GENDER AND PUBLIC ATTITUDES TOWARD CORRUPTION AND TAX EVASION

I. INTRODUCTION

A large literature on compliance with the law has demonstrated strong gender differences. Women are less likely to commit almost all kinds of criminal offenses and are less likely to be involved in and approve of corruption, tax evasion, and other illicit activities. The observed gender differences at the micro level may even carry over to macro outcomes. For example, Dollar, Fisman, and Gatti (2001) and Swamy et al. (2001) show that the level of corruption in a country decreases in the percentage of women in parliament. Furthermore, the belief that greater female representation in public institutions would reduce corruption has produced specific policy actions. In 1999, Mexico set up new female uniformed patrols and increased the number of women police officers to reduce corruption (TI, Press release, March, 2000). A similar policy has been introduced in Lima, Peru (Swamy et al., 2001).

The literature offers two major theories to explain the gender differences (Gottfredson and Hirschi, 1990; Zager, 1994). One theory attributes gender differences to fundamental differences at the cognitive, emotional, and behavioral levels due to biological, psychological, and experiential realities. The alternative theory attributes gender differences to the different involvement of men and women in the workforce and in government. For example, according to that view women are less corrupt because they are less likely to occupy positions of power and therefore they have less opportunity to become corrupt.

Establishing whether gender differences in terms of illicit activities can be explained by the different opportunities for men and women has profound implications for policy prescriptions based on the observed gender differences. If gender differences are related primarily to opportunities, then greater representation of women in the positions of authority would not lead to reduction in illicit activities. Women would simply develop attitudes and behaviors similar to men. Unfortunately, although the literature provides ample evidence for gender differences, it does not say much about why these differences exist. Therefore, it is hard to determine whether gender promotion policies would have the desired effect on corruption and other illicit activities.

Here, we make an effort to fill this gap using data from the World Values Survey (WVS) and the European Values Survey (EVS). As in the previous literature (Swamy et al., 2001; Torgler, 2007; Torgler and Schneider, 2007), we find that women are less likely to approve of corruption and tax evasion. However, we take a step further to investigate whether the gender effect can be explained by the opportunities for women to become involved in illicit activities. We accomplish this in three ways. First, we estimate several empirical models where we introduce variables that proxy for the opportunities to become involved in illicit activities. We find that adding these variables does little to change the effect of gender. Second, we investigate the effect of gender in 1980, 1990, and 2000 separately. All the countries in our sample have experienced an increase in the involvement of women in the labor markets and government during that period. Yet, we show that the gender effects persist in direction and size over time. Third, we estimate country-specific regressions and show that the gender effect is similar across countries with quite different levels of women involvement in the labor market and politics.

Furthermore, unlike previous studies (Swamy et al., 2001), we investigate the attitudes toward corruption as well as tax evasion. We find that the gender effect for corruption is much more robust across the various specifications. In addition, we extend the empirics in Swamy et al. (2001) by adding the 2000 survey. This gives us not only a longer time period to explore but also more recent data that can be compared with earlier data.

The rest of the article is structured as follows. Section II gives an overview of the existing literature and has the aim to outline our theoretical approach. The interdisciplinary phenomenon of corruption and tax evasion makes it also interesting to focus on research findings in different social science areas. Section III then presents the empirical findings, and Section IV concludes with final remarks.

II. EVIDENCE FOR GENDER DIFFERENCES

The correlation between gender and crime or delinquent behavior has been investigated extensively in the criminology literature. The following sweeping statement by Mears, Ploeger, and Warr (2000) summarizes the general finding that women are less likely to be involved in such activities compared with men:

In terms of tax compliance, the literature, including several experimental studies, shows a tendency for greater tax compliance among women (Alm, Jackson, and McKee, 2006; Baldry, 1987; Torgler and Schaltegger, 2005).¹ The experimental findings are particularly interesting because they control the opportunities set of the participants and can therefore provide direct evidence against the difference-in-opportunities theory.² Experimental studies also show that gender influences charitable giving, bargaining, household decision making, and contributing to public goods (Andreoni and Vesterlund, 2001; Brown-Kruse and Hummels, 1993; Eckel and Grossman, 2001; Nowell and Tinkler, 1994). Women are generally more likely to contribute to public goods or common objectives although there are some nuances to that finding. For example, Andreoni and Vesterlund (2001) found that women are more generous in experimental situations when the price of giving is high. Conversely, men become more generous when the price of giving is low.

On broader level, the evidence whether greater women involvement in government leads to lower corruption is mixed. Swamy et al. (2001) and Dollar, Fisman, and Gatti (2001) argue that greater female participation in government lowers corruption. In a related analysis, Mocan (2004) uses data from the International Crime Victim Survey to show that men are more likely to be asked for a bribe compared with women. However, Sung (2003) reports that the gender effect loses significance when the empirical models include measures of constitutional liberalism. In addition, Mukherjee and Gokcekus (2004) argue that there is an optimal level of women participation in public organizations: an increase in the proportion of women in public organizations reduces corruption if women's share is less than one-third. However, the gender effect on corruption disappears when women occupy about 45% of the positions in a public organization. An increase in women's share above 70% even raises corruption.

Large differences between men and women are observed in other circumstances as well. For example, motor vehicle accident rates are higher for men for all kinds of accidents. Men are also more likely to be the victim of an accidental drowning or accidents caused by fire (Junger 1994). Similarly, alcohol and drug abuse are more common among men than women (Gottfredson and Hirschi, 1990). Evidence about gender differences can also be found in helping behavior (Eagly and Crowley, 1986) or ethical decision making (Ford and Woodrow, 1994; Glover et al., 1997; Reiss and Mitra, 1998).

In summary, despite some caveats the literature finds that women are more compliant with laws and are less likely to be involved in crime and illicit activities. However, there is substantially less evidence on why men and women differ in this respect. According to one theory, gender differences are attributed to different biological, psychological, and experiential realities, whereas the other theory explains gender differences by the different external constraints and opportunities faced by men and women. Although the evidence on the two theories is limited, the available evidence seems to provide little support for the opportunities argument. For example, the theories on constraints and opportunities would suggest that crime rates for men and women would become more similar with the increasing equality of status between men and women over time (Gottfredson and Hirschi, 1990). Indeed, judging by arrest records the share of criminal offenses by women in the United States has increased and has become more similar to the crime rates of men. However, careful interpretation of this evidence by Steffensmeier and Schwartz (2004) based on pooling data from several sources reveals that “crime is still a man's world.” They found that the increase in arrests is due primarily to increasing arrest rates for females. Similar evidence is presented by Gottfredson and Hirschi (1990) who show that the differences in crime rates between men and women persist even after the rapid increase of female labor-force participation rates in the United States suggesting that the equality thesis cannot explain the gender differences.

Some researchers stress that female roles and crime can be seen as complex outcomes of socioeconomic, political, and historical factors that go beyond gender equality (Steffensmeier et al., 1989). For example, influenced by the sociological theory of Sutherland (1947) who argues that delinquency is learned behavior imitating social groups, Mears, Ploeger, and Warr (2000) found that men are more likely than women to have delinquent friends, and that they appeared to be more strongly influenced by delinquent peers.

The objective of the analysis in the following sections is to investigate further the validity of the opportunities theory. Unlike the previous studies that focus on crime rates, we investigate attitudes toward corruption and tax evasion, two topics of significant interest in economics and other social sciences. Furthermore, we use data from several countries over time unlike much of the literature on crime that has focused on the United States.

III. DATA

The data used in the present study have been obtained from the WVS and the EVS. The surveys were first conducted in 1981–1984. The researchers who conduct and administer the WVS/EVS in their respective countries are required to follow the methodological requirements of the World Values Association.

The surveys are generally based on national representative samples of at least 1,000 individuals, aged 18 and over (although sometimes people under the age of 18 participate). The samples are selected using probability random methods, and the questions contained within the surveys generally do not deviate from the original official questionnaire.³ The WVS/EVS inquires about the acceptability of various dishonest or illegal activities. The questions on the justifiability of corruption and tax evasion that are of primary interest in this article are stated as follows:

The variables based on these questions are not free from biases and problems as they inquire about self-reported attitudes and behavior and hypothetical choices. It is possible that individuals who are or have been involved in illegal activities would tend to excuse such behavior declaring high justifiability (Torgler and Schneider, 2007). Respondents may also overstate their degree of compliance (Andreoni, Erard, and Feinstein, 1998). Elffers, Weigel, and Hessing (1987), for example, found strong differences between actual and reported tax evasion. Nonetheless, we expect more honest reporting because the WVS and the EVS ask about social norms rather than personal involvement in illicit activities. Moreover, the dataset is drawn from wide-ranging surveys that cover multiple topics; this should reduce respondent suspicion and the framing effects of other tax-related questions.

In recent years, a number of studies have shown that values, norms, and attitudes affect economic behavior and institutions (Knack and Keefer, 1997). According to Ajzen and Fishbein (1980) and Lewis (1982), behavior can be predicted from attitudes and subjective norms. The tax compliance literature, for example, has documented a strong link between attitudes toward tax compliance and actual compliance (Cummings et al., 2009). Weck (1983) reports a negative correlation between tax morale (attitudes toward paying taxes) and the size of the shadow economy. When compared with other variables, tax morale has the most significant impact on the size of the shadow economy.⁴ For our purposes, it is also useful to note that the justifiability of corruption variable is positively correlated with well-known indexes of the actual level of corruption such as the Transparency International Corruption Perception Index (correlation coefficient is 0.358 and statistically significant) and the Quality of Government rating (control of corruption) developed by Kaufmann, Kraay, and Mastruzzi (2003) (correlation coefficient 0.380 and also statistically significant).⁵

We work with data from the following European countries: France, Great Britain, Italy, The Netherlands, Denmark, Belgium, Ireland, and Spain. There are two reasons why we work with this subsample of the WVS. First, working with a homogeneous region reduces potential biases from cross-country comparisons. Cross-cultural comparisons should be treated with caution. In countries where corruption and tax evasion are widespread and delays in transactions are long, additional payments to “speed up” the process may be justifiable.⁶ Second, we are interested in covering a period of almost 20 yr, which reduces the number of European countries that can be investigated.

Figure 1 plots the distribution of views on the justifiability of corruption and tax evasion for women and men. The ten-scale index with the two extreme points “never justified” and “always justified” was recoded into a four-point scale (0, 1, 2, 3), with the value 3 standing for “never justifiable.” The scores from 4 to 10 were chosen by few respondents and were therefore integrated into the value 0. The figure shows that women are more likely to report that tax evasion and corruption are never justifiable.

Table 1 provides further evidence of the gender differences. We use the Wilcoxon rank-sum (Mann-Whitney) test to investigate whether the two subsamples split along gender lines have the same distribution.⁷ We perform the tests for the overall sample as well as for each time period and each country in the sample and find statistically significant differences between women and men in all 12 cases. In addition, we also use the two-sample Kolmogorov-Smirnov test to determine whether there are any differences in the distribution between men and women. The p-values indicate that the difference is always statistically significant.

IV. EMPIRICAL RESULTS

This section tests whether the effect of gender is statistically significant in a multivariate context. We also discuss the size of the gender differences and the robustness of the results in various specifications. To simplify matters, we recoded the ten-scale index with the two extreme points “never justified” and “always justified” into a two-point scale (0, 1), with the value 1 standing for “never justifiable.” As Figure 1 shows, the share of respondents stating that corruption and tax evasion are never justifiable is quite high making that answer a natural cut-off point. The advantage of recoding the answers into a dummy variable is that we can report fewer coefficient estimates and, also, we can interpret them more directly in terms of probabilities. Nevertheless, as a robustness test we also report ordered probit estimations using the scale from Figure 1 (see Tables 5 and 6). In addition, we estimated all equations using an ordered probit model without recoding the answers. We obtained similar results that are available upon request.

To facilitate the interpretation of the results, we report the marginal effects of the probit estimations instead of the estimated probit coefficients, so that we can discuss not only the direction of the effects but also their sizes. The estimated probit coefficients are based on a non-linear estimation technique and cannot be interpreted readily in terms of the quantitative sizes of the effects. Calculating the marginal effects is a method to find the quantitative effect of an independent variable. The marginal effect indicates the change in the share of individuals (or the probability of) belonging to the “never justifiable” category when the independent variable increases by 1 unit. If the independent variable is a dummy variable such as gender, the marginal effect shows the difference in the likelihood of reporting that corruption and tax evasion are never justifiable for one group (e.g., female respondents) compared with the reference group (e.g., male respondents).

We estimate probit models that explain the justifiability of corruption and tax evasion using gender and different variables that proxy for respondents' opportunity to be involved in illicit activities in the different specifications. We are interested in the effect of including these variables on the gender differences reported in more parsimonious equations where we do not control for such opportunities. All models include country and time dummy variables.⁸

Tables 2 and 3 present the first results. EQ1 and 7 include gender as well as age and marital status. We find a strong gender effect. The likelihood that a respondent finds corruption and tax evasion justifiable is 5.6 or 7.1 percentage points lower for women than for men. These are strong quantitative effects. Then, we add variables that can be identified as opportunity/constraints proxies. We add education (EQ2 and 8), employment status (EQ3 and 9) and economic situation (EQ4 and 10). Education is related to taxpayer's knowledge about the tax and government system. Better-educated taxpayers might be less compliant because they better understand the opportunities for evasion, avoidance, or corruption. The employment status is also connected to opportunities. For example, self-employees have a higher opportunity to evade or avoid taxes or to bribe government officials. Furthermore, individuals with higher income are more likely to be asked for a bribe, as are those with better education. In countries with a progressive income tax rate, taxpayers with higher income realize a higher dollar return by evading and have more opportunities to hire excellent tax agents that explore the possibilities of reducing the tax burden through tax avoidance. Conversely, lower income taxpayers might be less in a position to take these risks because of a high marginal utility loss (wealth reduction) if they are caught and penalized.

Table 2, which reports the results on the justifiability of corruption, shows that the quantitative effect of gender is not affected substantially by the additional control variables. The marginal effect of gender is still 5.6 percentage points even after controlling for education and employment status. Adding income reduces it slightly to 5.4 percentage points. A stronger decrease of the marginal effect of gender is observed in Table 3, which reports the results on the justifiability of tax evasion. The marginal effect of gender decreases from 7.1 to 6.1 percentage points. Nonetheless, the gender effect remains statistically and economically significant, suggesting that it cannot be explained away by differences in opportunities.

Looking at the control variables, we find a statistically significant age effect. Greater age is associated with lower justifiability of corruption and tax evasion. We find a statistically significant effect of education on the justifiability of corruption (lower justifiability) but not on the justifiability of tax evasion. In both cases, married people report a lower justifiability of illicit activities compared to individuals with another marital status. Being married increases the share of respondents indicating that accepting a bribe is never justifiable by more than 3 percentage points and increases the probability of stating that tax evasion is never justifiable by more than 4 percentage points. Being self-employed increases the justifiability of evading taxes quite substantially (marginal effects around 6 percentage points). The effect of economic class is similar to that of education, that is, the highest economic class has the lowest justifiability of corruption with a marginal effect of 1.8 percentage points. However, the coefficient is not statistically significant in the equations explaining the justifiability of tax evasion. In EQ5 and 6 (corruption) and EQ11 and 12 (tax evasion), we use a subsample of women and men separately. Overall, the results are comparable. Interestingly, the education variable works stronger among women than men. Looking at corruption, we also observe that economic class matters for women but not for men.

Table 4 reports several robustness checks for the gender effect summarizing the results of 22 regressions (see EQ13–34). We estimate equations using each of the years in our sample (1981, 1990, and 1999) separately (EQ13–18) and also using data from each country separately (EQ19–34). For simplicity, we report only the marginal effect of the gender variable.

The opportunities theory would suggest that a growing equality of status between men and women over time would lead to decreasing gender differences. The share of women parliamentarians in our sample increased from approximately 9% in 1980 to approximately 21% in 2000. The share of women in the labor force of the countries in our sample increased from approximately 36% in 1980 to approximately 43% in 2000.⁹

However, such an argument is not supported by our results. Gender differences remain statistically significant in all three time periods. We observe some changes over time and differences between variables. The effect decreases in 1980s for corruption but increases for tax evasion. However, the effect in 1999 is stronger compared with 1990 when looking at corruption, but it is stable for tax evasion.

We also investigate each country in our dataset (EQ15–30) separately. If the opportunities theory holds, gender differences might be less pronounced in countries where women have established greater equality (e.g., stronger labor-force participation or stronger involvement in the political process). However, Table 4 reveals a similar gender effect across countries. Regional differences are detected only when we look at the quantitative effects. Looking at corruption, all coefficients are statistically significant with marginal effects between 4.1 (Spain) and 8.8 (The Netherlands). Similar results are observed for tax evasion. The marginal effects vary between 0.4 (Italy) and 13.6 percentage points (Denmark). Surprisingly, the results from social democratic states such as Denmark and The Netherlands reveal large gender differences. It is Italy, a country from the south with a certain history of patronage, where we find relatively small gender effects. This also seems to indicate that opportunity factors are less relevant.

Finally, in the next two tables we conduct further robustness tests. In EQ35–42, we add four additional variables in the specification: political interest, trust in the legal system, trust in the parliament, and national pride. Politically interested citizens will associate with one another and engage in discussion. Exchange of arguments and face-to-face interaction enhance group identification and give citizens the opportunity to identify others' preferences. As others' preferences become visible, the moral costs of free-riding or behaving illegally increase, reducing the justifiability of corruption and tax evasion. Our two trust variables allow us to analyze trust at the constitutional level (e.g., trust in the legal system) and trust in the current political process (trust in the parliament), thereby focusing on how the relationship between the state and its citizens is established. If the state is seen to be acting in a trustworthy way, individuals' willingness to comply may be higher. The relationship between them and the state (relational contract) can be maintained by positive actions, well-functioning institutions, and implementing a positive social capital atmosphere. Such a strategy will be honored with a higher tax morale. Scholz and Lubell (1998), for example, found that if American taxpayers trusted government or other citizens, they were more likely to comply with their tax obligations than taxpayers who did not trust. Thus, trust influences citizens' incentives to commit themselves to obedience. Identification with the state may induce cooperation within a society or a group and thus induces similar mechanisms as the trust variables. Tyler (2000) argues that pride influences people's behavior in groups, organizations, and societies. It gives a basis for encouraging cooperative behavior. However, contrary to the trust variables, which have been thoroughly analyzed by social capital researchers, the variable pride has been completely neglected although it is a widespread phenomenon (Boulding 1992). The results indicate that national pride strongly reduces the justifiability of illegal activities. However, trust in the legal system and the parliament is only relevant for the justifiability of tax evasion and interest in politics only for the justifiability of corruption (without being robust).

We also differentiate between probit (EQ35, 37, 39, and 41) and ordered probit models (EQ36, 38, 40, and 42) and explore the subsample 18–65 separately (EQ37–38 and EQ41–42) as we may observe stronger self-reported biases and measurement error problems among very young or very old individuals.¹⁰ In all the eight regressions reported in Tables 5 and 6, the coefficient for women is statistically significant with marginal effects between 4.8 and 5.9 percentage points.

In summary, our results are in line with previous studies that report strong gender differences. However, we find little support for the opportunities theory that gender differences in terms of attitudes toward illicit activities would decline if men and women face more similar opportunities to be involved in such activities.

V. CONCLUDING REMARKS

This empirical study uses the WVS and the EVS data covering eight Western European countries spanning the period from 1981 to 1999 to shed some light on the extent to which citizens perceive corruption and tax evasion as a justifiable phenomenon. The major goals of this article are to investigate whether gender matters and whether greater equality of status and opportunities reduces gender differences. We find evidence for strong gender differences. Women are significantly less likely to agree that corruption and cheating on taxes can be justified. The results remain robust after investigating different time periods and extending the specification with several opportunity factors such as education, employment status, or income.

We focused on a relatively stable region but it would be interesting to analyze developing and transition countries, where women have experienced more dynamic changes during our investigated period of 20 yr. The strong economic, social, and cultural changes in these regions lead to new opportunities and a new role for women in society. For example, in Latin America 33 million women joined the labor force between 1990 and 2004 (Abramo and Valenzuela 2005, p. 373). Future studies could also use information on the sector of employment. In many European countries, women are overrepresented in the public sector which might influence their faith in the government system.¹¹

The results have interesting political implications. Increasing the number of women in the government or the public administration may help reduce the level of corruption with important benefits for society. However, such recommendation should be treated with caution. Although we perform robustness tests it is still possible that other factors are causing the gender differences. Moreover, the limited number of studies on corruption provide a somewhat mixed picture and more evidence is required to formulate a solid policy recommendation.