Gender and corruption: The neglected role of culture

Debski, Julia; Jetter, Michael; Mösle, Saskia; Stadelmann, David Article — Accepted Manuscript (Postprint) Gender and corruption: The neglected role of culture European Journal of Political Economy Provided in Cooperation with: Kiel Institute for the World Economy – Leibniz Center for Research on Global Economic Challenges Suggested Citation: Debski, Julia; Jetter, Michael; Mösle, Saskia; Stadelmann, David (2018) : Gender and corruption: The neglected role of culture, European Journal of Political Economy, ISSN 0176-2680, Elsevier, Amsterdam, Vol. 55, pp. 526-537, https://doi.org/10.1016/j.ejpoleco.2018.05.002 This Version is available at: https://hdl.handle.net/10419/233994 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by-nc-nd/4.0/deed.de Accepted Manuscript Gender and corruption: The neglected role of culture Julia Debski, Michael Jetter, Saskia Mösle, David Stadelmann PII: S0176-2680(16)30194-X DOI: 10.1016/j.ejpoleco.2018.05.002 Reference: POLECO 1715 To appear in: European Journal of Political Economy Received Date: 25 September 2016 Revised Date: 4 May 2018 Accepted Date: 7 May 2018 Please cite this article as: Debski, J., Jetter, M., Mösle, S., Stadelmann, D., Gender and corruption: The neglected role of culture, European Journal of Political Economy (2018), doi: 10.1016/ j.ejpoleco.2018.05.002. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. © 2018. This manuscript is made available under the CC BY-NC-ND 4.0 license. This is the accepted version of an article in European Journal of Political Economy 55 (2018), pp.526-537, available online at: https://doi.org/10.1016/j.ejpoleco .2018.05.002 MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 1 Gender and Corruption: The Neglected Role of Culture ∗ Julia Debski a Michael Jetter b,c,d Saskia Mösle a David Stadelmann a,e Mai 2018 Abstract: Empirical findings of a negative association between female participation in politics and the labor market, and levels of corruption have received great attention. We reproduce this correlation for 177 countries from 1998 to 2014. However, once taking account of countryspecific heterogeneity via fixed effects, these negative associations disappear, both in terms of statistical significance and magnitude. This suggests that female participation rates in politics and the labor market are not directly linked to lower corruption. Exploiting countryspecific dimensions of culture, we then present evidence from pooled estimations suggesting that power distance and masculinity are systematically associated with both corruption and female participation rates. In fact, these two cultural characteristics are sufficient to fully explain the link between gender and corruption. Therefore, culture is an important dimension to consider when analyzing the relationship between female participation in society and corruption since the omission of cultural characteristics can produce a spurious correlation between increased female participation rates alone and reduced corruption levels. Keywords: corruption, culture, development, female participation, gender, masculinity, power distance. JEL Classification: J16, D73, Z10. ∗ Acknowledgements: We would like to thank Reiner Eichenberger, Arye Hillman, Marco Portmann, Niklas Potrafke, and the participants of the 2016 Annual Meeting of the European Public Choice Society for illuminating discussions and highly constructive remarks on an earlier version of this paper. We are also grateful to Toke Aidt, the editor in charge of this paper, for helpful suggestions to improve the paper. Corresponding author: Michael Jetter, [email protected] a University of Bayreuth, Universitätsstraße 30, Bayreuth, Germany. b University of Western Australia, 35 Stirling Highway, Crawley, WA 6009, Australia. c IZA Research Fellow – Institute for the Study of Labor, Bonn, Germany. d CESifo Research Affiliate – Center for Economic Studies, Munich, Germany. e CREMA – Center for Research in Economics, Management, and the Arts, Zurich, Switzerland. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 2 Women aren’t better men. They only had fewer opportunities to get their hands dirty. (Alice Schwarzer, author and feminist Frankfurter Allgemeine Zeitung, April 2008, translated from German) I. I NTRODUCTION Over the past decades, governments and international organizations around the world have implemented policies aimed at increasing the share of women in politics and the labor market. In 2015, the UN member states renewed their commitment to “achieve gender equality and empower all women and girls” by adopting the Sustainable Development Goals (United Nations 2015). The Copenhagen Consensus Center (2015) estimates that the benefits per US-Dollar spent on increasing women’s political representation are “likely to be high”. Similarly, the McKinsey Global Institute (2015) highlights the importance of integrating women into the labor market, suggesting potential gains in terms of annual GDP of $28 trillion in 2025, equivalent to 26 percent of global output. Achieving gender equality is an important goal in itself and desirable to achieve for a numerous reasons. One particular positive and appealing effect of increased female participation in society has been linked to corruption. Two influential studies suggest that women, on average, seem to be less corrupt than their male counterparts (Dollar, Fisman, and Gatti 2001; Swamy et al. 2001). This hypothesis, buttressed by the associated research findings from cross-country correlations, has influenced governments to bring more women into public offices. 1 At the same time, the view of attributing women a positive role in lowering overall corruption levels found a 1 The Peruvian government decided to replace a part of the male traffic police of Lima by a new team consisting exclusively of female officer and a women-only traffic force was established by the chief police officer of Mexico City (Goetz 2007, Anozie et al. 2004). Similarly, but to a lesser extent, El Salvador, Panama, Ecuador, and Bolivia incorporated women in their transit divisions (Karim 2011). The idea of feminizing important decision-makers to fight corruption was also put into practice in Uganda, where president Museveni assigned the majority of treasury positions in the new local government system to women (Goetz 2007). MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 3 lively response in academic research. 2 However, existing studies usually rely on cross-country evidence, as repeated country-level information on both female representation and corruption has been limited. Ideally, we would like to know whether increasing female participation within a given country decreases corruption levels and to do so we need to observe a set of countries for a certain period of time. 3 As for women’s roles in society throughout the world, we now have substantial evidence of historical and cultural factors playing an important role (Alesina, Giuliano, and Nunn 2013). Specifically, cultural roots have been strongly associated with corruption levels (Fisman and Miguel 2007; Barr and Serra 2010). Thus, the previously conducted cross-country studies on the link between gender and corruption could potentially be traced to country-specific cultural differences. As a first contribution, this article analyzes the relationship between gender and corruption using annual panel data for up to 177 countries from 1998 to 2014. Specifically, we distinguish between the role of women in politics and in the labor force when analyzing potential links to corruption. We begin by replicating the key result of Dollar, Fisman, and Gatti (2001) and Swamy et al. (2001), which suggests that higher female participation in parliament and the labor force is associated with less corruption. We then introduce country fixed-effects to generally account for time-invariant unobserved heterogeneity on the country level that may affect the link between gender and corruption, such as cultural, geographical, and institutional factors. Female participation in society has substantially risen over the sample period and the change in participation varies notably across countries and over time. The corresponding empirical results from fixed-effects regressions differ systematically from the existing evidence derived from pure 2 This response is also reflected by the numerous citations the original papers receive. 3 Unobserved heterogeneity between countries has been shown to matter. For instance, determinants of economic growth (Islam 1995), democracy (Acemoglu et al. 2008), and government size (Ram 2009; Jetter and Parmeter, 2015) change fundamentally once country-specific, time-constant characteristics are taken into account. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 4 cross-country studies: Changes in the share of women in parliament (WIP) or the female labor force participation (FLFP) within a country do not to predict within-country changes in corruption levels. In fact, we find relatively precisely estimated zero effects for the link between gender and corruption once we account for country-fixed effects. 4 As a second contribution, we then focus on the interplay between culture, gender, and corruption. To test the hypothesis that distinct cultural factors could simultaneously drive the level of female representation in society and corruption, thereby causing a spurious correlation in the cross section, we turn to Hofstede’s (1980, 2001, 2011) cultural dimensions. We identify two cultural factors that mediate the link between gender and corruption: Power distance and masculinity. Once we control for these cultural dimensions in a pooled setting with regionand time-fixed effects (since cultural variables are only available once per country), the relationship between female participation in the public sphere and corruption vanishes entirely. These findings are confirmed when instrumenting power distance with genetic distance from the United Kingdom (Gorodnichenko and Roland, 2011, 2016) to alleviate potential endogeneity concerns. 5 In sum, disregarding the role of distinct cultural attributes can lead to biased results and the idea that increasing female participation per se could be an effective means to directly alleviate corruption levels. Additional robustness tests with different corruption measures substantiate these results. The paper proceeds as follows: Section II reviews the literature and presents theoretical considerations on the interplay between gender, culture, and corruption. Section III describes our data and empirical methodology. In Section IV, we present our main findings, address endogeneity concerns, and discuss a range of robustness tests. Section V concludes. 4 The precisely estimated zero effects of the effect of gender on corruption imply a high power for our statistical tests which is relevant as our first result is a negative one. 5 To measure genetic distance from the United Kingdom, Gorodnichenko and Roland (2018, p. 403) “use a measure of genetic distance between the population in a given country and the population in the United Kingdom.” They further propose that “measures of genetic distance can be seen as a proxy measure of differences in cultural values.” MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 5 II. L ITERATURE AND THEORETICAL CONSIDERATIONS Two influential cross-country studies suggest that women might be less prone to corruption than men. Dollar, Fisman, and Gatti (2001) find a negative and statistically significant association between the share of women in parliament and the level of corruption. Swamy et al. (2001) find the same association for female representation in senior positions in government bureaucracy and in the labor force. These results are confirmed when using more recent data, as shown by Watson and Moreland (2014). 6 Torgler and Valev (2010) analyze compliance data from the World and European Values Survey for Western European countries and find that women are less tolerant of corruption. Taken literally, these results suggest policy initiatives aimed at increasing the number of women in public life will alleviate corruption levels. Similar results are suggested by experimental studies examining gender differences towards corruption (e.g., Schulze and Frank 2003; Frank, Lambsdorff, and Boehm 2011; Chaudhuri 2012; Rivas 2013; Barnes and Beaulieu 2014). Yet, while laboratory studies help us to better understand certain behavioral mechanisms, results need to be interpreted with caution, especially with respect to women in public office: Politicians are a specific, self-selected group (Ruske 2015; Kauder and Potrafke 2016) that may not necessarily be comparable to participants in laboratory studies. Recently, a growing body of research casts doubt on a causal effect of female representation on corruption. Some studies question the direction of causality and argue that male-dominated patronage networks make it more difficult for women to enter politics and engage in corrupt practices in the first place (e.g., Alhassan-Alolo 2007; Goetz 2007; Stockemer 2011; Sundström and Wängnerud 2014). Others doubt the general existence of a universal link between gender and corruption, as well as gender and governance (Branisa and Ziegler 2011; Stadelmann et al. 2014). Research has also focused on potentially mediating factors. Sung (2003) proposes a “fairer 6 Related to that, numerous studies (e.g., Glover et al. 1997; Eckel and Grossman 1998; Grove et al. 2011; Hicks et al. 2016) suggest women to be more trustworthy, less opportunistic, and more public-spirited than men. May et al. (2018) point to systematic differences between economic policy views held by male and female economists. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 6 system” rather than “fairer sex” hypothesis. 7 Further cross-country evidence suggests the relationship to be specific to democracies (Esarey and Chirillo 2013) and particularly present when electoral accountability is high (Esarey and Schwindt-Bayer 2017). In this context, studies aiming to cope with the problem of unobserved heterogeneity across countries are scarce (e.g., Sung 2012). We contribute to this literature by conducting an extensive panel data analysis of the potential link between the participation of women in parliament and in the labor market, and corruption, accounting for country fixed-effects. This allows us to control for country-specific time-invariant characteristics, isolating a potential link between female participation in society and corruption in a much cleaner fashion. We then advance existing research by proposing an alternative explanation for the crosscountry association between gender and corruption: Cultural heterogeneity. Hinting at this idea, Alatas et al. (2009) provide evidence that gender differences in corruptibility might be culturespecific, using an experimental setting. Employing cultural characteristics on the country level, we turn to Hofstede’s conceptualization of culture (Hofstede 1980, 2001, 2011; Hofstede, Hofstede, and Minkov 2010). Hofstede (2011: 3) defines culture as “the collective programming of the mind that distinguishes the members of one group or category of people from others.” Initially, his research on differences in cultural values was based on employee surveys collected within IBM in about 40 countries between 1967 and 1973, but by now the cultural concept covers more than 100 countries. Applying factor analysis, Hofstede identifies four core dimensions of culture: 1. First, power distance measures the degree to which less powerful members of society accept and expect that power is distributed unequally. 2. Second, uncertainty avoidance expresses a society’s tolerance when it comes to ambiguity and uncertainty. 8 7 Sung’s (2003) results suggest that the correlation between gender and corruption shrinks and loses statistical significance once measures of liberal democracy are accounted for. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 7 3. Third, the individualism vs. collectivism dimension measures the degree of interdependence between the members of society. 9 4. Forth, the masculinity vs. femininity dimension refers to the social preference for male values generating a more competitive social environment as opposed to female values that are more consensus-oriented. 10 This conceptualization has been also used in the literature, including research on corruption (e.g., Husted 1999; Park 2003; Sanyal 2005; Seleim and Bontis 2009; Gorodnichenko and Roland 2011). For example, Yeganeh (2014) provides evidence that high levels of power distance, uncertainty avoidance, collectivism, and masculinity tend to promote corruption. Getz and Volkema (2001) argue that the clear separation between socioeconomic classes in countries with high power distance increases the likelihood of corrupt behavior. In highly masculine societies, Sanyal (2005: 144) hypothesizes that “an aggressive pursuit of success and achievement appears to accompany corrupt conduct”. Several studies address Hofstede’s cultural dimensions in the context of gender equality (e.g., Luthar and Luthar 2002; Cheung and Chan 2007; Parboteeah, Hoegl, and Cullen 2008). Following Hofstede’s (2001) definition of masculinity, Cheung and Chan (2007) argue that the gender gap is smaller in low-masculinity countries, gender roles are more progressive, and consequently women are elected into parliament more frequently. Based on the existing literature, we hypothesize that particularly power distance and masculinity play a relevant role in influencing the relationship between female representation and corruption. 8 Uncertainty avoidance is not the same as the economic concept of risk avoidance but rather measures whether the members of a culture feel comfortable in unstructured, unknown, or new situations. 9 In collectivist societies, members are integrated into strong cohesive groups, whereas in individualist societies only loose interpersonal connections exist, and everybody is supposed to care for themselves. 10 While feminine countries are characterized by overlapping social gender roles, “masculinity stands for a society in which social gender roles are clearly distinct” and material success is highly valued (Hofstede 2001: 297). MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 14 Table 1: Female participation and corruption – Zero effect once country fixed-effects are introduced. 1998-2014. Dep. variable: (1) (2) (3) (4) (5) (6) (7) (8) CPI OLS OLS FE FD 14/98 OLS OLS FE FD 14/98 WIP 0.076 *** 0.023 *** 0.003 0.014 (0.016) (0.007) (0.005) (0.015) FLFP 0.015 ** 0.009 ** -0.015 -0.010 (0.007) (0.004) (0.009) (0.016) Ln (GDPp.c.) -3.027 *** 1.286 2.773 * -2.949 *** 1.200 2.785 * (0.426) (1.003) (1.495) (0.486) (0.961) (1.417) Ln (GDPp.c.) 0.247 *** -0.015 -0.073 0.245 *** -0.024 -0.108 Squared (0.026) (0.068) (0.100) (0.030) (0.064) (0.091) Time FE Yes Yes Yes No Yes Yes Yes No Country FE No No Yes No No No Yes No # of observations 2350 2302 2302 74 2443 2389 2389 83 # of countries 181 177 177 74 180 177 177 83 R 2 0.152 0.769 0.147 0.204 0.032 0.765 0.135 0.145 Notes: In columns (1), (2), (6) and (7) OLS estimates are reported. In columns (3) and (8), the fixed effects estimator is applied to the whole panel ranging from 1998 to 2014. In columns (4) and (9), and (5) and (10), respectively, the data are first differenced over the whole panel (2014-1998), and ten years (2014-2004). Robust standard errors are reported in parentheses. They are clustered at the country level in columns (1)-(3) and (6)-(8). * p < 0.10, ** p < 0.05, *** p < 0.01. When focusing on women in the labor force as the explanatory variable of interest, even a negative coefficient emerges (column 7) though of a small magnitude. Again, the coefficient is relatively precisely estimated and the power of the test is substantial compared to any positive benchmark value. In specifications (4) and (8) we additionally look at first differences from the beginning to the end of our sample. When first differencing the data over the whole sample period, the measures of female participation turn statistically insignificant at conventional levels and the magnitude of the implied effect is small; for the female labor force participation it is, if anything, negative. To increase the sample size, we also looked at first differences over the ten-year period from 2004 to 2014 and the interpretation of the results remains unchanged, i.e., we do not find any relevant association, neither in terms of statistical significance nor of economic relevance. These findings are important, especially in the light of possible policy conclusions: Once we control for country-specific heterogeneity, the positive correlation between female participation in society and corruption disappears. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 15 IV.2 Making sense of the evidence: The importance of culture The results from first difference estimates and the panel data analysis show that a fixed effect can explain the association between female representation and decreased corruption levels previously assumed in the literature. This suggests that the relationship between gender and corruption is mediated by a time-invariant source of heterogeneity across countries. Such a factor may influence both a country’s level of corruption and the participation of women in politics and the labor force. Cultural, institutional, or geographical characteristics are the most obvious candidates for time-invariant differences across nations, as they have been shown to influence economic development at a fundamental level. While some of these factors, such as culture and institutions, can change over time, they are likely changing slowly, especially in light of our sample period of 17 years. In the following, we aim to identify the source of time-invariant unobserved heterogeneity across countries. In other words, we want to know which factor captured by the fixed effects is responsible for rendering the relationship between gender and corruption irrelevant. First, with respect to institutions, democracy has been shown to affect the association between gender and corruption (Sung 2003; Esarey and Chirillo 2013). Specifically, we enrich our regression analysis with the Polity IV index (variable polity2). Second, we account for trade openness, region fixed-effects, a binary indicator for landlocked countries, and a tropical climate index to capture differences across countries due to geographic conditions. To give the previous results of the literature the best chance to emerge, we analyze pooled data again over the whole period of time. Third, regarding cultural aspects, we include Hofstede’s four main cultural dimensions, i.e., individualism, masculinity, uncertainty avoidance, and power distance. While culture can change over time, it is a very slow-moving process. In fact, Hofstede (2011: 22) specifically argues that MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 16 cultural change basic enough to invalidate his country scores will only happen over a very long time period of at least 50 to 100 years, or as a result of dramatic outside events. Following this line of argument and using Hofstede’s data, we consider culture within a given country as constant over our sample period and as a potential candidate to explain the fixed effects finding. Table 2: Women, culture and corruption: Power distance and masculinity render the effect of female participation in politics and the labor market statistically insignificant. 1998-2014. (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) OLS OLS OLS OLS OLS OLS OLS OLS OLS OLS WIP 0.029 *** 0.029 *** 0.004 0.017 0.016 (0.007) (0.011) (0.010) (0.011) (0.011) FLFP 0.009 * 0.017 * 0.004 0.012 0.011 (0.005) (0.009) (0.008) (0.010) (0.008) Ln (GDP/cap) -2.430 *** -2.023 ** -1.590 ** -1.683 ** -1.941 *** -2.517 *** -1.975 ** -1.509 ** -1.502 ** -1.785 ** (0.488) (0.872) (0.672) (0.732) (0.669) (0.553) (0.803) (0.707) (0.725) (0.691) Ln (GDP/cap) 2 0.211 *** 0.193 *** 0.166 *** 0.170 *** 0.197 *** 0.220 *** 0.194 *** 0.162 *** 0.161 *** 0.190 *** (0.029) (0.049) (0.039) (0.042) (0.039) (0.033) (0.046) (0.041) (0.041) (0.040) Polity IV 0.048 *** 0.034 * 0.048 *** 0.029 (0.012) (0.020) (0.014) (0.021) Openness 0.003 0.003 0.002 0.003 (0.002) (0.002) (0.002) (0.002) Landlocked 0.084 -0.012 0.055 -0.191 (0.161) (0.277) (0.159) (0.263) Tropical climate -0.363 -0.545 -0.410 -0.515 (0.287) (0.395) (0.312) (0.436) Uncertainty avoidance -0.005 -0.005 (0.005) (0.005) Individualism 0.008 0.008 (0.008) (0.008) Power distance -0.012 ** -0.018 *** -0.012 ** -0.019 *** (0.005) (0.005) (0.005) (0.006) Masculinity -0.020 *** -0.020 *** -0.020 *** -0.021 *** (0.006) (0.005) (0.005) (0.005) Region FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # of observations 1,987 1,317 1,317 1,317 1,317 2,057 1,317 1,317 1,317 1,317 # of countries 152 90 90 90 90 152 90 90 90 90 R 2 0.819 0.812 0.843 0.817 0.827 0.807 0.807 0.843 0.816 0.826 Notes: Dependent variable: CPI. Robust standard errors clustered at the country level are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. Table 2 investigates whether any of these country-specific variables are responsible for the zero-effect estimated in the fixed-effects regressions in Table 1. Columns (1) and (6) begin by replicating our main regression for employing WIP and the FLFP. We then re-estimate the MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 17 respective regression when only employing those observations for which the cultural variables are available (specification 2 and 7). This comparison aims to demonstrate that none of the upcoming results are driven by potential sample selection. While a positive correlation between gender and the absence of corruption holds when including our institutional and geographic variables, the association entirely disappears once cultural variables are introduced in columns (3) – (5) and (8) – (10). (The corresponding results associated with power distance and masculinity in columns (3) – (5) and (8) – (10) remain virtually unchanged when also controlling for the institutional and geographical covariates.) Thus, cultural characteristics turn out as valid omitted variables that systematically affect the gender-corruption nexus. More precisely, the results of columns (4) and (5), as well as columns (9) and (10), indicate that higher levels of power distance and masculinity in society facilitate corruption. At the same time, just one of these two indicators is sufficient to render the effects of female participation statistically meaningless. When power distance is included in specifications (4) and (9), the effect of women in society vanishes and the same holds when only masculinity is included in specifications (5) and (10). Figure 2 helps to explain the statistically insignificant effects of women in the public sphere on corruption, once cultural factors are accounted for. Power distance is negatively correlated with both measures of female participation and with the CPI, and the same holds for masculinity. Thus, analyses omitting power distance or masculinity, respectively, are likely to overestimate the effect of female participation on the absence of corruption. Consequently, earlier studies may have led to interpretations that exaggerate the role of female participation in explaining corruption levels. At the very least, this highlights that policy interventions aimed at increasing female participation alone are unlikely to be successful in directly alleviating corruption. Next, we now turn to addressing potential endogeneity concerns in order to test the robustness of our findings. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 18 Figure 2: Correlations: Women, culture, and corruption (1998-2014 average) PANEL A. Correlation of power distance with women in parliament, female labor force participation, and corruption PANEL B. Correlation of masculinity with women in parliament, female labor force participation, and corruption MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 19 IV.3 Addressing endogeneity concerns A common concern in the cross-country literature relates to endogeneity issues. In our case, reverse causality could influence our point estimates, as it is possible that corruption in turn affects cultural characteristics. Further, societies characterized by high power distance or masculinity may also be more corrupt because of other, unobserved dynamics, such as distinct historical developments or institutional arrangements previously not considered. To alleviate such concerns as well as possible, we present instrumental variables for our cultural variables. In particular, we follow Gorodnichenko and Roland (2011, 2016) by using genetic distance as an instrumental variable for Hofstede’s cultural dimensions. We use the Mahalanobis distance of the frequency of blood types A and B in a given country relative to the frequency of blood types A and B in the UK. 15 As Gorodnichenko and Roland (2011, 2016) argue, using the frequency of blood types as an instrument for cultural attributes is attractive for two reasons: First, it is a neutral genetic marker and second, it is the most broadly available genetic information, allowing for wide country coverage of the instrument. We want to briefly sketch the intuitive reasoning and validity of the instrument and refer to Gorodnichenko and Roland (2011, 2016) for further details. The authors argue that parents transmit both their genes and their cultural values to their descendants. Populations that mix should consequently be genetically and culturally similar as in both cases the transmission mechanism is at work. Thus, measures of genetic distance can be seen as proxies for differences in cultural values and can serve as a relevant instrument for cultural characteristics. The exclusion restriction is plausibly satisfied, as it remains difficult to identify a meaningful additional channel through which genetic distance may influence corruption levels. Nevertheless, we can of course not completely eliminate that possibility. Finally, to strengthen our interpretation we also provide alternative instrumentation strategies in the appendix Table A2. 15 Data on genetic distance to both the UK and the US are available and we follow Gorodnichenko and Roland’s (2018) recommendation that the UK is more suitable as a reference because it is genetically more homogeneous. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 20 Table 3 presents the results of the instrumental variable regressions for power distance and provides evidence for our central finding that culture mitigates the relationship between female representation and corruption. As we only have one observation per country for genetic distance, we average our data over the period from 1998 to 2014. Throughout Table 3, we only consider those countries for which Hofstede’s cultural dimensions are available to ensure the comparability of our results. In columns (1) and (5), we only include the key variables WIP and FLFP, respectively. Table 3: Culture and corruption: Cross-sectional regressions – Power distance instrumented with genetic distance (1) (2) (3) (4) (5) (6) (7) (8) OLS OLS IV IV OLS OLS IV IV Second stage – Dependent variable: Corruption Perceptions Index WIP 0.107 *** 0.029 -0.049 -0.058 (0.023) (0.020) (0.031) (0.039) FLFP 0.030 *** 0.003 -0.015 -0.012 (0.011) (0.009) (0.011) (0.012) Power distance -0.065 *** -0.130 *** -0.148 *** -0.071 *** -0.119 *** -0.132 *** 0.009 (0.024) (0.045) (0.009) (0.020) (0.038) Controls Yes Yes First stage – Dependent variable: Power distance Genetic distance 9.901 *** 6.190 *** 12.583 *** 7.665 *** (2.056) (2.110) (2.245) (2.505) WIP -0.899 *** -0.686 *** (0.174) (0.181) FLFP -0.301 *** -0.158 (0.097) (0.110) Controls Yes Yes # of observations 88 88 88 88 88 88 88 88 F-Stat (first stage) 23.195 8.604 31.416 9.364 Endogeneity test 9.34 8.17 9.77 7.71 AR test statistic 18.71 14.26 25.98 15.04 Notes: Data are averaged over 1998-2014. In columns (3), (4), (7) and (8), power distance is instrumented with Mahalanobis distance of frequency of blood types A and B in a given country relative to the frequency of blood types A and B in the UK from Gorodnichenko and Roland (2011, 2016). Control variables include the polity IV index (variable polity2), openness, a dummy for landlockedness, and tropical climate. The endogeneity test reports the robust score test by Wooldridge (1995). The AR test statistic is the Anderson-Rubin test statistic obtained from Stata’s weakiv command. Robust standard errors are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 21 Both independent variables show a positive association with corruption (columns 1 and 5) that vanishes once the cultural factor power distance is introduced (columns 2 and 6). Power distance is statistically significant at the one percent level and produces a negative coefficient in predicting the absence of corruption. Specifications (3), (4), (7), and (8) report the results from IV regressions for our two measures of female participation in politics and the labor market. Analyzing the test statistics for the quality of the instrument (Wooldridge 1995, Finlay and Magnusson 2009), using genetic distance is econometrically appropriate. When power distance is instrumented with genetic distance, it remains statistically significant and actually further increases in size. 16 Thus, it is possible that a conventional OLS framework regressing corruption on power distance underestimates the underlying relationship. Confirming previous results, both gender variables lose their explanatory power as soon as power distance is included. The respective coefficients are not significant on conventional levels of statistical relevance, indicating a zero effect. When re-estimating Table 3 with masculinity as the main cultural attribute, we obtain similar results, i.e., the statistical relevance of women in the public sphere vanishes (results available on request). However, the instrument suggested by Gorodnichenko and Roland (2011; 2016) is too weak for masculinity, which is why we restrict our presentation to power distance. IV.4 Further robustness tests and extensions In the appendix, we first present further strategies to deal with potential endogeneity issues concerning the share of women in parliament and the labor force, as well as culture (see Table 16 With first stage F-statistics ranging from 8.5 to 31.4, depending on the control variables used, our instrument produces a reasonable accuracy in predicting power distance, though marginally weak in some specifications, as it does not clear the conventional threshold of ten in some instances (see Staiger and Stock 1997 and Stock and Yogo 2005 for tests on weak instruments). MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 22 A2). 17 All analyses confirm our main findings, as virtually no effect for WIP or FLFP emerges once cultural attributes are accounted for, both in terms of statistical significance and magnitude. Second, we include additional control variables used in the literature on corruption (Dollar, Fisman, and Gatti 2001; Svensson 2005; Treisman 2007) into our framework in Tables A3 and A4. In addition to the Polity IV index, we add civil liberties and a press freedom index as institutional controls that might influence the interplay between culture, gender, and corruption. Furthermore, we control for historical factors. Following La Porta, Lopez-de-Silanes, and Shleifer (2008), we add a set of dummy variables to control for a country’s legal origin. We also include a binary variable indicating whether a country was colonized. As Treisman (2000) finds that traditionally Protestant countries are perceived to be less corrupt, we control for the share of Protestants in the total population. To account for the possibility that the degree of ethnolinguistic diversity could simultaneously affect culture and the dynamics of political networks, we add the fractionalization index provided by Alesina et al. (2003). Further, we include the average years of schooling as a measure of educational attainment. Third, we introduce lagged instead of contemporaneous versions of our measures of female representation. If female representation affects corruption perceptions, there is reason to believe that this occurs with a delay. For example, before corruption perceptions (as measured by our main dependent variable CPI) change, corruption levels would likely have to change first. To take into account that the transmission from women in parliament and the labor force to corruption perceptions probably takes time, we use oneand four year lags of both measures in our regression framework. Tables A5 and A6 show that this strategy produces consistent results. 17 We use the log of years of agriculture in 1,500 BC to instrument female representation. Hansen, Jensen, and Skovsgaard (2015) argue that societies with a long history of agriculture developed stronger male-dominated norms and cultural beliefs due to historical gender division of labor. Thus, women in agricultural societies became more dependent on men as compared to hunter-gatherer societies where both men and women contributed to the provision of food (Iversen and Rosenbluth 2010). Being persistent over time, these patriarchal values still influence contemporary gender roles. Conditional on current economic development and geographical controls, an early Neolithic revolution does not only have a significant negative effect on today’s FLFP, but also on the introduction of female suffrage and WIP. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 23 Forth, as the share of women in parliament may vary less outside electoral years, we take four-year averages of all incorporated variables in order to ensure sufficient variation stemming from the electoral cycles. Table A7 shows that the effects of both measures of female representation remain statistically indistinguishable from zero, while power distance produces a negative and statistically powerful effect on the absence of corruption. When masculinity is included as the cultural dimension of interest, we also derive consistent results. Sixth, we use the World Bank’s Control of Corruption index (CoC) as an alternative measure for corruption to ensure that our findings are not specific to the CPI (see Table A8). The statistically significant effect of the share of women in parliament and the labor force is present in a cross-sectional setting, but entirely vanishes once we add country fixed-effects. The inclusion of power distance and masculinity affects the relationship between women and corruption similarly as the inclusion of fixed-effects. Again, we estimate relatively precise zero effects of female participation on corruption when culture is accounted for. Instrumenting power distance with blood distance also yields similar results irrespective of the corruption measure used. Finally, we rule out the possibility of limited comparability of the CPI over time, as from 2012 onwards a different methodology has been used to construct the CPI. In our final robustness test in Table A9, we check the restricted sample with data until 2011. Again, all findings support our interpretations. V. C ONCLUDING REMARKS Does greater representation of women in politics and business systematically decrease corruption levels? This study takes a detailed look at that relationship and our results suggest no statistically or economically significant association once we account for country-specific timeinvariant factors. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 30 O NLINE A PPENDIX Figure A1: Boxplots by year, showing mean and median for the three main variables of interest. 0 2 4 6 8 10 Corruption Perceptions Index 19981999200020012002200320042005200620072008200920102011201220132014 Median Mean 0 20 40 60 Perceentage of women in parliament 19981999200020012002200320042005200620072008200920102011201220132014 Median Mean MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 31 20 40 60 80 100 Female to male labor force participation rate 19981999200020012002200320042005200620072008200920102011201220132014 Median Mean MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 32 Table A1: Descriptive statistics Variable N Mean SD Min Max Source Description % women in parliament 3,014 15.77 10.68 0 63.8 Inter-Parliamentary Union (IPU) Percentage of seats occupied by women in the lower and upper chamber Female to male labor force participation rate 3,162 69.01 19.81 15.42 107.12 International Labor Organization (ILO) Number of women divided by number of men in the labor force (population ages 15 and older that is economically active) CPI 2,480 4.23 2.17 0.4 10 Transparency International Corruption Perceptions Index, measuring the absence of corruption. 0 (highly corrupt) - 10 (very clean) CoC 3,017 -0.02 1.00 -1.92 2.59 World Bank Worldwide Governance Indicators Control of Corruption index; -2.5 (highly corrupt) – 2.5 (very clean) Ln(GDP/cap) 3,839 8.15 1.63 4.24 11.97 World Development Indicators, World Bank Log of GDP per capita in constant 2011 international $, PPP Polity IV 3,224 3.40 6.50 -10 10 Polity IV Project, Center for Systematic Peace Variable polity2; Regime indicator, combining democracy and autocracy indices. -10 (highly autocratic) - 10 (highly democratic) Civil liberties 3,249 3.33 1.84 1 7 Freedom House Measure of freedom of expression, assembly, association, education, and religion. 1 (high degree of civil liberties) - 7 (no civil liberties) Free press 3,746 46.49 24.34 0 100 Freedom House Measure of overall press freedom. 1 (the most free) - 100 (the least free) Openness 3,656 90.04 52.93 0.02 531.74 The World Bank Sum of imports and exports of goods and services (% of GDP) Schooling 1,947 7.69 3.10 1.10 13.10 UNDP Average number of years of education received by people ages 25 and older Ethnolinguistic fractionalization 3,500 0.44 0.26 0 0.93 Alesina et al. (2003) Degree of ethno-linguistic fractionalization. 0 (homogenous) - 1 (highly diverse) Protestants 3,540 0.14 0.21 0 0.90 Alesina, Giuliano and Nunn (2013) Share of protestants in the total population Tropical climate 3,480 0.74 0.42 0 1 Alesina, Giuliano and Nunn (2013) Fraction of land in the tropics or subtropics Landlocked 4,357 0.20 0.40 0 1 The World Bank Binary variable indicating whether a country is landlocked Region dummies The World Bank Set of dummy variables for seven world regions Legal origin La Porta, Lopez-deSilanes and Shleifer (2008) Set of dummy variables for UK, French, German, Scandinavian, and Socialist legal origin Colonial dummy 3,600 0.63 0.48 0 1 Teorell et al. (2013) Dummy variable indicating whether a country used to be a colony Power distance 1,840 63.27 21.19 11 100 The Hofstede Centre Degree to which less powerful members of society accept and expect that power is distributed unequally Uncertainty avoidance 1,840 63.71 21.32 8 100 The Hofstede Centre Society’s tolerance when it comes to ambiguity and uncertainty Individualism 1,840 40.29 22.37 6 91 The Hofstede Centre Degree of interdependence between the members of society Masculinity 1,840 47.88 18.76 5 100 The Hofstede Centre Social preference for male values generating a more competitive social environment Genetic distance 154 1.74 0.81 0.00 3.59 Gorodnichenko and Roland (2018) Mahalanobis distance of frequency of blood types A and B in a given country relative to the frequency of blood types A and B in the UK Years agriculture 152 4,783.63 2412.08 362 10,500 Putterman (2008) Ln (years of agriculture in 1,500 CE) (years since the Neolithic revolution) MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 33 Table A2: Female representation instrumented with years since Neolithic Revolution (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) OLS OLS IV IV IV OLS OLS IV IV IV WIP WIP WIP/PD FLFP FLFP FLFP/PD Second stage – Dependent variable: Corruption Perceptions Index WIP 0.107 *** 0.029 -0.183 0.019 -0.179 (0.023) (0.020) (0.221) (0.094) (0.247) FLFP 0.030 ** 0.003 -0.037 0.005 -0.032 (0.011) (0.010) (0.033) (0.023) (0.029) Power -0.065 *** -0.115 ** -0.053 *** -0.179 * -0.071 *** -0.082 *** -0.056 *** -0.128 *** distance (0.009) (0.056) (0.016) (0.097) (0.009) (0.012) (0.007) (0.037) Controls Yes Yes Yes Yes First stage (fs1) – Dependent variable: Female participation (WIP/FLFP) Ln(years of -1.871 -3.168 ** -4.097 *** -9.146 *** -13.021 *** -13.435 *** agriculture) (1.437) (1.529) (1.434) (3.129) (3.504) (3.131) Genetic -1.577 2.321 distance (1.261) (2.427) First stage (fs2) – Dependent variable: Power distance (PD) Ln(years of 5.955 * 5.955 * agriculture) (3.236) (3.236) Genetic 7.027 *** 7.027 *** distance (2.300) (2.300) # of obs. 87 87 87 87 87 87 87 87 87 87 F-Stat (fs1) 1.696 4.293 4.293 8.542 13.808 13.808 F-Stat (fs2) 8.523 9.393 Notes: Data are averaged over 1998-2014. In columns (3), (4), (5), (8), (9) and (10) WIP and FLFP, respectively, is instrumented with the log years since the Neolithic revolution from Putterman (2008). Additionally, in columns (5) and (10), power distance is instrumented with Mahalanobis distance of frequency of blood types A and B in a given country relative to the frequency of blood types A and B in the UK from Gorodnichenko and Roland (2011, 2016). Controls include the polity IV index (polity2 variable), openness, a dummy variable for landlockedness, and tropical climate. All variables included in the second stage are also included in the first stage. To obtain the F-statistics of the first stage, we ran the regressions reported in columns (5) and (10) separately with just one variable instrumented. Robust standard errors are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 34 Table A3: Robustness tests for power distance (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) OLS OLS OLS OLS OLS OLS OLS OLS OLS OLS WIP 0.014 0.017 0.018 0.010 0.015 (0.010) (0.011) (0.011) (0.009) (0.009) FLFP 0.003 0.003 0.011 0.005 0.009 (0.007) (0.009) (0.007) (0.007) (0.007) Power distance -0.020 *** -0.019 *** -0.017 *** -0.015 ** -0.017 *** -0.022 *** -0.022 *** -0.018 *** -0.016 ** -0.018 *** (0.005) (0.005) (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) (0.006) Ln (GDP/cap) -2.559 *** -2.014 *** -2.315 *** -2.504 *** -3.537 *** -2.508 *** -2.110 *** -2.003 *** -2.337 *** -3.066 *** (0.658) (0.710) (0.634) (0.621) (0.699) (0.711) (0.706) (0.679) (0.613) (0.797) Ln (GDP/cap) 2 0.201 *** 0.186 *** 0.199 *** 0.205 *** 0.259 *** 0.198 *** 0.192 *** 0.183 *** 0.196 *** 0.235 *** (0.041) (0.041) (0.038) (0.038) (0.042) (0.043) (0.040) (0.040) (0.038) (0.045) Institutions Yes Yes & Openness Geography Yes Yes History Yes Yes Population Yes Yes Education Yes Yes Region FE Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # of observations 1305 1305 1305 1305 885 1305 1305 1305 1305 885 # of countries 89 89 89 89 89 89 89 89 89 89 R 2 0.827 0.831 0.819 0.819 0.804 0.824 0.827 0.819 0.819 0.804 Notes: Dependent variable: CPI. The following control variables are included. Institutions & Openness: Polity IV index, civil liberties, free press index, imports and exports as share of GDP. Geography: Landlocked, tropical climate, region fixed effects. History: Legal origin, colony dummy. Population: Share of Protestants in total population, ethnolinguistic fractionalization. Education: Average years of schooling. Robust standard errors clustered at the country level are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 35 Table A4: Robustness tests for masculinity (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) OLS OLS OLS OLS OLS OLS OLS OLS OLS OLS WIP 0.015 0.016 0.025 ** 0.011 0.015 (0.011) (0.010) (0.012) (0.010) (0.011) FLFP 0.004 0.003 0.012 0.007 0.010 (0.007) (0.008) (0.008) (0.007) (0.007) Masculinity -0.012 ** -0.022 *** -0.016 *** -0.012 ** -0.016 *** -0.014 *** -0.024 *** -0.017 *** -0.013 *** -0.017 *** (0.005) (0.006) (0.006) (0.005) (0.005) (0.005) (0.005) (0.006) (0.005) (0.005) Ln (GPD/cap) -3.122 *** -1.703 ** -2.704 *** -2.875 *** -4.222 *** -3.081 *** -1.802 ** -2.468 *** -2.647 *** -3.779 *** (0.612) (0.684) (0.630) (0.564) (0.641) (0.688) (0.732) (0.684) (0.601) (0.795) Ln (GDP/cap) 2 0.241 *** 0.181 *** 0.230 *** 0.234 *** 0.302 *** 0.240 *** 0.190 *** 0.220 *** 0.221 *** 0.281 *** (0.038) (0.040) (0.038) (0.034) (0.037) (0.041) (0.042) (0.041) (0.036) (0.044) Institutions Yes Yes & Openness Geography Yes Yes History Yes Yes Population Yes Yes Education Yes Yes Region FE Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes # of observations 1,305 1,305 1,305 1,305 885 1,305 1,305 1,305 1,305 885 # of countries 89 89 89 89 89 89 89 89 89 89 R 2 0.820 0.841 0.820 0.818 0.806 0.818 0.837 0.817 0.818 0.806 Notes: Dependent variable: CPI. The following control variables are included. Institutions & Openness: Polity IV index, civil liberties, free press index, imports and exports as share of GDP. Geography: Landlocked, tropical climate, region fixed effects. History: Legal origin, colony dummy. Population: Share of Protestants in total population, ethnolinguistic fractionalization. Education: Average years of schooling. Robust standard errors clustered at the country level are reported in parentheses. * p < 0.10, ** p < 0.05, *** p < 0.01. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 36 A5: Main findings allowing for lagged adjustment of corruption levels (1 year lags) (1) (2) (3) (4) (5) (6) (7) (8) OLS FE OLS OLS OLS FE OLS OLS WIP t - 1 0.031 *** 0.008 0.031 *** 0.007 (0.011) (0.007) (0.010) (0.010) FLFP t - 1 0.017 ** -0.016 0.016 * 0.006 (0.008) (0.012) (0.009) (0.009) Ln (GDP/cap) -3.299 *** 2.694 ** -2.049 ** -1.655 ** -2.907 *** 2.746 * -2.008 ** -1.534 ** (0.617) (1.306) (0.877) (0.672) (0.669) (1.391) (0.811) (0.720) Ln (GDP/cap) 2 0.260 *** -0.076 0.193 *** 0.171 *** 0.241 *** -0.093 0.196 *** 0.166 *** (0.037) (0.087) (0.049) (0.039) (0.040) (0.089) (0.046) (0.041) Polity IV 0.033 0.031 (0.021) (0.021) Openness 0.003 0.003 (0.002) (0.002) Landlocked -0.012 -0.181 (0.283) (0.259) Tropical climate -0.534 -0.516 (0.395) (0.432) Power distance -0.016 *** -0.016 *** (0.005) (0.006) Masculinity -0.019 *** -0.019 *** (0.005) (0.005) Country FE Yes Yes Region FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Yes Yes # of observations 1219 1219 1219 1219 1291 1291 1291 1291 # of countries 80 80 80 R 2 0.778 0.179 0.811 0.838 0.777 0.186 0.809 0.840 Notes: Dependent variable: CPI. Robust standard errors clustered at the country level are reported in parentheses. Measures of female participation included as lagged variables dated t-1. * p < 0.10, ** p < 0.05, *** p < 0.01. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 37 A6: Main findings allowing for lagged adjustment of corruption levels (4 year lags) (1) (2) (3) (4) (5) (6) (7) (8) OLS FE OLS OLS OLS FE OLS OLS WIP t - 4 0.032 *** 0.002 0.030 *** 0.007 (0.011) (0.007) (0.011) (0.010) FLFP t - 4 0.017 ** -0.019 0.014 0.005 (0.007) (0.013) (0.009) (0.009) Ln (GDP/cap) -3.308 *** 4.616 *** -1.988 ** -1.633 ** -2.983 *** 3.972 ** -2.123 ** -1.640 ** (0.629) (1.534) (0.901) (0.690) (0.674) (1.589) (0.849) (0.736) Ln (GDP/cap) 2 0.259 *** -0.193 * 0.189 *** 0.170 *** 0.245 *** -0.162 0.202 *** 0.171 *** (0.038) (0.103) (0.051) (0.040) (0.040) (0.105) (0.048) (0.042) Polity IV 0.026 0.030 (0.022) (0.021) Openness 0.003 0.003 (0.002) (0.002) Landlocked -0.005 -0.173 (0.288) (0.261) Tropical -0.576 -0.592 climate (0.391) (0.439) Power distance -0.015 *** -0.016 *** (0.005) (0.006) Masculinity -0.019 *** -0.019 *** (0.005) (0.005) Country FE yes yes Region FE yes yes yes yes Time FE yes yes yes yes yes yes yes yes # of observations 1022 1022 1022 1022 1087 1087 1087 1087 # of countries 80 80 80 80 80 80 80 80 R 2 0.778 0.203 0.811 0.836 0.775 0.221 0.808 0.838 Notes: Dependent variable: CPI. Robust standard errors clustered at the country level are reported in parentheses. Measures of female participation included as lagged variables dated t-4. * p < 0.10, ** p < 0.05, *** p < 0.01. MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 38 Table A7: Main findings incorporating four year averages of all incorporated variables (1) (2) (3) (4) (5) (6) (7) (8) OLS FE OLS OLS OLS FE OLS OLS WIP 0.028 ** 0.006 0.028 ** 0.005 (0.012) (0.010) (0.011) (0.010) FLFP 0.017 ** -0.019 0.018 ** 0.007 (0.007) (0.013) (0.009) (0.009) Power distance -0.016 *** -0.016 *** (0.005) (0.006) Masculinity -0.019 *** -0.019 *** (0.005) (0.005) Ln (GDP/cap) -3.454 *** 2.869 ** -1.952 ** -1.728 ** -2.914 *** 2.454 * -1.903 ** -1.561 ** (0.637) (1.333) (0.930) (0.689) (0.670) (1.425) (0.869) (0.767) Ln (GDP/cap) 2 0.270 *** -0.096 0.190 *** 0.176 *** 0.241 *** -0.081 0.191 *** 0.167 *** (0.038) (0.090) (0.052) (0.040) (0.040) (0.092) (0.049) (0.044) Polity IV 0.028 0.025 (0.023) (0.023) Openness 0.003 0.003 (0.002) (0.002) Landlocked 0.094 -0.087 (0.300) (0.273) Tropical climate -0.526 -0.491 (0.401) (0.438) Country FE No Yes No No No Yes No No Region FE Yes Yes Yes Yes Period FE Yes Yes Yes Yes Yes Yes Yes Yes # of observations 335 335 335 335 335 335 335 335 # of countries 90 90 90 90 90 90 90 90 R 2 0.781 0.192 0.817 0.845 0.783 0.199 0.814 0.846 Notes: Dependent variable: CPI. Robust standard errors clustered at the country level are reported in parentheses. The data are averaged over 5 and 4 periods, respectively. More precisely, the first period is 1998-2002, the second period is 20032006, the third period 2007-2010, and the fourth period 2011-2014. * p < 0.10, ** p < 0.05, *** MANUS CRIP T ACCEP TED ACCEPTED MANUSCRIPT 39 Table A8: Main findings using alternative measure of corruption: Robustness tests with CoC index (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) OLS FE OLS OLS IV IV OLS FE OLS OLS IV IV WIP 0.014 *** 0.000 0.013 *** 0.003 -0.023 -0.027 (0.005) (0.002) (0.005) (0.004) (0.015) (0.019) FLFP 0.010 *** 0.008 0.010 ** 0.005 -0.006 -0.005 (0.003) (0.005) (0.004) (0.004) (0.005) (0.005) Ln (GDP/cap) -1.595 *** 0.476 -0.818 * -0.769 ** -1.274 *** 0.660 -0.758 * -0.666 * (0.288) (0.455) (0.412) (0.304) (0.312) (0.431) (0.390) (0.348) Ln (GDP/cap) 2 0.125 *** -0.001 0.084 *** 0.081 *** 0.107 *** -0.010 0.082 *** 0.076 *** (0.017) (0.028) (0.023) (0.018) (0.019) (0.026) (0.022) (0.020) Polity IV 0.008 -0.051 0.010 -0.045 (0.010) (0.038) (0.010) (0.036) Openness 0.001 0.005 * 0.001 0.004 (0.001) (0.003) (0.001) (0.003) Landlocked 0.100 -0.239 0.007 -0.114 (0.134) (0.332) (0.120) (0.312) Tropical -0.190 -0.267 -0.166 -0.252 climate (0.182) (0.347) (0.189) (0.361) Power distance -0.006 ** -0.061 *** -0.069 *** -0.006 ** -0.056 *** -0.061 *** (0.002) (0.012) (0.021) (0.003) (0.009) (0.018) Masculinity -0.009 *** -0.009 *** (0.003) (0.002) Country FE Yes Yes Region FE Yes Yes Yes Yes Times FE Yes Yes Yes Yes Yes Yes Yes Yes # of obs. 1,237 1,237 1,237 1,237 88 88 1,284 1,284 1,284 1,284 88 88 # of countries 90 90 90 90 88 88 90 90 90 90 88 88 R 2 0.779 0.096 0.812 0.838 0.194 0.138 0.783 0.105 0.812 0.839 0.278 0.261 F-Stat (first stage) 23.195 8.604 31.416 9.364 Notes: Dependent variable: CoC. Robust standard errors clustered at the country level are reported in parentheses. In columns (5), (6), (11) and (12) power distance is instrumented with blood distance. Data averaged over 1998-2014 is used for the IV regressions. * p < 0.10, ** p < 0.05, *** p < 0.01.