Gender and Corruption in Business

Abstract Are firms with female owners or managers less corrupt than other firms? We test this question using firm-level data on corruption, ownership, and management. We find that women in positions of influence are associated with less corruption: female owners are associated with a lower incidence of bribery and report smaller levels of bribery. Moreover, corruption is seen as less of an obstacle in companies where women are represented in top management. By providing evidence that women are associated with lower levels of corruption in business our research contributes to the literature on development, gender equality, and corruption.


Introduction
Are firms with female owners or managers less corrupt than other firms? Many studies have identified a significant association between women and lower levels of corruption in different social contexts. This basic pattern has been observed across different time periods, indicators of corruption, and a variety of micro-and macro-data (Dollar, Fisman, & Gatti, 2001;Rivas, 2013). However, recent studies using experimental approaches have presented mixed evidence, arguing that if gender has an effect on corruption it may depend on institutional and cultural contexts (Armantier & Boly, 2011;Frank, Lambsdorff, & Boehm, 2011). We contribute to the literature by exploring the association between gender and corruption in business using firm-level data from the World Bank's Enterprise Surveys, a series of global surveys that contain data on firms' experience of corruption. The surveys record the gender of owners and top managers, allowing us to contribute to an emerging literature which explores how the 'tone at the top' of organisations is important for corrupt outcomes (d 'Adda, Darai, & Weber, 2014;Lambsdorff, 2015). We find that women are associated with less corruption: female-owned businesses are associated with less bribes and a lower incidence of bribery. Moreover, female managers are associated with a reduction in the extent to which corruption is seen as an obstacle to the operations of the company.
Corruption remains a persistent problem in both developed and developing countries and women are still under-represented in senior management, even in countries that have achieved high levels of development and gender equality. 1 There has been a strong push within the NGO sector and by global organisations like the World Bank and the United Nations to support gender equality as a way of work can be differentiated from theirs as we focus on bribery within the business world, which differs considerably to bribery by individuals who wish to gain access to public services. Corrupt firms may use bribery to gain competitive advantages in the market place, as well as to gain access to public services.
Second, researchers have used field and lab experiments to explore the relationship between gender and bribery. Unlike many macro studies, experimental work has tended to focus on corrupt transactions rather than perceptions. These studies have also yielded mixed evidence about the effect of gender on corruption. Frank et al. (2011) report that women are not necessarily more honest or averse to corruption in the lab or the field but are less likely to engage in corrupt behaviour if there is a high risk of being caught. Armantier and Boly (2011) present a similar result in their experimental analysis of exam-grading in Burkina Faso. They find that women accept bribes more than men when they are not monitored but otherwise follow similar patterns to men. Alhassan-Alolo (2007) reports no gender difference in terms of condoning gift acceptance by public officials in Ghana's passport office. Alatas, Cameron, Chaudhuri, Erkal, and Gangadharan (2009) observe that women in Australia are less tolerant of corruption, but that there are no significant gender differences in India, Indonesia, and Singapore. Rivas (2013) explores the role of women as both bribers and bribees using an experiment where participants take the role of a firm or a public official. She finds that both the frequency of bribes and the amount offered are higher if a participant is male and they are assigned to a firm. Rivas (2013) concludes that women offer a bribe less frequently than men but even when women do, the amount is still lower than when males offer a bribe. Each of the experimental studies that we review recognises that studying corruption is challenging because it is very difficult to mimic the real world. Indeed, experimental methods have become increasingly sophisticated to meet this challenge as many studies now consider both the opportunity to engage in bribery and the chance of being caught. Ultimately, however, the subjects in most studies are participating in a game where bribery may appear more acceptable than in real life, and the consequences cannot be approximated: no one will be sent to jail or lose their jobs. 3 In short, further evidence from the field is necessary to complement and corroborate experimental studies.
To summarise, the literature is split between macro studies that usually report that women are associated with lower levels of perceived corruption and studies that use field and lab experiments, which sometimes find mixed evidence. Despite some contradictory findings, these studies contain important lessons about how to deal with the theoretical and methodological challenges inherent in the study of gender and corruption. The first lesson is that we must be cautious about whether our data are really measuring corruption, as perception and experience-based indicators are very different (Razafindrakoto & Roubaud, 2010;Treisman, 2007). The second lesson is that corruption occurs at the individual level, so we need to be careful about drawing inferences from macro-level indicators. The third lesson is that we must consider external validity: field and lab experiments take place within a wide array of cultural and institutional contexts, which may explain the diverse findings. More carefully designed cross country studies based on reliable micro-data can complement existing work using lab and field experiments.

Corruption in business: how gender matters
Both conventional wisdom and previous research suggest that individuals at the top of the corporate hierarchy have the power to affect corruption (Clarke & Xu, 2002). If top managers support bribery or do not treat the issue seriously, by using policies and procedures to detect, eliminate and sanction illegal activities, this may foster a culture of corruption within an organisation. According to Lambsdorff (2015), the 'tone at the top' is perhaps the most important factor in fighting corruption. Experimental evidence from a recent study suggests that dishonest behaviour by leaders can induce cheating behaviour in subordinates (d' Adda et al., 2014). Indeed, corrupt activities are widely tolerated in many contexts and are sometimes considered mandatory for doing business. One study, for example, found that top executives who have engaged in corruption tend to rationalise their actions as a necessity for being competitive (Collins, Uhlenbruck, & Rodriguez, 2009). Nevertheless, by definition, corruption is never a legitimate act, no matter how widely tolerated.
There are several potential channels through which gender may affect corruption. One channel is through its effect on other employees' expectations. For example, employees that work in companies where women are better represented in upper management may determine that corruption is generally less tolerated within the firm. Another channel is by favouring business strategies that rely on less corruption or focusing business activities in areas where corruption is less prevalent. However, these channels assume that women have a preference for less corruption and that they always tend to support business strategies that are in line with their preferences. Both preference formation and the link between preferences and outcome are not straightforward. Indeed, there is a controversial debate about whether gender differences are socially constructed, biologically determined, or some combination of both (Gilligan, 1982;Kohlberg, 1969).
In line with many of these studies, we expect to find that female business owners and female top managers are associated with smaller bribes, a lower incidence of bribery, and a more optimistic outlook regarding the effect of corruption on doing business. However, there are circumstances where the relationship may not hold. For example, activities that are normally considered corrupt may not be considered corrupt in some contexts. Previous work has suggested that institutional and cultural arrangements may activate the relationship between gender and corruption (Esarey & Chirillo, 2013;Truex, 2011). Corrupt activities like bribery might seem virtuous if they hasten the process of vital business activities like registering property or obtaining permits. At the same time, not engaging in corrupt activities might be seen as a risky strategy. For example, failing to pay a bribe to a corrupt politician or official might be seen to threaten the dominant regime and provoke retaliation. For this reason, our empirical models examine the relationship between female business leaders and corruption not only in the entire sample but also in specific institutional or cultural subsets of countries. 4

Data and method
Like many secretive activities, corruption is difficult to quantify. However, it is possible to obtain reliable data from well-designed surveys and appropriate interview techniques (Reinikka & Svensson, 2006, p. 8). Our main dependent variable comes from the World Bank's Enterprise Surveys (ES; http:// www.enterprisesurveys.org/) and is based on a survey question designed to capture a firm's total annual informal payment. The ES, which began in 2002, are representative firm-level surveys that are carefully designed and implemented according to the recommendations in the literature. Our dataset contains observations on over 105 countries, though our sample size depends primarily on our choice of dependent and independent variables. 5 The dependent variable (Bribe) is derived from the following survey question: We've heard that establishments are sometimes required to make gifts or informal payments to public officials to 'get things done' with regard to customs, taxes, licences, regulations, services and so forth. On average, what per cent of total annual sales, or estimated total annual value, do establishments like this one pay in informal payments or gifts to public officials for this purpose?

Gender and corruption in business 1489
The literature on corruption and survey design suggests that it is better to ask this type of question than one that seeks information about the respondents' own company (Reinikka & Svensson, 2006). While some respondents may take the question at 'face value' and report the amount of bribes they believe similar establishments pay, many will provide the amount paid by their own company. The strategic placement of the question after a battery of similar questions that seek information about the respondents' own company helps to induce this type of response. Moreover, the phrase 'get things done' and 'establishments like this one' help to reduce the respondents' anxiety about whether they are implicated in any wrongdoing. Furthermore, recent studies have found that women are no more likely to lie about corruption than men (Azfar & Murrell, 2009;Clausen, Kraay, & Murrell, 2010). These studies find that reticent respondents are more likely to lie than non-reticent respondents but that women are no more likely to be reticent than men. Therefore, the difference between male and female respondents is probably not due to women lying more frequently. 6 In brief, this ES question was designed carefully to elicit a truthful response; however, we acknowledge the potential for measurement error.
From this question, we use the total annual informal payment as our dependent variable. 7 While the ES data also contain the information of informal payments as percentage of total annual sales, we decided not to use it because of the possibility of measurement error; too many non-zero responses to the survey question are in multiples of five, suggesting that respondents may have been rounding off their estimates. Clarke (2011) reports that firm managers overestimate bribes when they report them in percentage terms. Furthermore, we drop 13 observations of firms that reported bribes in excess of one million dollars. The statistical significance of our main results is not substantively affected when we include these observations, however, they considerably inflate the estimated relationships in most of our samples.
Additionally, we utilise another measure of corruption from the ES data. Our second dependent variable (Obstacle) is derived from a survey question that asks 'Is [Corruption] No Obstacle, a Minor Obstacle, a Major Obstacle, or a Very Severe Obstacle to the current operations of this establishment?' The question should capture general perceptions regarding corruption but it may also capture individuals' lived experience of real corruption in business, such as having to source inputs from connected suppliers. From the answers, we create a dummy variable which takes a value of one if the firm feels that corruption is a major or very severe obstacle and zero otherwise. This variable allows us to extend our study to consider firm-level perceptions regarding corruption, which may differ considerably to total bribes paid.
Our explanatory variables of interest come from questions which record gender, including (i) 'are any of the owners female?' and (ii) 'is the top manager female?' Note that the World Bank's manual defines the top manager as the 'highest management individual', and that this 'person may also be the owner if he/she works as the Manager of the firm'.
We control for several important firm characteristics which may potentially affect corruption. According to Rand and Tarp (2012), the incidence of bribery is associated with firm-level differences in visibility, sunk costs, ability to pay, and level of interaction with public officials. The World Bank data allow us to control for several similar or closely related variables. First, we control for the degree of foreign ownership. Foreign-owned firms may not have access to the same social networks as domestic firms, and may have to pay more bribes to do business. Conversely, foreign firms may be less prone to bribes because of a risk that they can be punished for corrupt activities in their home countries as well as the host country in some cases. Second, we use a dummy variable that takes a value of one if some of the firm's sales are not national sales, allowing us to control for the possibility that export-oriented firms may come into contact with a greater variety of public officials. Third, we control for firm size using the natural logarithm of sales (see Fan et al., 2009). Fourth, we include a variable which measures the extent of state ownership (see Billon & Gillanders, 2016;Fan et al., 2009). 8 Like our foreign ownership variable, state-owned companies have access to different social and political networks, potentially affecting the level of bribery. Finally, we control for GDP per capita in the firm's home country, in order to control for likelihood that the level of development contributes to perceptions of corruption (see for example, Ades & Di Tella, 1999;Svensson, 2005) and the possibility that higher income countries may be less prone to tolerating corruption. The average firm in our dataset pays 2746 dollars in informal payments per year though this varies substantially within and between global regions. The average firm with a female owner pays 2179 dollars and the average firm with a female top manager pays approximately 1349 dollars. On face value, these figures suggest that companies where women are in positions of influence pay less in bribes. However, the descriptive statistics mask the important sources of variation in corruption which we described above, as well as the fact that approximately 80 per cent of firms report zero informal payments. This means that Bribe is a left-censored continuous variable, posing a challenge to standard linear models such as OLS. 9 For this reason, when examining the level of bribery we use the tobit estimator, which can model a linear relationship with a censored dependent variable. Recent firm-level studies that consider corruption have also used a similar empirical strategy (Diaby & Sylwester, 2015;Jagger & Shively, 2015). 10 When studying the binary Obstacle outcome, we employ probit models. In addition, we include dummies for industry type, as some industries may be more likely to engage in (or be targeted for) bribery. For the same reason, we cluster our standard errors by industry and country groups, allowing errors to be correlated within industry-country groups.

Main results
The main findings from our econometric analysis are presented in Table 2. Columns 1-3 report tobit coefficients for our main dependent variable of interest (Bribe). The first column includes a femaleowner dummy variable; the second column includes a female top manager dummy variable; the third column includes both owner and manager variables, and columns 4-6 repeat these specifications for our second dependent variable (Obstacle). 11 With regard to our first dependent variable, we find that female owners, but not female managers, are significantly associated with smaller bribes. This relationship holds even when both variables are added to the same specification. 12 Moreover, the effect of female ownership is substantial: the presence of a female owner is associated with 6785 dollars less in bribes on average, a very substantial sum when one considers that the mean bribe in our sample is just 2746 dollars (standard deviation is 25,760 dollars). For our second dependent variable, we find that female managers, but not female owners, are significantly associated with a reduction in the probability of holding the perception that corruption is an obstacle to the operations of the company (approximately 3.6% lower). This relationship also holds even when both variables are added to the same specification. In this specification, female owners also become statistically significant at the 0.1 level. Gender and corruption in business 1491 Notes: Columns 1-3 report tobit coefficients. Columns 4-6 report probit marginal effects. Standard errors are clustered at the country-sector level and are reported in parentheses. *, ** and *** indicate significance at the 10 per cent, 5 per cent and 1 per cent levels respectively.
From these results, we conclude that there is indeed a relationship between gender and corruption in business, but that it varies depending on the specific aspect of corruption that we focus on, and on the specific business role. For example, the top manager, as one of the most visible members of any company, might create the perception of clean operations among other members in the company, while owners may not always take an active role in day-to-day operations. On the other hand, the owner may have the power to constrain bribery, while the top manager may not have as much of an effect unless she is also the owner, as top managers may themselves be constrained by opposing groups within the corporate structure.

Alternative dependent variable, corruption perceptions, and respondents' gender
In Table 3, we test our results using alternative measures of corruption and additional control variables. As self-reported information on the extent of bribery may be subject to measurement error, columns 1 and 2 report estimates of the effect of gender on whether any bribe has been paid or not. Our main finding is robust to this test: female ownership reduces the probability of paying a bribe by 4 per cent. The next test, presented in columns 3 and 4, introduces Obstacles as an additional control variable. This allows us to control for corruption perceptions, which may potentially affect the level of bribery. Again, our main results hold even when this variable is included. 13 In the final test, we add the gender of the ES survey respondent to our base model. Columns 5 and 6 show that female ownership remains a statistically significant determinant of Bribe, even controlling for the gender of the respondent. However, our finding regarding the importance of female management for Obstacle does not pass this test. All else equal, female respondents to the survey were associated with a reduction in the likelihood of having the perception that corruption was an obstacle to the operations of the company, regardless of whether there was a woman owner or top manager. This result does not necessarily mean that women are more likely to hide corruption than men, and therefore does not necessarily contradict recent studies which have found that women are no more likely to lie about corruption than men (Azfar & Murrell, 2009;Clausen et al., 2010). Rather, since the respondents were usually at a management level, if not necessarily the owners or top managers themselves, the statistical significance of female respondents suggests that companies which have women at a management level are less likely to view corruption as an impediment. That said, since the variable causes us to lose approximately half of our sample we do not include it in other specifications. Sung (2003) has argued that the association between women in government and reduced corruption is spurious and caused mainly by context, namely liberal democracy. In Table 4, we introduce additional country-level control variables that address this concern. Following Fan et al. (2009) we include the quality of institutions, trade openness, democracy, and the importance of natural resources. Our reason for including these variables is that low quality institutions, trade restrictions, and non-democratic regimes give rise to opportunities for rent seeking, as does an abundance of natural resources. These factors will also likely correlate with the role of women in the economy.

Gender and corruption: country-level controls
We use the Rule of Law from the World Governance Indicators 2014 (Kaufmann, Kraay, & Mastruzzi, 2011) as a proxy for quality of institutions. Trade openness is measured as the sum of exports and imports of goods and services as a percentage of GDP, whose data are from the World Development Indicators (databank.worldbank.org/wdi). Democracy is captured by the Polity index (1 if Polity2 ≥ 6, 0 otherwise) (http://www.systemicpeace.org/polity/polity4.htm). Each column introduces one country-level variable and the final column includes all of them. Table 4 shows that our main result on the association between gender and bribery is robust to the inclusion of these variables. Female ownership is negatively associated with bribery levels but the gender of the top manager does not seem to matter. Table 5 repeats the specifications using Obstacles as the dependent variable. Again, our findings are broadly consistent with what we have seen in earlier tests; however, the association between female top managers and Obstacles disappears in the full model (in column 10). In light of these findings, and the earlier tests we Gender and corruption in business 1493 Table 3. Gender and corruption: alternative dependent variable, attitudes toward corruption, and respondents' gender Notes: Columns 1-2 and 7-8 report probit marginal effects. Columns 3-6 report tobit coefficients. Standard errors are clustered at the country-sector level and are reported in parentheses. *, ** and *** indicate significance at the 10 per cent, 5 per cent and 1 per cent levels respectively.   Notes: Columns report tobit coefficients. Standard errors are clustered at the country-sector level and are reported in parentheses. *, ** and *** indicate significance at the 10 per cent, 5 per cent and 1 per cent levels respectively. Table 5. Gender and obstacles to corruption: country-level controls Notes: Columns report probit marginal effects. Standard errors are clustered at the country-sector level and are reported in parentheses. *, ** and *** indicate significance at the 10 per cent, 5 per cent and 1 per cent levels respectively. conducted, the association between gender and bribery appears to be robust. The association between gender and corruption perceptions is less so, and is particularly sensitive to the inclusion of country-level controls and the gender of the survey respondent.
6. Sample splits 6.1. Firm size, ownership, and industry type The previous section gives us confidence that, in general, there is a meaningful relationship between gender and corruption at the firm-level. In this section, we ask whether this relationship holds in subsamples of firms defined by three variables: firm size, ownership structure, and industry type. Dividing the sample along these lines helps us to understand how the association between gender and corruption varies within the business world, thus informing policy initiatives aimed at addressing corruption in this context. 14 Table 6 presents our findings from the sub-samples. Estimates are reported separately for female owners and managers in order to maximise sample size. The first three columns present estimates where the sample is divided by firm size; columns four and five present ownership type, and the remaining columns divide the sample by industry type. 15 We find that female ownership but not top management is associated with a significant reduction in bribery in many sub-samples. In particular, it is associated with a significant reduction in small-and medium-sized firms although the association is statistically significant at the 10 per cent level in the sample of large firms. While the magnitude of the effect increases with firm size, this is good news for existing policy initiatives, like microfinance schemes, that are targeted at women in the small business sector. Our findings suggest that such interventions may help to promote enterprises that are relatively less corrupt.
There is more evidence that the effect of gender varies across the business world: female owners are associated with a reduction in bribery in the manufacturing sector. This association is weaker but also present at the 10 per cent level in the services sector. However, gender is not associated with less bribery in either foreign or state-owned companies. This result means that the negative association between female owners and bribery is experienced mainly by domestic or private firms. Managers and owners of state and foreign owned enterprises probably differ from their counterparts in many regards and have a different relationship to government and its officials. Further results where we use Obstacles as the dependent variable, and where we split the sample by a range of cultural and institutional variables are available in the Online Appendix.

Conclusion
Many supporters of gender equality assert that it has the potential to reduce corruption. Our findings lend some empirical support to this assertion: firm-level evidence from the World Bank shows that female owners are associated with a lower incidence of bribery and report smaller levels of bribery. In addition, firms where women are represented in top management tend to see corruption as less of an obstacle to doing business, although this finding is not significant in the models where survey respondents' gender is included. Reassuringly, all of our findings are robust to the determinants of corruption at the firm-level, including state ownership, foreign ownership, firm size, and exporter status. Our findings are also robust to a range of country-level variables and alternative measures of our key concepts.
Taken together, our findings help to shed light on a longstanding puzzle in social science about whether women are less corrupt than men. We suggest that women in leadership roles in business are indeed less corrupt. However, our study also helps us to better understand where the effect is greatest. In particular, we found that female ownership is associated with a significant reduction in bribery in the manufacturing sector, and that the effect was apparent even in small-and medium-sized firms. These findings can help us to better understand the costs and benefits of existing policy initiatives that are targeted at women in developing countries. Finally, it is worth reiterating the point that the negative Notes: Estimates from tobit regression. Control variables (same as Table 2) are not displayed, and sector dummies are dropped for the firm size and ownership subsets to avoid estimation failure. Standard errors are clustered at the country-sector level and are reported in parentheses. *, ** and *** indicate significance at the 10 per cent, 5 per cent and 1 per cent levels respectively.
consequences of corruption are widespread and extend much further than the economy (Brown, Touchton, & Whitford, 2011). Efforts to secure gender equality in business may help us to fight corruption as well as advance human development more generally.