are influential in tax compliance decisions is consistent with the moral psychology literature (Kohlberg, 1969; Rest, 1986). Kohlberg (1969) identifies three distinct hierarchical levels of moral development: (1) preconventional, where judgments are motivated by self-interest, (2) conventional, where individuals form moral judgments based upon referent norms and the expecta-

3 Social norms of tax compliance tions of significant others, and (3) principled, where self-chosen principles guide decisions. Cross-cultural research has indicated that most adults are at the conventional level of moral reasoning (Rest, 1986), and would therefore be influenced by referent social norms in their tax compliance decisions. Jones (1991) proposes that ethical decision-making is issue-contingent, whereby social consensus reduces the moral ambiguity of alternate actions, and provides guidance as to appropriate behavior. Thus, we maintain throughout this study that tax compliance is an ethical as well as economic decision and that social norms influence individual ethical decisionmaking. Social norms are defined as rules and standards that are understood by members of a group, and that guide and/or constrain social behavior without the force of law (Cialdini and Trost, 1998, p. 152). The four categories of social norms identified by Cialdini and Trost are: descriptive norms, injunctive norms, subjective norms, and personal norms. Descriptive norms are standards that develop out of observation of how others actually behave in particular situations. Injunctive norms, on the other hand, specify what should be done and are therefore the moral rules of the group. Subjective norms relate specifically to the expectations (i.e., injunctive norms) of referent others (e.g., family, friends, coworkers). Finally, personal norms are one s own self-based standards or expectations of appropriate behavior, which may arise from the internalization of injunctive, subjective and/or descriptive norms. Individuals may have different motivations to comply with social norms (Cialdini and Trost, 1998). These motivations include building and maintaining social relationships (injunctive and subjective norms) and making effective decisions in novel situations (descriptive norms). On the other hand, personal norms are primarily an internal influence on decision-making that relate to one s desire to maintain a positive self-image and live up to self-based standards or expectations (Schwarz, 1977). Table I details the differences between these four constructs. Some tax compliance researchers (e.g., Alm et al., 1999; Cowell, 1990; Davis et al., 2003; Porcano and Price, 1993) use terms similar to social norm to represent one or more of the social norm constructs discussed above, without specifying which type of norm may be affecting behavior. For example, analytically, Cowell (1990) modeled the tax compliance dilemma as including a social stigma influence that serves to increase compliance. Porcano and Price (1993), using a survey methodology, found that subjects were more likely to comply if a social sanction (e.g., cheaters being publicized in the newspaper) supplements enforcement efforts. Further, the effect of social sanctions increased as the likelihood of being caught (i.e., the audit rate) increased. They also concluded that many of their subjects did not exhibit internalized norms (i.e., personal norms) toward compliance while participating in their experiments because when the audit rate was very low and no social sanction was present, cheating was very high. Alm et al. (1999), in an experimental economics setting, manipulated social norms via voting. They found that when the social norm (i.e., the results of the vote) tended toward non-compliance, almost all the subjects cheated. However, when cheap talk 1 was allowed prior to voting, subjects generally voted in favor of stronger enforcement mechanisms, which then lead to more compliance. 2 Alm and Sanchez (1995) examined cross-cultural tax compliance in an experimental setting. Subjects in the U.S. and Spain participated in virtually identical experimental economics tasks and yet, controlling for all other factors, compliance was different between the participants in the two countries. While they did not explicitly measure the social norms of the participants, Alm and Sanchez attributed the unexplained disparity in compliance to differences in societal attitudes toward tax compliance (Alm and Sanchez, 1995, p. 4). Further, they argued that a multi-faceted approach to improving compliance requires taxing authorities to emphasize, among other things, the social obligation of paying one s taxes. While these prior studies focussed on an inferred general societal compliance pressure, other studies have addressed the impact of moral influences and referent others on compliance decisions. Scholz and Pinney (1995) investigate the concept of tax duty and find that it not only has a positive effect on compliance, but also increases the subjects subjective probability of the likelihood of detection. Hanno and Violette (1996) and Bobek and Hatfield (2003) employ two different versions of Ajzen s

4 Donna D. Bobek et al. Descriptive norm Definition Descriptive norms are derived from observation of what other people do in any given situation. Watching others provides information about what is normal in a novel or ambiguous situation. Social goal norm helps to achieve Descriptive norms are helpful for maximizing effectiveness in social situations. They help individuals make accurate choices about how to behave. When will the norm be influential? Descriptive norms will be most influential when the situation is novel, ambiguous, or uncertain. Individuals are most effective when they model behavior of those who are similar to them but also successful. Individuals should be more likely to imitate others who have visible signs of success. TABLE I Social norm constructs Injunctive norm Injunctive norms represent the perceived behavior of which most people approve (or disapprove). They specify what should (should not) be done. Injunctive norms prescribe appropriate behavior and thus aid in building and maintaining social relationships. Injunctive norms motivate behavior by promising social rewards or punishments. They need not be expressed in order to direct behavior. There is evidence that they can be primed and are more powerful when they are brought into focus. Subjective norm Subjective norms represent an individual s perception of the injunctive norms held by referent others (e.g., family, friends, and coworkers). Subjective norms represent the injunctive norms of those closest to an individual. Thus, they aid in building and maintaining social relationships with people whose opinions matter the most. Subjective norms will influence behavior when individuals are motivated to comply with the norms of referent others. Personal norm Personal norms are selfbased standards or expectations for behavior that flow from internalized values. Behaving consistent with one s personal norms helps to enhance or maintain one s sense of self-worth. Personal norms are enforced through the anticipation of selfenhancement or self-deprecation (i.e., they are self-reinforcing ).

5 Social norms of tax compliance (1991) Theory of Planned Behavior. In both of these studies, moral obligation (i.e., personal norms) and the opinions of important others (i.e., subjective norms) influence compliance decisions. Kaplan et al. (1997) find that taxpayers moral development is significantly correlated with compliance and also influences the effectiveness of certain educational communications to improve compliance. Bosco and Mittone (1997) test the role of moral obligation in an experimental economics setting and, while they were only partially successful at creating an artificial sense of moral obligation, they find that there is a natural cultural constraint that deters evasion. We found only one published study that examines the influence of both general social norms 3 and personal norms 4 on tax compliance behavior (Wenzel, 2004). In this study, conducted in Australia, personal norms are more influential than social norms. In addition, the importance of social norms is mediated by how strongly the subjects identified with being Australian. The present study extends prior research (e.g., Wenzel, 2004) in several ways. First, we consider compliance in three different countries: Australia, Singapore, and the U.S. Second, we develop and measure a more comprehensive set of social norm constructs. Third, Wenzel asks his subjects about their actual compliance behavior. This is often problematic as subjects are hesitant to admit to unethical and illegal behavior (Wenzel, 2004), potentially contributing to the relatively low R 2 (11%) of his compliance model. In the present study we consider taxpayers compliance intentions, rather than requesting subjects to reveal their actual tax compliance behavior. Behavioral intentions have been shown to be highly correlated with actual behavior (Ajzen, 1991). 5 The methodological improvements incorporated in this study enable our tax compliance model to explain more than three times the variance explained in Wenzel (2004). Hypotheses Two of the four types of social norms identified by Cialdini and Trost (1998), descriptive and injunctive, relate to the views/behavior of society in general. Alm and Sanchez (1995) and most other economic researchers are likely referring to these two types of norms when modeling tax compliance behavior. However, as noted, Wenzel (2004) finds that personal norms (he did not measure subjective norms) are more influential than general social norms. Knowing which specific social norms influence tax compliance intentions is critical if the ultimate goal is to understand and change behavior. Based on this review of prior research, we offer the following two hypotheses (in the alternate): Hypothesis One. Social norms (descriptive, injunctive, subjective and personal) are related to tax compliance intentions. Specifically, stronger non-compliance social norms are related to an increased likelihood of cheating intentions. Hypothesis Two. Personal norms will be more important than injunctive and descriptive norms for explaining subjects compliance intention. Most inter-country accounting research focusses on the construct of cultural norms, defined by Hofstede s (1980) seminal work, as opposed to social norms. [See Harrison and McKinnon (1999) and Chanchani and MacGregor (1999) for a review]. However, Hofstede s operationalization of nations as the unit of analysis, implying that each nation has its own singular culture, has been criticized as overly simplistic and inaccurate (Baskerville, 2003; Gernon and Wallace, 1995; Kohn, 1996). Most critics contend that a single nation represents multiple cultures, and that individuals are likely influenced more by the norms of their immediate social system. An indirect test of this proposition is to examine the influence of participants referent country on tax compliance intentions. If Alm and Sanchez (1995) are correct in attributing the difference in tax compliance behavior to social norms, then after accounting for social norms, we would expect that country would not be significant in explaining compliance. This leads us to the following hypothesis: Hypothesis Three. Differences in compliance intention among countries will be accounted for by differences in the social norms (descriptive, injunctive, subjective and personal) of the subjects of the countries.

6 Donna D. Bobek et al. TABLE II Sample demographics Australia Singapore United States Total sample Average age (in years) 22 a Gender (% male) 55% 49% 54% 53% Who prepares tax return Self or spouse 63% 96% 37% 63% Paid preparer 17% 54% 24% Other 20% 4% 9% 13% Education High school or less 16% 2% 6% 9% Some college 39% 44% 31% College graduate 37% 31% 35% 35% Post-graduate study 8% 67% 15% 25% Average annual household income $67,808 $78,272 $56,537 (in local currency) b Number of subjects a Significantly different from other two countries at significance level. b At the time the data was collected, $1 Singapore was approximately equal to $1 Australian. However, $1 American was equal to $1.75 Singapore. Thus, the income of U.S. subjects was actually higher (in equivalent currency) than in Singapore and Australia. There was no statistical difference between Singapore and Australia. Methodology Subjects Convenience samples of student and non-student subjects were obtained in three countries: Australia, Singapore, and the U.S. 6 Three countries were selected for this cross-country comparison because it allows us to test for differences in social norms. These three countries were selected because all three represent developed nations with relatively sophisticated income taxation systems. 7 In addition, the author team had contacts in each of these countries and, thus, was able to have the instruments administered reliably. Table II reports the overall sample demographics and the demographics of subjects in each country. As expected, there are some differences in the demographics among countries that will be controlled for in our statistical tests (discussed later). Specifically, in Australia the subjects are predominantly students, and consequently the average age was much lower (22 years) than in the U.S. (35.7 years) and Singapore (35.9 years). The U.S. has the highest percentage of paid preparer use, while virtually all of the Singaporean subjects prepare their own tax returns. 8 The Singaporean subjects are also better educated than the subjects from the U.S. and Australia. Finally, household income is highest (once the exchange rate is considered) in the U.S. To investigate for non-response bias, we compared the mean item responses between early and late U.S. and Singaporean subjects (all of the Australian data was collected at the same time). There were no differences in the means of any of the variables between early and late responders from the U.S. Early responders in Singapore, however, were less compliant than late responders. The dependent variable mean was 1.95 for early responders vs for late responders. This difference was significant at a p-value = This difference appears to relate to differences in personal norms. Specifically, their response to the following question: I think it would be morally wrong to take the additional $1,400. The response was measured on a 7-point scale with 1 = strongly agree and 7 = strongly disagree. The mean response for early (late) responders was 3.04 (1.95). This difference was

7 Social norms of tax compliance significant at p-value = The fact that nonresponding Singaporeans might be more compliant than responders does not change the interpretation of our results. Experimental instrument Subjects responded to an experimental instrument that presented a hypothetical tax compliance dilemma. Due to differences in tax laws among the three countries, the dilemma was modified somewhat to conform to a realistic compliance decision for subjects in each country. In Australia and the U.S., subjects responded to a scenario regarding business use of their personal automobile. In Singapore, subjects responded to a scenario regarding international travel expenses. 9 The scenarios are reproduced in the Appendix. The instrument was initially pilot tested with 91 students, who also answered debriefing questions. Based on their feedback, some of the questions were revised for clarity. After reading the scenario, subjects determined what course of action they would take in response to the tax compliance dilemma. Specifically subjects were asked How much of the additional $1,400 in automobile expense [overseas travel expense] do you think is closest to the amount you would deduct if you were in a similar situation? The response scale had six choices from $0 to $1,400. Each subject s response to this question forms the dependent variable: Compliance Intention. Subjects then responded to questions designed to assess each of the four types of social norms articulated by Cialdini and Trost (1998). Multiple questions were included for each construct. We also measured subjects familiarity with the deduction, perceived detection rates, and risk preference in order to control for each of these potential influences on compliance. Perceived probability of detection is included to control for differences in perceived enforcement levels, both among and within countries. Prior research has shown that perceived risk of detection is related more to compliance choice than actual detection rates (Carnes and Englebrecht, 1995). We also control for risk preferences as it may be related to compliance choice, and has been found to be different across cultures (Hsee and Weber, 1999). Finally, demographic information was collected. Results Descriptive statistics Table III reports descriptive statistics for the dependent variable and the control variables by country. On average, subjects said they would deduct approximately 36% of the additional $1,400 in expenses. Singaporean subjects had the lowest noncompliance rate at just over 25.8%, and Australian subjects had the highest at 44.8%. There was a sizable minority from all three countries who indicated they would be completely compliant. Across all three countries, 39.4% of subjects said they would not cheat at all. Of course, that implies that approximately 60% would potentially cheat under certain conditions. Complete compliance was highest in Singapore (53.3%) and lowest in Australia (30.3%). We proceed with our test of the effect of social norms as follows: first, we develop a baseline model which includes only control and demographic variables as explanatory variables so that we can later assess the relative importance provided by adding the social norm variables. Second, we construct the social norm variables and finally, we test our research hypotheses. Baseline model with control variables Regarding the control variables, in general the subjects were not familiar with the deduction in question. The average response was 6.2 on a 7-point scale, where 7 indicated they had never taken a similar deduction, and 1 indicated that they had taken a similar deduction very often. U.S. subjects were the most familiar (average 5.87), and Singaporean subjects were the least familiar with the deduction (6.56). T-tests of the means between countries indicated that the U.S. mean was different from the Singapore mean (p < 0.000). Risk preference was measured with the following question: If you want to claim a deduction, but the tax law is unclear, how certain (as a percentage) would you want to be that the taxing authority would allow the deduction before you would deduct it? Responses ranged from 100 to 0% at 10-point increments. The overall mean was 3.06, representing

8 Donna D. Bobek et al. TABLE III Descriptive statistics Australia Singapore United States Total sample Compliance intention a 2.24* Percentage of completely compliant subjects 30.3% 53.3% 40.7% 39.4% Familiarity with deduction b *** 6.18 Risk preference c Perceived likelihood of being audited (in general) d ** Perceived likelihood of being audited (specific) e Perceived likelihood of disallowance (specific) f *Significantly different than other two countries at 0.10 significance level; **significantly different than other two countries at 0.05 significance level; ***significantly different from Singapore at 0.05 significance level. a Measured on a 6-point scale where 0 is no cheating at all and 5 is cheating as much as possible. Thus, 2.5 is the midpoint of the scale. b Measured on a 7-point scale with 1 = has taken a similar deduction very often and 7 = has never taken a similar deduction. c Measured on a 11-point scale of 0 100%. Response to the following question: If you want to claim a deduction, but the tax law is unclear, how certain (as a percentage) would you want to be that the taxing authority would allow the deduction before you would deduct it? d Measured on a 9-point scale where 8 = less than 1%, 7 = 1%, 6 = 5%, 5 = 10%, 4 = 20%, 3 = 30%, 2 = 20%, 1 = 50%, and 0 = Over 50%. e Measured on a 7-point scale with 1 = if take deduction very likely will get audited and 7 = if take deduction very unlikely will get audited. f Measured on a 7-point scale with 1 = if take deduction and get audited very likely that deduction would be disallowed and 7 = if take deduction and get audited, very unlikely that deduction would be disallowed. approximately 70% certainty. U.S. subjects wanted to be the most certain (approx 75%), and Singaporean subjects the least certain (approx 67%). However, t-tests of the means indicated that there was no significant difference among the three countries. The final control variable (other than country) was perceived risk of detection. We measured this in several ways. We asked a general question,...how likely do you think it is that your 2000 tax return will be audited...? The response scale had nine choices: Less than 1%, 1%, 5%, 10%, 20%, 30%, 40%, 50%, and more than 50%. The average response for all subjects was 6.27, indicating that the average subject believed their tax return had a 5% chance of being audited. The average in the U.S. was 6.78 (which is closer to 1% and approximates the actual audit rate in the U.S.) and in Singapore 5.50 (which is closer to 7.5%). The mean for the Singaporean subjects was significantly different from the other two countries at the 0.05 significance level. We also asked subjects: (1) how likely they thought it was that they would get audited if they took the deduction in the case scenario; and (2) if they were audited, how likely would it be that the taxing authority would disallow the deduction. Both of these questions were on a 7-point likert scale with 1 = very likely, and 7 = very unlikely. There were no statistical differences in the responses among the three countries. The overall mean of 4.83 for the likelihood of audit question, indicates that, on average, subjects thought it was somewhat unlikely that their return would be audited. The overall mean of 3.47 for the likelihood that the deduction would be disallowed question suggests that, on average, subjects thought it was slightly more likely than not that the deduction would be disallowed. In order to get a baseline model, we ran a regression analysis with Compliance Intention as the dependent variable, and the control and demographic variables as explanatory variables. With all demographic and control variables included, this

9 Social norms of tax compliance model was significant (p-value = 0.000, Adjusted R 2 = 0.18). Risk preference (0.005 p-value), perceived likelihood of disallowance if audited (0.012 p-value), 10 and age (0.008 p-value) were significant determinants of compliance intention. 11 Specifically, older, better-educated, more risk-averse subjects, who perceived a greater risk of the deduction being disallowed, were more likely to be compliant. 12 Social norms Prior tax research has not directly addressed the potential influence of specific social norms on tax compliance. We therefore developed 14 questions to assess subjects social norms for compliance based upon a review of the relevant literature. Eight of these questions dealt specifically with the scenario in question (e.g., I think it would be morally wrong to take the additional $1,400 deduction ). Six of the questions were more general in nature (e.g., Most people will do anything to avoid paying taxes ). The questions were designed, a priori, to address each type of social norm. However, due to high correlations among the items, we subjected the questions to factor analysis in order to identify interpretable orthogonal factors. Table IV reports the results of this procedure. We eliminated three questions because their factor loadings were less than 0.6 on any of the factors, leaving 11 questions that loaded on three different factors. Table IV reports the rotated factor loadings based on a varimax rotation. The three factors explained 64.8% of the variance in the data. As can be seen by an inspection of the questions that loaded on each factor, the four distinct types of norms identified by Cialdini and Trost (1998) did not emerge. Questions representing personal norms (e.g., I think it is morally wrong and tax evasion is morally wrong in any amount ) and subjective norms (e.g., if my friends knew... they would think it was wrong, and most people who are important to me would think it was wrong ) loaded together on Factor 1, and explained 33.3% of the variance in the data. Cialdini and Trost (1998) point out that personal norms may arise from shared expectations in social interactions (p. 160). Since it is important others that we interact with, it is not surprising that personal and subjective norms are similar, and thus load on the same factor. Factors 2 and 3 are interpreted to represent injunctive norms and descriptive norms, respectively. Factor 2 explained 18.8% of the variance. Three questions loaded on this factor: most people will do anything to avoid paying taxes, it is socially acceptable to avoid paying taxes by whatever means possible, and it is ethical for me to do anything I think I can get away with to avoid paying taxes. All three of these questions relate to what society thinks is acceptable, thus Factor 2 is best described as injunctive norms. Finally, two questions loaded on Factor 3: Tax evasion is widespread in..., and I think most people think it is okay to pad deductions by a small amount... This factor explained 12.7% of the variance in the data and we believe this factor best captures descriptive norms. Panel A of Table V reports the mean factor scores by country. Based on the coding of the data, a negative factor score can be interpreted as the norms being unfavorable toward cheating. Thus, for example, the U.S. subjects indicated that their personal/subjective norms and injunctive norms were negative toward cheating. However, their view of what actually occurs is that taxpayers do cheat (positive value for descriptive norms). The Australian subjects had positive scores for all three factors, and their personal/subjective mean score was marginally significantly different than the other two countries (p-value = 0.07). Finally, the Singaporean subjects have negative personal/subjective norms toward cheating (similar in magnitude to the U.S. subjects); however, their view of society in general is much different than the U.S. subjects view. The Singaporean mean scores for Factors 2 and 3 suggest that while they believe that society does not view cheating as wrong (positive value for Factor 2), the mean for descriptive norms (what is actually going on) is sharply negative, and significantly different from the other two countries. Hypotheses testing Panel B of Table V reports the results of a general linear model with Compliance Intention as the dependent variable, and the three social norm factor scores along with factor by country interaction terms as explanatory variables. In addition, we include age,

10 Donna D. Bobek et al. TABLE IV Factor analysis social norm questions Question description Factor loadings Factor 1 Factor 2 Factor 3 I think it would be morally wrong... If I got caught... I would feel ashamed If I got away with... I would feel guilty If my friends knew... they would think it was wrong Most people who are important to me would think it was wrong... Tax evasion is morally wrong in any amount It is socially acceptable to avoid paying taxes by whatever means possible It is ethical for me to do anything I think I can get away with to avoid paying taxes Most people will do anything to avoid paying taxes Tax evasion is widespread in the U.S., [Singapore], and [Australia] I think most people think is okay to pad deductions by a small amount to reduce taxable income % of Variance explained 33.3% 18.8% 12.7% Factor description Personal and subjective norms Injunctive norms Descriptive norms education, risk preference, and perceived likelihood of detection as control variables, since they were significant in our baseline model. The model was significant (p < 0.000) with an R 2 = 0.445, and an adjusted R 2 = Including the social norm factor scores more than doubles the adjusted R 2 from the baseline model (adjusted R 2 = 0.180). The results reported in Panel B of Table V provide evidence in support of all three hypotheses. Personal/subjective norms (Factor 1) and injunctive norms (Factor 2) were significantly related to compliance intentions. This provides support for Hypothesis One. 13 Regarding Hypothesis Two, personal/subjective norms (Factor 1) were much more strongly related to subjects compliance intentions than either injunctive norms (Factor 2) or descriptive norms (Factor 3). Injunctive norms were significant (p-value 0.01), but the amount of variance explained by injunctive norms (Factor 2) was only 1/6th the amount explained by personal/subjective norms (Factor 1) [standardized regression coefficient 0.231(Factor 1) vs (Factor 2)]. Descriptive norms were not a significant influence on compliance intentions. Thus, Hypothesis Two is supported. As shown in Table III, there were significant differences in compliance intentions among the countries. Hypothesis Three states that social norms will account for the differences in compliance intentions across countries, after controlling for other explanatory variables (e.g., perceived probability of detection). To test this hypothesis, we include a country indicator variable in the regression reported in Table V. We also include interaction

11 Social norms of tax compliance TABLE V Social norm constructs Panel A mean factor score by country Australia Singapore United States Factor 1 personal and subjective norms * ) ) Factor 2 injunctive norms ** ) Factor 3 descriptive norms ) 0.564*** Panel B regression analysis ab Variable name F-Statistic Significance level Partial Eta-squared Intercept Country Factor Factor Factor Factor 1 Country Factor 2 Country Factor 3 Country Perceived risk of detection Risk preference Age Education Note: See Table IV for questions that make up each factor. *Significantly different from other two countries at 0.07 significance level; **Significantly different from U.S. at 0.10 significance level; ***Significantly different from other two countries at significance level. a Model tested: Compliance Intentions = b 0 + b 1 Country + b 2 Factor 1 Score + b 3 Factor 2 Score + b 4 Factor 3 Score + b 5 Country Factor 1 + b 6 Country Factor 2 + b 7 Country Factor 3 + b 8 Perceived Detection Risk + b 9 Risk Preference + b 10 Age + b 11 Education. b Model statistics Mean square F statistic (p-value = 0.000) Adjusted R 2 = terms between the indicator variable and each of the social norm factors. The results provide partial support for Hypothesis Three. The indicator variable was not significant, however, the interaction term between personal/subjective norms and country was significant. To explore the interaction terms, we examine the correlation coefficients between compliance intention and the factor scores for each country individually. Inspection of these correlations shows that it is only in the U.S. that injunctive norms are related to compliance intention (correlation coefficient of 0.362, p-value = 0.000). The correlations between injunctive norms (Factor 2) and compliance intention were not significant for the other two countries. Further, the correlation between personal/subjective norms (Factor 1) and compliance intention, while highly significant in all three countries, was slightly lower in the U.S. (r = 0.426), than in Australia (r = 0.570) and Singapore (r = 0.594). Thus, while social norms are strongly related to tax compliance and do explain inter-country compliance differences, the strength of the effect does vary across countries. Regarding the control variables, only age remained highly significant (p-value 0.005), with older subjects being more compliant. This is consistent with results from prior research (Andreoni et al., 1998). Perceived detection risk and education were marginally significant (p-values < 0.10).

12 Donna D. Bobek et al. Discussion Inclusion of the social norm variables in the model (adjusted R 2 = 0.39) considerably improved the variance explained over the baseline model (adjusted R 2 = 0.18). Furthermore, the taxpayer subjects country of origin was not significant in the model. All of our hypotheses regarding social norms were supported. In a setting where cheating was possible, 14 the factor representing personal/subjective norms was the primary determinant of subjects compliance choices. This result holds across all three countries. The direct effect of injunctive norms, while significant, was not nearly as influential as personal/subjective norms. Although not tested here, there may be an indirect effect from injunctive norms because it is likely that they are one of the influences on personal and subjective norms. In fact, Cialdini and Trost (1998) suggest that one way personal norms are developed is as a result of internalization of injunctive norms. Finally, the factor representing descriptive norms was not related to compliance. Cialdini and Trost (1998) characterize descriptive norms as arising out of observation of others behavior. In novel or ambiguous settings, watching others provides information. Thus, the fact that the descriptive norms factor was not useful in explaining the compliance intentions of our subjects seems reasonable for at least two reasons. First, tax compliance is not easily observable, and second, it is also not a novel setting. A striking observation from Panel B of Table V is the magnitude of the effect of personal/subjective norms. The influence of personal/subjective norms was much greater than the influence of traditional economic variables on cheating intentions. This observation, combined with the result in Cowell (1990) and Davis et al. (2003) who found, analytically, that compliance equilibriums are sticky, should give policymakers grave concerns that declining norms toward compliance could create an epidemic of non-compliant behavior. Limitations Our results cannot be generalized beyond the three countries studied. Additionally, the fact that the scenario used in Singapore was slightly different than the scenario used in Australia and the U.S. may have introduced additional error into our statistical tests, which could have muted the effects of the independent variables. The differences in the demographics of the subjects across countries, although controlled for in the statistical tests, may also have introduced additional error. 15 It is also possible that subjects were not completely honest about what they would do if they were in the situation, although the high level of cheating (over 60% agreed they would cheat at least a little) indicates that, at a minimum, they did not appear to fear admitting that they would cheat. Additionally, as part of the pilot test, we asked subjects if they believed they could answer the questions in the questionnaire truthfully (e.g. did you feel like your identity was sufficiently protected?). All but one subject responded yes. Finally, in each country we obtained convenience samples of taxpayers. Thus generalization of results within each country must also be made with caution. Summary, conclusion, and future research The main purpose of the present study is to test the conjecture, presented by prior economics and accounting research, that a primary determinant of tax compliance is social norms. Further, our aim is to provide more specificity to what exactly is meant when we attribute an effect to social norms. We draw on social psychology research (Cialdini and Trost, 1998) and identify four potential types of social norms: descriptive norms, injunctive norms, subjective norms, and personal norms. A secondary objective of this study is to determine whether social norms can explain differences in tax compliance among countries. We test our research hypotheses by obtaining responses to a hypothetical compliance dilemma from taxpayers in three different countries: Australia, Singapore, and the U.S. Our results show that the four different types of social norms are highly correlated; however, when subjected to factor analysis and varimax rotation to obtain orthogonal factors, three factors emerge. The first, and most important factor, is a combination of personal and subjective norms. Our subjects did not show a distinction between their own personal norms and the norms of important others. We identify this factor as representing both personal

13 Social norms of tax compliance and subjective norms. The second factor appears to reflect injunctive norms, which are defined as what one believes society, in general, views as proper behavior. This factor also influenced subjects compliance choices. Finally, factor three reflects descriptive norms, which are defined as what one perceives to represent actual behavior. This factor was not significant at explaining our subjects compliance choices. We included a number of control variables in our regression analysis, including perceived risk of detection, risk preference, country, subject age, and education. Consistent with our hypotheses, the social norms variables displayed the most influence on the subjects compliance choice, with personal/ subjective norms being four times as important as any other variable. While social norms were consistently important in explaining compliance across all three countries, there were some differences in social norms between the countries and also in the relative influence on compliance of the different types of social norms in each of the countries. Specifically, the personal/subjective norms of the Australian subjects were much more favorable toward cheating than were the subjects norms in the other two countries. This may have been, in part, because of their age. Additionally, subjects in the U.S. were more likely than those in Singapore to have negative injunctive norms (i.e., to believe that society in general views cheating as unacceptable). Conversely, the U.S. and Australian subjects were more likely to agree that cheating was actually occurring (i.e., positive descriptive norms) than were the Singaporean subjects. In none of the countries were the subjects influenced by their descriptive norms. The most important influence was personal/subjective norms. The correlation coefficients between personal/subjective norms and compliance intentions in Australia and Singapore were more than They were slightly less influential in the U.S. (but still more than 0.40). Finally, only in the U.S. was there a significant correlation between injunctive norms and compliance intentions. These results present a dilemma for policymakers. How can policymakers influence taxpayers personal and subjective norms? It is apparent from recent news accounts (e.g., Dalrymple, 2003) that, at least in the U.S., taxpayers personal norms toward compliance are declining. Cowell (1990) and Davis et al. (2003) demonstrate analytically that compliance equilibriums are sticky, and that when compliance rates change, they may do so in epidemic proportions. Due to this possibility, future research should seek to determine both ways to influence norms directly (e.g., through advertising appeals) and the effect public policy decisions have on citizens social norms regarding tax compliance. For example, do cuts in the Revenue Department s budget (e.g., in the U.S., the IRS) send an unintended message to taxpayers that cheating is okay? Do complex tax laws undermine taxpayers compliance norms? Does perceived unfairness in the distribution of the tax burden or in tax law changes affect compliance norms? Does the level of government spending and or spending priorities influence social norms? These are all important questions for future research. Notes 1 Cheap talk refers to the ability of subjects to informally discuss specific aspects of the experiment. 2 It should be noted that most experimental economics studies use neutral terminology. For example, instead of words like tax and audit they use terms like fee and check (Alm et al., 1999). Thus, any effect from a social norm construct arises from the experimental setting and not as a result of the subjects tapping into their own social norms of tax compliance schema. 3 Measured with three questions most people think it is acceptable..., most people think they should honestly declare, most people think... is a trivial offense. 4 Measured with three questions do YOU think you should honestly..., do YOU think it is acceptable..., do YOU think... is a trivial offense. 5 Ajzen (1991) frames his discussion of the link between behavioral intention and actual behavior in terms of individuals perceived behavioral control. Specifically, to the extent that a behavior is under an individual s volitional control (e.g., voting choice), there is a strong link from behavioral intention to actual behavior, and perceptions of behavioral control have been found to be largely irrelevant (Ajzen 1991). However when an action is outside one s volitional control (e.g., winning a gold medal in the Olympics), regardless of how strongly one intends to engage in the behavior, his/her actual behavior will be strongly, and directly, influenced by his/her actual (or perceived) ability to engage in the behavior. Thus, in a tax compliance scenario, as long as the taxpayer believes he has the ability (or opportunity)