The World Bank Economic Review, 37(1), 2023, 74–92 https://doi.org10.1093/wber/lhac021 Article Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Legal Bans, Female Genital Cutting, and Education: Evidence from Senegal Jorge García-Hombrados and Edgar Salgado Abstract A law that banned the practice of female genital cutting (FGC) in Senegal in 1999 reduced its prevalence and increased educational investments in girls. These results are not driven by mechanisms like health, broader changes in empowerment, or child marriage. Suggestive evidence indicates that results could be driven by some parents of future brides reacting to the increase in the cost of FGC caused by the law by abandoning this practice and investing in their daughter’s education to compensate for smaller bride prices among uncut women. JEL classification: I25, K42, O12 Keywords: Female genital cutting, education, harmful norms, social norms, impact evaluation 1. Introduction The practice of female genital cutting (FGC)1 affects more than 200 million women worldwide (UNICEF 2016). Although many governments and international organizations have mobilized large amounts of resources in the last decades to fight FGC (UNICEF 2016), the practice remains a social norm in many countries, affecting around 3 million girls annually (Novak 2020). Despite the attention gathered from policy makers and civil society, evidence on what policies are effective to fight this cultural practice is scarce (Crisman et al. 2016; De Cao and La Mattina 2019; Harari 2019). This article documents the beneficial effects of a law that banned the practice of FGC in Senegal in January 1999 on the prevalence of this practice and on educational investments. In a context where the majority of the cuts occur during infancy and where FGC is deeply rooted in the tradition of some ethnic groups, but hardly exists in others, our difference-in-differences strategy compares women Jorge García-Hombrados (corresponding author) is an assistant professor at the Universidad Autónoma de Madrid and a research affiliate at the Max Planck Institute for Demographic Research, Madrid, Spain; his email address is jorge.garciah@uam.es. Edgar Salgado is an economist at the International Finance Corporation; his email address is esalgadochavez@ifc.org. The research for this article was financed by the ERC grant 336475 to Mikko Myrskylä. The au- thors are thankful for comments from Esther Duflo, Luke Chicoine, David Evans, Florian Auferoth, Stefan Leefers, Berkay Ozcan, Lindsey Novak, Vikram Pathania, Mikko Myrskylä, two anonymous reviewers at the WBER along with participants at seminars at the NOVAFRICA Conference, CSAE Conference, University of Kent, University of Newcastle, University of Reading, Universidad de Alicante, Universidad Autónoma de Madrid, UCL, LSE (STICERD, Alpha workshop, and Depart- ment of Social Policy), Max Planck Institute, EEA annual meeting, Nordic Conference of Development Economics, and IDB. A supplementary online appendix is available with this article at The World Bank Economic Review website. 1 Defined as the ritual cutting of some or all of the external female genitalia for reasons unrelated to health (see https://www.who.int/news-room/fact-sheets/detail/female-genital-mutilation). © The Author(s) 2022. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com The World Bank Economic Review 75 and girls from these groups born before and after the introduction of the law. Results show that the ban was followed by a reduction in the probability of experiencing FGC and an increase in the extensive Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 margin of education. While the main focus of the paper is the estimation of the effects of the law, we complement this analysis using the variation in the probability of undergoing FGC generated by this law as an instrumental variable to estimate the implied effect of FGC on education. Under the assumption that the ban only affected education through its effect on FGC, a reduction of 10 percentage points in the prevalence of FGC (mean is 32 percent) would lead to a reduction of 2.6 percentage points in the probability of never attending school (mean is 37 percent). On the other hand, we find limited effects of the law on the intensive margin of education: while the law substantially decreased the number of women receiving one or two years of education, it did not affect the probability of achieving three or more years of education. Falsification tests suggest that the estimates are unlikely to be driven by differential trends in FGC or educational investments across different ethnic groups before the introduction of the law. The underre- porting of FGC among younger cohorts of girls is also unlikely to be driving the results according to further tests. More importantly, even if the law led to underreporting among women born after the law, this could only bias downwards the estimates of the effect of FGC on education, which would not affect the main conclusions of the paper. Other interventions confounding the effects are also ruled out. The results are robust to the use of different analytical samples and regression specifications and to a placebo test examining the effects of the law on the education of a group of boys that should be unaffected by the law. Four mechanisms could explain the results. First, the ban may promote a broader change in gender norms regarding education or shifting the power in communities or households towards individuals with higher consideration for women’s education. Second, the law may raise educational investments through improving girls’ health, which may lead to higher returns on educational investments. Third, the law may reduce early marriage, a related social practice, reducing early school dropout. Results indicate that none of these three mechanisms plays a major role. On the other hand, based on anthropological literature, the paper shows suggestive evidence of a fourth mechanism operating through the marriage market. Among ethnic groups practicing FGC, women who have undergone this practice are perceived as purer and more loyal (Wagner 2015; Karumbi and Jacinta 2017). Kolawole and Anke (2010) and Karumbi and Jacinta (2017) review several qualitative studies on different sub-Saharan countries and conclude that, in those communities where this practice is common, FGC increases marriage payments received by the parents of the bride, also known as the bride price, and in some cases, FGC also seems to improve marriage prospects. Although their broad roles in the marriage market are different,2 the effect of FGC on bride price resembles the one reported for education, with different studies showing that even a few years of education seem to increase the bride price in Zambia, Indonesia, and Senegal (Ashraf et al. 2020; Hotte and Lambert 2020). In contexts where both FGC and education increase the marriage payments received by the parents of the bride, and even if FGC and education signal different treats or skills in the marriage market and would lead to marrying a different type of husband, FGC and education might be perceived by non-altruistic parents of brides as substitute pre-marital investments. If the ban raises the cost of FGC, some parents might decide not to subject their daughters to FGC and invest in their education to compensate for reductions in the bride price that parents will receive at their daughter’s marriage. Lack of data prevents an assessment of the effects of FGC on bride price or other marriage-market outcomes, but evidence indicates that while the law has a large effect on the prevalence of FGC in both rural and urban areas, its effect on education is negligible in urban areas, where the value of FGC in the marriage market is much smaller. This result should be interpreted as suggestive evidence in favor of this mechanism. Providing conclusive evidence 2 For example, FGC might signal loyalty and education independence and productivity. 76 García-Hombrados and Salgado is however not possible because Demographic and Health Surveys do not provide information on bride price and the vast majority of the women born after the introduction of the law remained unmarried Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 when the surveys were implemented. This study contributes to the growing evidence on the effects of social norms and cultural traits (Nunn 2012; Efferson et al. 2015; La Ferrara and Milazzo 2017; Michalopoulos, Putterman, and Weil 2019). First, it adds to the literature that investigates whether legal changes can be effective instruments to address harmful traditional practices in developing countries (Crisman et al. 2016; Acemoglu and Jack- son 2017; Garcia-Hombrados 2021). While bans on FGC are widespread political instruments, there is little evidence on their effectiveness (Crisman et al. 2016; De Cao and La Mattina 2019; Cetorelli et al. 2020). This study contributes to the literature by showing that FGC bans could reduce this practice even in countries with limited capacity to enforce the law. In particular, this study provides crucial insights regarding implementation. The Senegalese strategy focused on expanding a lasting fear of prosecution through sanctioning a few high profile cases that received high coverage, which shows the way for other countries willing to fight FGC. Second, this article contributes to the understanding of the consequences of FGC. Current research has mainly focused on the effect of FGC on health, documenting a lack of association with health impairment or fertility, while there is some relation to worse reproductive and sexual health, especially higher preva- lence of sexually transmitted diseases, genital sores, or hard labor at birth (Berg et al. 2014; Wagner 2015). However, because the reasons for FGC are socioeconomic, and some studies suggest it plays an important role in marriage markets (Mackie 1996; Kolawole and Anke 2010; Karumbi and Jacinta 2017), FGC may have consequences beyond health. The ban on FGC not only reduced its prevalence, and because of that education among girls increased. While the effects on the intensive margin of education are modest, results show that interventions that reduce FGC could indirectly improve women’s human capital accumulation in countries where this practice is prevalent. Third, existing empirical studies are mostly correlations. The causal identification of the effects of FGC on health and socioeconomic outcomes is however methodologically challenging, as the adoption of this practice might be correlated with unobserved factors such as attitudes towards tradition that may lead to endogeneity. The natural experiment used in this study provides, under plausible assumptions, an adequate framework to extend the analysis conducted in this paper to the study of the broader socioeconomic effects of FGC.3 Fourth, this study contributes to the literature that explores the nature, origin, and persistence of this practice (Bellemare, Novak, and Steinmetz 2015; Wagner 2015; Poyker 2016; Vogt et al. 2016; Becker 2019; Becker 2019; Diabate and Mesplé-Somps 2019; Efferson, Vogt, and Fehr 2020; Novak 2020) by providing empirical and theoretical insights into how FGC interacts with other pre-marital investments. The suggestive results on mechanisms are consistent with the hypothesis that parents react to the increase in the cost of FGC through abandoning this practice and investing in education to compensate for po- tential losses in bride price. These findings imply that parents react to the relative price of FGC and other pre-marital investments and call for more research assessing the impact of other interventions that could decrease the price of alternative pre-marital investments. 2. FGC in Senegal and the Introduction of the Ban The prevalence of FGC is widespread in West African countries, where it ranges between 3 percent in Niger and 99 percent in Guinea. In Senegal, 25 percent of women aged 15–49 report being cut (Allen et al. 2015). Unlike other social norms, such as bride price or polygamy, common across all ethnic groups 3 This exercise is not possible given that health and marriage-market outcomes in the DHS are only available for women at least 15 and there are few girls born after the introduction of the law. The World Bank Economic Review 77 in Senegal, FGC is embedded in of some them such as the Soninke, Mandingue, Diola, and the Poular, while it is sparse among the Wolof and the Serer. Unlike most East African countries, FGC in Senegal is Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 mainly conducted during infancy. Yoder and Wang (2013) show that 61 percent of FGC occurs before the end of the first year of life and that it is infrequent after early childhood. The data used in this study show that FGC occurs mostly during the first year of life among all practicing ethnic groups in Senegal, and the percentage of women who experienced FGC after the sixth year is less than 7 percent. In Senegal, FGC usually involves the partial or total removal of the clitoris and the labia, and, in most cases, the cuts are performed without anesthetics by traditional practitioners who have little knowledge of female anatomy and use crude, unsterile instruments (Berg et al. 2014). Tackling FGC topped Senegal’s policy agenda for decades. The flagship measure of the government was the approval of Law No. 99-05 which sanctions those who provoke sexual mutilations or give instruc- tions for their commission with six months to five years in prison, or hard labor for life if FGC results in death. The law was enacted on 29 January 1999, following the anti-FGC speech of US First Lady Hillary Clinton in Senegal, and after months of intense campaigning of Senegalese civic organizations. The gov- ernment raised awareness about the law, and knowledge of the law was widespread (Shell-Duncan et al. 2013). Following the introduction of the law, the government conducted some high-profile arrests and convictions of FGC perpetrators, which received widespread media coverage. Although prosecution was scarce Shell-Duncan et al. (2013), (Kandala and Komba 2015) argue that the strategy of the government set a lasting fear of prosecution. We examine the effectiveness of the law empirically in the Results section and find that, although the law decreased the prevalence of FGC remarkably, a non-negligible share of women born after the introduction of the law was subject to this practice. The Empirical Strategy section discusses and rules out – as potential limitations of the results – the existence of anticipation effects of the law and the possibility that the results are simply capturing an eventual effect of the law on misreporting of the FGC status. 3. Data The analysis exploits the 2016, 2015, 2014, 2012, and 2010 waves of the Senegalese DHS. Unlike in previous rounds of the Senegalese DHS, the 2010–2016 waves collected information on the FGC status of every female aged 0–49.4 The information on FGC is self-reported for respondents aged 15 or older, while reported by the mother for girls younger than 15. When conducting the survey, enumerators are instructed to seek privacy and to avoid the presence of other members of the family in the same room, although this might not always be possible. Additionally, the survey collects individual information on education and other basic demographic characteristics. On the other hand, information on health is only available for girls younger than 5 or older than 14 and on marital status, bargaining power, and labor-force participation for women aged at least 15. The analytical sample includes information from 28,425 females born since January 1990 who, by the time of the survey, were at least 7 years old. We restrict the analysis to females born since 1990 because the accuracy of the reporting of year of birth is likely to be higher among relatively younger cohorts. However, the selection of the exact birth date used as a threshold for being excluded from the sample is arbitrary and we also examine the robustness of the results to the inclusion of older cohorts of women (born from 1980). We focus on girls and women aged 7 or older because 7 is the legal age for starting school in Senegal. Additionally, the vast majority of FGC occurs during infancy, thus setting the threshold at 7 avoids the possibility that the reduction in the prevalence of FGC simply reflects that many girls in 4 FGC status was collected for females aged 0–49 in the 2014, 2015, and 2016 rounds of the DHS; the 2010 round did not gather FGC information for girls aged 11–14 and the 2012 round did not report FGC status for women aged 15 or older. Although the 2005 Senegalese DHS also includes a module on FGC, this information is only collected for women aged 15 or older. Because the main sample is restricted to females born since 1990, we do not use the 2005 round of the DHS in the main analysis. 78 García-Hombrados and Salgado Table 1. Prevalence of Female Genital Cutting (FGC) across Ethnic Groups in Senegal for Cohorts Born Before and After the Introduction of the Law Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Women born Women born Reduction before FGC ban after FGC ban incidence of FGC (1) (2) (3) (4) FGC Share of ethnic FGC Percentage Ethnic group prevalence group in sample prevalence Wolof 0.017 0.322 0.008 0.009 Poular 0.642 0.341 0.507 0.135 Serer 0.020 0.119 0.012 0.008 Mandingue 0.812 0.086 0.603 0.209 Diola 0.553 0.039 0.440 0.113 Soninke 0.679 0.019 0.523 0.156 Not a Senegalese 0.744 0.023 0.563 0.181 Other 0.450 0.051 0.314 0.136 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: The prevalence of FGC among each ethnic group before the introduction of the law and the share of each ethnic group in the sample are calculated using the sample of women and girls born between 1990 and January 1999 who were interviewed in the Senegalese DHS rounds 2016, 2015, 2014, 2012, and 2010. The prevalence of FGC among each ethnic group born after the introduction of the law is calculated using the sample of women and girls born after January 1999 that were 7 or older by the time of the survey and were interviewed in the Senegalese DHS rounds 2016, 2015, 2014, 2012, and 2010. the sample have yet to undergo FGC by the time of the survey. Age ranges between 7 and 26 years and 36 percent report never attending school. The average number of years of schooling is three. Thirty-three percent of girls and women have undergone FGC and are poorer, older, and more likely rural than girls and women who have not undergone FGC. Column 1 of table 1 reports the prevalence of FGC across ethnic groups, calculated from the group of girls and women born before the introduction of the law. The table also reports the prevalence of FGC among those cohorts born after the introduction of the law. In a country where inter-ethnic marriages are rare,5 the data confirm a wide variation in the prevalence of FGC across ethnic groups, which ranges between 81 percent among Mandingue and 1.7 percent among Wolof. On the other hand, ethnic groups in Senegal do not differ in terms of other social norms or religion. Figure 1 shows the evolution of FGC and school outcomes. The left-hand graph shows that while the prevalence of FGC hardly changed across cohorts of girls and women from ethnic groups where FGC is rare (e.g. Wolof and Serer), it dropped sharply among girls from FGC-prevalent groups born after the introduction of the law. The right-hand graph reveals that the percentage of individuals who never attended school is lower in ethnic groups that practice FGC. Among the cohorts of girls and women born after the introduction of the law, the upward trend in the probability of never having attended school is mechanically originated by the fact that many girls start school after the age of 7.6 The exclusion of the youngest cohorts from the analytical sample—to test whether the estimate is simply reflecting the fact that girls from non-FGC ethnic groups start school at a later age—does not change the results. 4. Empirical Strategy The first equation explores whether girls and women from ethnic groups with a larger prevalence of FGC born after the introduction of the law experienced larger reductions in the probability of undergoing FGC 5 Only 24 percent of marriages are between ethnic groups. In rural areas it is 17 percent. 6 While starting school after 7 is common (31 percent of first graders are older than 7), starting school after 10 is rare: less than 1.5 percent of children aged 11 at the time of the survey had started school in that year. The World Bank Economic Review 79 Figure 1. Female Genital Cutting (FGC) and Never in School across Year of Birth Cohorts (All Ethnic Groups) Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: Each dot represents the average prevalence of FGC and the proportion of girls and women who have never attended school, by year of birth, for ethnic groups in which the practice of FGC is traditional (i.e. Poular, Diola, Mandingue, Soninke, Non-Senegalese, and “Other” ethnic groups) and for ethnic groups in which FGC is not traditional (non-FGC; i.e. Wolof and Serer). The size of the dots reflects the quantity of data in our sample. We overlay local polynomial curves (bandwidth of 2) to show trends across FGC and non-FGC ethnic groups. compared with girls and women from ethnic groups with a lower prevalence of FGC: FGCikrst = α0 + α1 (POSTt × LawIntensityk ) + α2 (YearBirtht × Waves ) + α3 EthnicGroupk + α4 Villager + μikrst , (1) where FGCikrst indicates whether woman or girl i from ethnic group k, living in village r, interviewed in survey round s, and born in year t, experienced FGC, and YearBirtht × Waves , EthnicGroupk , and Villager are vectors of dummy variables to indicate survey round-year of birth, ethnic group, and DHS cluster (equivalent to village of residence). The variable POSTt equals 1 if the individual was born after the introduction of the law (January 1999), and LawIntensityk measures the mean prevalence of FGC within ethnic group k calculated using the cohorts of individuals born before the introduction of the law. As reported in column 1 of table 1, α 1 measures the differential change in the prevalence of FGC after the introduction of the law for ethnic groups with higher and lower levels of FGC before the introduction of the law. In other words, α 1 yields the effect of the degree of exposure to the law, which is believed to be larger among ethnic groups with a larger prevalence of FGC, on the probability of experiencing FGC.7 The variable μikrt is the error term. 7 Our approach requires the drop in the prevalence of FGC among females born after the law be larger for ethnic groups with a larger pre-law prevalence of FGC in percentage points, not in percentage terms. 80 García-Hombrados and Salgado The second equation estimates the reduced form effect of the law on education: Educationikrst = γ0 + γ1 (POSTt × LawIntensityk ) + γ2 (YearBirtht × Waves ) Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 + γ3 EthnicGroupk + γ4 Villager + θikrst , (2) where Educationikrst measures the education of woman or girl i from ethnic group k, living in village r, interviewed in survey round s, and born in year t. The parameter γ 1 yields the effect of the degree of exposure to the law on education. While the focus is on the effect of the ban on education and FGC, it is also possible to complement this analysis by providing an estimate of the effect of FGC on education. This involves using the interac- tion term POSTt × LawIntensityk in equation (1) as an instrumental variable or first stage to investigate whether the reduction in FGC caused by the law increased education. Using a two-stage least-squares (2SLS) procedure, we estimate the following second stage equation Educationikrst = β0 + β1 FGCikrst + β2 (YearBirtht × Waves ) + β3 EthnicGroupk + β4 Villager + uikrst , (3) where FGC is the predicted probability of FGC estimated from equation (1). The parameter of interest is β 1 , which yields the effect of FGC on education attained by girls and women who did not undergo FGC because they were exposed to the law. In this empirical strategy, the 2SLS parameter β 1 is the ratio between the parameters γ 1 and α 1 estimated in equations (1) and (2). The estimations of equations (1), (2), and (3) follow Cameron, Gelbach, and Miller (2008) and use wild bootstrapped standard errors clustered at the ethnic group × year of birth level. Furthermore, we also estimate the regressions with a vector of regional dummies and region-specific time trends to account for differential trends in FGC or education across regions in Senegal.8 The causal interpretation of the parameters α 1 and γ 1 relies on two conditions: that trends before the law are parallel, and that coefficients are not confounded by other interventions after the law. The first identification condition requires that, without the introduction of the law, the evolution of the prevalence of FGC and educational investments over year of birth cohorts should have been the same across ethnic groups. It is possible to test for parallel trends by examining whether the evolution of the prevalence of FGC and educational outcomes across cohorts of girls and women born before the introduction of the law is similar across the different ethnic groups. This assumption would fail if the prevalence of FGC among FGC ethnic groups started to decline among cohorts born before the introduction of the law, relative to girls from non-FGC ethnic groups. Figure 1 suggests that the evolution of both the prevalence of FGC and the probability of never having attended school was not different between FGC and non-FGC ethnic groups for pre-ban cohorts.9 Moreover, fig. 2 shows the evolution of these variables over cohorts of birth by ethnic group. The evolution of educational outcomes for girls and women born before the introduction of the law was remarkably similar across every ethnic group. However, although the prevalence of FGC remains constant across cohorts of girls and women born before the introduction of the law in six out of the eight ethnic categories in the sample, the prevalence of FGC was already declining among girls and women from Soninke and the ethnic category “Other” who were born before the introduction of the law. The main analysis estimates the above equations restricting the sample to the six ethnic groups for which the prevalence of FGC among cohorts born before the law did not decrease. Results are robust 8 Results including household fixed effects were very similar to those obtained with cluster FE. The coefficient of the 2SLS FGC variable is 0.255, statistically significant at the 5 percent level. We however do not think this strategy is optimal since relevant variation (at least one woman born between 1990 and 1998 and at least one woman born from 1999 aged 7 or older) is only present in 28 percent of the households. Results of this analysis are available upon request. 9 Figure S3.1 in the supplementary online appendix shows the graphs for the rest of the educational outcomes used in the study for these two groups, showing similar patterns. The World Bank Economic Review 81 Figure 2. Female Fenital Cutting (FGC) and Educational Outcomes for all Ethnic Groups: Continuous Treatment Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: The graphs show local polynomial curves (bandwidth of 2) to show trends across ethnic groups in terms of the prevalence of FGC and educational outcomes in order to assess the parallel trends condition. to the inclusion of girls and women from these two excluded ethnic groups in the analytical sample and find similar results. The two ethnic groups removed from the sample in the main estimations account for 7 percent of the girls and women in the sample and, as can be seen in figs S3.2 and S3.3, excluding these from the analysis ensures the pre-law parallel trends in terms of educational variables and FGC. The parallel trends condition is also tested empirically using a leads and lags equation in fig. S3.4. The results of a Wald test for joint significance of the lead variables show no evidence of differential trends across ethnic groups for cohorts born before the introduction of the law in terms of the prevalence of FGC, the probability of never having attended school, and the probabilities of having completed at least one year of school, two years of school, and four years of school. On the other hand, the Wald test for pre-trends in the probability of having completed at least three years of school is marginally significant at 10 percent (p-value equal to 0.094). Further checks in the supplementary online appendix examine whether the results could be driven by differential pre-law parallel trends or anticipatory effects. First, fig. S3.5 shows that law intensity is only associated with the extensive margin of education for girls and women born after the introduc- tion of the law. Second, results are not susceptible to placebo bans introduced in 1993, 1995, and 1997 (see table S1.2). The second identifying condition for the estimation of α 1 , γ 1 , and β 1 requires ruling out confounding factors. Supplementary online appendix S1 shows that the educational reforms in Senegal that occurred 82 García-Hombrados and Salgado during the last decades and the large anti-FGC programme implemented by the NGO Tostan are unlikely to drive the results. Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 The identification of the 2SLS estimate of β 1 in equation (3) requires two additional conditions: that the instrumental variable is relevant and that it satisfies the exclusion restriction. The relevance condition re- quires that the parameter α 1 in equation (1) is large and statistically significant. The Results section shows that the statistical significance of α 1 largely satisfies standard relevant thresholds in most specifications. The exclusion restriction requires that the law should only affect educational investments through its effect on the prevalence of FGC. It is important to highlight that the law that banned FGC was a self- contained law and did not include specific provisions aimed at increasing female empowerment, economic conditions, or education. The Mechanisms section rules out the possibility that the effect of the law on education is caused by a broader effect of the new legislation on gender norms regarding education or by a shift in power in these communities and households towards individuals more favorable towards empowering women, which affects more intensely girls and women from FGC ethnic groups who were born after the introduction of the law. While the exclusion restriction is ultimately untestable and requires taking with caution the interpretation of the 2SLS estimates, it is difficult to think about potential channels through which a higher exposure to the law could affect education other than through reducing FGC. A final concern is error in measuring FGC since the DHS is either self-reported or reported by mothers, which is possible in settings where this cultural practice has been contested (De Cao and Lutz 2018). Misreporting could challenge the results biasing downwards the 2SLS and the estimated effect of the ban on the prevalence of FGC if the law affected the reporting of the practice among individuals exposed to the law. This is ruled out in supplementary online appendix S1. 5. Results 5.1. Effect of the Law on the Prevalence of FGC The first estimation is the effect of exposure to the law on the prevalence of FGC using the main analytical sample. This excludes the two ethnic groups for which the prevalence of FGC started decreasing for cohorts born before the introduction of the law: Soninke and the category “Other” ethnic group. These two groups account for 7 percent of the girls and women and excluding them from the sample ensures parallel pre-law trends in the prevalence of FGC and educational outcomes across ethnic groups. Panel B in table 2 reports the results of equation (1) using region fixed effects, then adding region-specific time trends, and finally using village fixed effects, which is our preferred estimation. The coefficient of the variable Intensity × PostLaw in all regressions shows that girls and women from ethnic groups with a higher pre-law incidence of FGC born after the introduction of the law experienced larger reductions in the probability of undergoing FGC than girls and women from less prevalent FGC groups. The F- statistics of the coefficient are larger than 10 in all specifications. The estimated reduction in the probability of undergoing FGC for girls and women that belong to an ethnic group with a 100 percent pre-law prevalence of FGC is nearly 25 percentage points in our preferred specification. At the ethnic group level, the mean prevalence of FGC among women born before the introduction of the law is 46 percent. For such an ethnic group, the estimated reduction in the prevalence of FGC caused by the law would be 11.5 percentage points. In a robustness check, the reestimation of the regressions using the sample that includes girls and women from all ethnic groups confirms the large effect of exposure to the law on the prevalence of FGC. Results are reported in table S3.1 in the supplementary online appendix. The magnitude and statistical significance of the impact of the law on FGC in this analysis is similar to that reported in table 2. Overall, these results suggest that the law led to a reduction in the prevalence of this practice. The World Bank Economic Review 83 Table 2. Impact of the Law on Female Genital Cutting (FGC) and on the Extensive Margin of Educational Investments (1) (2) (3) (4) (5) (6) (7) Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Dependent variable: never in school (0/1) OLS RF SS RF SS RF SS Panel A: reduced form and second stage Intensity × PostLaw — −0.099*** — −0.051* — −0.061*** — (0.022) (0.027) (0.018) FGC 0.081*** — 0.416*** — 0.307** — 0.263*** (0.008) (0.094) (0.146) (0.059) Mean dep. var. 0.374 0.374 0.374 0.374 0.374 0.374 0.374 Panel B: first stage. Dependent variable: prevalence of FGC Intensity × PostLaw — — −0.237*** — −0.166*** — −0.246*** (0.054) (0.050) (0.056) F — — 19.33 — 10.99 — 19.33 Mean dep. var. — — 0.318 — 0.318 — 0.318 Regional dummies Yes Yes Yes Yes Yes No No Regional time trends No No No Yes Yes No No DHS cluster dummies No No No No No Yes Yes N 26,517 26,517 26,517 26,517 26,517 26,517 26,517 No. ethnic groups 6 6 6 6 6 6 6 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: The analysis is conducted using the restricted sample of ethnic groups with pre-law parallel trends. All the regressions include as control variables a vector of year of birth × survey round dummies and a vector of ethnic group dummies. Standard errors clustered at the ethnic group × year of birth level using wild bootstrapping are reported in parentheses. Sample weights are not used in the analysis because they are not designed for representativeness of the analytical sample. *** p < 0.01, ** p < 0.05, * p < 0.1. OLS, Ordinary Least Square naive estimation; RF, Reduced form estimation; SS, Second Stage estimation. 5.2. Effects of the Law on Educational Investments and 2SLS Results Under certain assumptions, the predicted values of FGC can be used as an instrumental variable to esti- mate of the effect of FGC on women’s education. Columns 2, 4, and 6 in Panel A of table 2 report γ 1 from (2) using the main analytical sample. The preferred estimation in column 6 reports that girls and women born after the introduction of the law that belong to an ethnic group with a 100 percent pre-law prevalence of FGC would experience a reduction in the probability of never having attended school of 6.1 percentage points. The estimated effect varies between 5.1 and 9.9 percentage points, depending on the specification. For the average ethnic group prevalence of 46 percent, the estimated reduction would be of 3 percentage points. Table S3.2 shows that the effects are consistently negative and statistically significant at conventional confidence levels when all ethnic groups are included in the analysis. In other words, the estimates on the extensive margin of education suggest that many parents reacted to the rise in the cost of FGC caused by the law by abandoning this practice and investing in their daughters’ education. Columns 1, 3, 5, and 7 of table 2 provide further insights about the link between FGC and the exten- sive margin of educational investments. The ordinary least-squares (OLS) estimates reported in column 1 show that, conditional on ethnic group, year of birth, survey round, and DHS cluster (equivalent to village of residence), girls and women who have experienced FGC are approximately 8 percentage points more likely to have never attended school. Further descriptive analyses are provided in tables S3.9 and S3.10, showing that FGC is associated with worse educational outcomes. However, these estimates should not be interpreted as the causal effect of FGC because the statistical association could be driven by unobservable factors, such as attitudes towards tradition. To overcome the endogeneity in the link between FGC and educational investments and to estimate the causal effect of FGC on education, the estimation procedure exploits the variation in the prevalence of FGC caused by across-ethnic-group variation in exposure to the law as an instrumental variable, as discussed in the previous section. If the ban on FGC only affected edu- 84 García-Hombrados and Salgado Table 3. Intensive Margin: Impact of Female Genital Cutting (FGC) on Education (1) (2) (3) (4) (5) (6) (7) (8) Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Years of educ≥1 Years of educ≥2 Years of educ≥3 Years of educ≥4 RF SS RF SS RF SS RF SS Intensity × PostLaw 0.054*** — 0.039** — 0.019 — 0.003 — (0.018) (0.017) (0.024) (0.026) FGC — −0.237*** — −0.193** — −0.103 — −0.020 (0.054) (0.081) (0.125) (0.170) Mean dep. var. 0.612 0.612 0.589 0.589 0.556 0.556 0.526 0.526 DHS cluster dummies Yes Yes Yes Yes Yes Yes Yes Yes N 24,060 24,060 21,726 21,726 19,900 19,900 17,945 17,945 No. ethnic groups 6 6 6 6 6 6 6 6 First-stage regression Intensity × PostLaw — −0.227*** — −0.200*** — −0.181*** — −0.148*** (0.051) (0.045) (0.041) (0.034) F — 19.82 — 19.73 — 19.51 — 19.05 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: The analysis is conducted using the restricted sample of ethnic groups with pre-law parallel trends. All the regressions include as control variables a vector of year of birth × survey round dummies and a vector of ethnic group dummies. Sample weights are not used in the analysis because they are not designed for representativeness of the analytical sample. Standard errors clustered at the ethnic group × year of birth level using wild bootstrapping are reported in parentheses. *** p < 0.01, ** p < 0.05, * p < 0.1. OLS, Ordinary Least Square naive estimation; RF, Reduced form estimation; SS, Second Stage estimation. cation through its effects on education, the 2SLS analysis provides the causal effect of FGC on educational outcomes. This is reported in columns 3, 5, and 7 of table 2. The coefficient of the variable FGC in these regressions suggests that FGC has a negative effect on the extensive margin of educational investments. If the exclusion restriction holds, the estimates of the 2SLS should be interpreted as the causal effect of FGC on education for the compliers, that is, for those women that were not cut because they have a higher level of exposure to the law. For them, an increase of 10 percentage points in the prevalence of FGC increases the probability of never having attended school by 2.6 percentage points in our preferred specification, reported in column 7. The estimates of this effect range between 2.6 and 4.2 percentage points, depending on the regression specification. In percentage changes, a decrease by 30 percent in the prevalence of FGC would reduce the probability of never having attended school by approximately 6–11 percent depending on the specification. The coefficient measuring the effect of FGC on the probability of never having at- tended school is statistically significant at conventional confidence levels in all specifications.10 Table S3.2 reports the analysis using the full sample of ethnic groups for which we find similar results. On the other hand, the results suggest limited effects of the law on the intensive margin of educational investments. Table 3 reports the estimates of equations (2) and (3) using the village fixed effects specifica- tion and whether the girl or woman has at least one, two, three, or four years of education as dependent variables. The use of these outcome variables imposes constraints on the minimum age of girls who are included in the analysis. For example, for the estimation of the effects on having three or more years of education, we only include in the analytical sample girls who are 10 or older (rather than 7), because this is the age at which girls who started school on time should have had three years of education. The 10 The coefficient reported in columns 1 (OLS) and those reported in 3, 5, and 7 (IV) are not comparable since in the OLS regression the explanatory variable is a dummy variable while the explanatory variable in the second stage of the IV is a continuous predicted probability. It is not possible to conclude anything on the correlation between FGC and omitted variables comparing the OLS and IV coefficients. Furthermore, differences between OLS and IV coefficients might also be caused by measurement error in FGC in the OLS regression. The World Bank Economic Review 85 Figure 3. Summary of the Effects of the Law and of Female Genital Cutting (FGC) on the Probability of Receiving At Least X Years of Education Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: The graphs show the reduced form estimates of the effect of the law on the probability of receiving X years of education and the second-stage estimates of the effect of FGC on the probability of receiving X years of education. results of these regressions suggest that the law increased the probability of receiving one or two years of education. However, the effect of the law on the probability of receiving three or more years of education was much smaller and statistically indistinguishable from 0. The strong gradient of the effect is graph- ically illustrated in fig. 3. We provide further descriptive evidence of this pattern in figs S3.5 and S3.6. Figure S3.5 shows that while the variable that measures law intensity is not associated with educational outcomes for those girls and women born before the introduction of the law, it is strongly associated with the variables never in school and at least one year in school for women born after the introduction of the law. The correlation then decreases and completely vanishes for the variables measuring the probability of having at least two, three, or four years of education. Finally, fig. S3.6 shows consistent patterns display- ing the survival in school for four different groups: females born before the introduction of the law from FGC groups, females born after the introduction of the law from FGC groups, females born before the introduction of the law from FGC groups, and females born after the introduction of the law from FGC groups. Additionally, we reestimate the intensive margin analysis using cluster, region fixed effects, and region-specific time trends for both samples in tables S3.3, S3.4, and S3.5. The results of these analyses are consistent with those reported in table 3, showing statistically significant results only for the probability of receiving one or two years of education, and small and insignificant coefficients for the effect of the ban on the probability of receiving three or more years of education in most specifications. We extend the analysis of the effect of FGC on educational outcomes in supplementary online ap- pendix S1 through assessing the effect of the law and of FGC on current school attendance (at the time of 86 García-Hombrados and Salgado Table 4. Impact of the Law on Empowerment of Women in Households of Girls Affected by the Law (1) (2) (3) (4) (5) (6) (7) Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Participation in decisions on Never in Years of Own health Large household Family Husband Participation school education≥4 care (0/1) purchases (0/1) visits (0/1) earns (0/1) index (0–4) Intensity × PostLaw −0.019 0.042 0.057 0.069 0.017 0.053 0.203* (0.028) (0.028) (0.039) (0.043) (0.037) (0.033) (0.116) Mean dep. var. 0.612 0.338 0.223 0.195 0.254 0.148 0.810 DHS cluster dummies Yes Yes Yes Yes Yes Yes Yes N 29,773 29,773 21,816 21,816 21,816 21,471 21,471 No. ethnic groups 6 6 6 6 6 6 6 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: All the regressions include as control variables a vector of year of birth × survey round year dummies for the older female member of the household and for the girl or women affected by the law, a vector of ethnic group dummies, and a vector of DHS cluster fixed effects. Standard errors clustered at the ethnic group × year of birth level using wild bootstrapping are reported in parentheses. Sample weights are not used in the analysis because they are not designed for representativeness of the analytical sample. ****** p < 0.01, ** p < 0.05, * p < 0.1. the survey) for women and girls aged 7–17, number of years of education, and age for grade. The results of these analyses are reported in tables S1.6 and S3.7 and reveal consistent positive effects of the law on school attendance outcomes, number of years of education, and age for grade outcomes. In supplementary online appendix S1, we rule out alternative explanations for the main results. We test further the parallel trends condition, assess whether the effects of the law might be confounded by other interventions implemented around the same time in Senegal, examine whether the results might be driven by across ethnic group differences in age at starting of school age, and the extent to which misreporting of FGC might be driving the results. These results presented in the supplementary online appendix reassure our confidence in the main conclusions of the study. 6. Mechanisms 6.1. Change in Education Social Norms, Health, and Age of Marriage Although the awareness campaign that followed the introduction of the law was focused on the sanctions, the law was introduced in a context of an increasing awareness of women’s rights. Thus, one may argue that the law itself may have led to a broader change in gender norms. If the law changed gender norms favoring investment in daughters’ education in households of girls affected by the law, the 2SLS estimates would violate the exclusion restriction. While the law could have changed gender norms, this would only be problematic if this affected partic- ularly the households of girls from FGC groups born after the introduction of the law. We believe this is unlikely. The cohorts of girls and women born in the year of the law only started school seven years after the introduction of the law. Thus, several cohorts of girls and women born before the introduction of the law started school after the introduction of the anti-FGC legislation. The lack of educational effects for girls and women born before the introduction of the law but that started school after it is evidenced in table S1.2. We examine the effect of the law on the empowerment outcomes of women who live in house- holds with girls or women exposed to the law. Table 4 shows no effect on any of the individual education or empowerment indicators of female members of households with girls exposed to the law, except for the aggregate index of empowerment, which is marginally significant at the 10 percent confidence level. This indicates that the effects of the law on education do not seem to be driven by the law promoting a differential change in gender norms or a shift in power among families of girls and women born after the introduction of the law who belong to ethnic groups in which the practice of FGC is traditional. We The World Bank Economic Review 87 reexamine this test in table S1.13 using all DHS databases available, including those without information on FGC (e.g. 2005, 2008, 2006, 1997, and 1992) and find reassuring results. Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 FGC may impact education by affecting the age at which women are married. In the context of a competitive marriage market where parents prefer to marry their daughters as soon as possible after menarche (Wahhaj 2015), girls and women who have undergone FGC could be in a better position to marry younger. Because early marriage could lead to early school dropout (Field and Ambrus 2008), FGC could also reduce educational investments by lowering the age of marriage. We cannot test the direct effect of the law on age at marriage with the data available because informa- tion on marital status is only available for a very small subsample of females born after the introduction of the law and the vast majority of these girls remain unmarried.11 However, although FGC may decrease the age of marriage, we believe this mechanism is unlikely to drive the observed educational effects of the ban. The results presented in the Results section suggest that the ban has a limited effect on the intensive margin of educational investments. Specifically, the estimates reveal no effect of the law on the probability of receiving three or more years of education, which should in principle be completed by the age of 10. Given the small impact on the intensive margin of education and the fact that only 5 percent of women were cohabiting with their partner by the age of 14, it seems unlikely that the increase in educational investments caused by the law is driven by women who have not undergone FGC marrying at older ages. Alternatively, if health during childhood is a strong determinant of educational achievements and even- tually reduces returns to education (e.g. Grantham-McGregor et al. 2007), FGC could affect educational investments and school dropout rates by severely affecting the health of women and girls. Although the literature on the health consequences of FGC is mixed, most studies suggest that FGC is associated with worse sexual and reproductive health outcomes, such as a higher prevalence of sexually transmitted dis- ease, painful intercourse, and difficulties during childbirth (Berg et al. 2014; Wagner 2015). In light of the evidence of the health consequences of FGC, we cannot dismiss the possibility that the effect of FGC on education may be driven by the lasting effects of FGC on health, eventually leading to lower educational achievements. The most straightforward way to test the health mechanism would be to reestimate equations (1), (2), and (3) using health outcomes as dependent variables. However, the data do not provide health infor- mation on anthropometric measurements, incidence of anemia and diarrhea for girls older than 5 years old, implying that we do not observe key health outcomes for children born before the introduction of the anti-FGC law. To cope with this limitation, we add to the sample rounds 2005 and 1997 of the Sene- galese DHS and rounds 2008 and 2006 of the Senegalese Malaria Indicator Survey (MIS). These databases include information on ethnic affiliation and on health outcomes for girls born before and after the in- troduction of the law that banned FGC. However, these rounds of the DHS and the MIS do not include appropriate information on FGC status, which limits the analysis to reduced-form estimations. To test the health mechanism, we estimate our reduced-form equation (2) using as dependent variables weight, height, anemia, diarrhea, and whether the girl has a health card, which aims to measure health status and investments. The health variables analyzed correspond to the health-related questions that are available for girls aged below 5 in all survey rounds included. The results of this analysis are reported in table 5. The coefficients that measure the effect of the law on weight, anemia, diarrhea, and the probability of having a health card are all statistically indistinguishable from 0 at conventional confidence levels. Only the height is statistically significant at 5 percent. Overall, the results suggest that the law had limited effects on the health outcomes analyzed. However, the young age of most girls and women born after the introduction of the law hampers the assessment of the effect of the ban on sexual and reproductive health outcomes, which have been found to be associated with FGC. 11 Information on marital status in the DHS survey is only available for women aged 15 or older. Only 70 women in our sample of females born after the introduction of the law were ever married at the time of the survey. 88 García-Hombrados and Salgado Table 5. Impact of the Law Banning Female Genital Cutting (FGC) on Health Outcomes and Investments (1) (2) (3) (4) (5) Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Ln weight Ln height Not Diarrhea in last (kg 1 dec) (cm 1 dec) anemic (0/1) 24 h(0/1) Health card (0/1) Intensity × PostLaw 0.001 0.016** −0.037 −0.054 0.001 (0.029) (0.008) (0.035) (0.034) (0.077) Mean dep variable 177 965 0.328 0.175 0.905 DHS cluster FE Yes Yes Yes Yes Yes N 13,965 13,763 15,158 19,656 18,413 No. ethnic groups 6 6 6 6 6 Source: Own calculations using rounds 1997, 2005, 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys and rounds 2006 and 2008 of the Senegalese Malaria Indicator Survey (MIS). Note: Analytical sample restricted to the ethnic groups with parallel trends in terms of FGC among women born before the introduction of the law. All the regressions include as control variables a vector of year of birth × survey round dummies, a vector of ethnic group dummies, and a DHS cluster fixed effects. Standard errors clustered at the ethnic group × year of birth level using wild bootstrapping are reported in parentheses. Sample weights are not used in the analysis because they are not designed for representativeness of the analytical sample. *** p < 0.01, ** p < 0.05, * p < 0.1. Indeed, one could argue that the effect of the law on education is caused by teenage girls who have not undergone FGC staying in school because of improved sexual and reproductive health outcomes, and this might not be reflected in the health variables analyzed. However, most of the effects of FGC on sexual and reproductive health (e.g. irregular menstruation, painful intercourse) are unlikely to affect girls’ education before their teenage years. Hence, the fact that the law has an impact on the probability of never having attended school but little or no impact on the probability of finishing three or more years of education (which should be completed well before teenage years) suggests that the beneficial effect of reducing FGC on educational investments observed does not seem to be driven by improving sexual and reproductive health. 6.2. Marriage Markets We propose a fourth mechanism: marriage markets. Various qualitative studies investigate the role that FGC plays in them. Kolawole and Anke (2010) and Karumbi and Jacinta (2017) conclude that in com- munities where FGC is common, it increases marriage payments received by the bride’s parents. Wagner (2015) finds that FGC is associated with a higher probability of marriage. Boyden, Pankhurst, and Tafere (2013) conclude that FGC improves marriage prospects in Ethiopia and Mackie (1996) documents that, in some societies, FGC works as a precondition of marriage. Although FGC and education might signal different features in the marriage market and they might be lead to marrying different types of husbands, both increase ceteris paribus the bride price (Ashraf et al. 2020; Hotte and Lambert 2020). Thus, non-altruistic parents of future brides may react to the law— interpreted as an increase in the cost of FGC—by abandoning FGC and investing in their daughters’ education to compensate for future lower bride price. The fact that both education and FGC increase bride price does not mean that they work as perfect substitutes in the marriage market. The broader role of FGC and education in the marriage market might not be the same, since married women with higher levels of education and with FGC might have partners with different preferences regarding education and FGC. The latter however does not invalidate this mechanism if parents of the bride are not perfectly altruistic and FGC and education both increase the payment received by the parents of the bride from the family of the groom, assumptions that are well backed by the anthropological and economics literature. An additional requirement of this mechanism is that not all the girls born after the introduction of the law were uncut because in a marriage market where all women would be uncut, there would arguably not be much incentives to raise investments in daughters’ education for uncut girls. This was however not the case for Senegal. Although the law reduced the prevalence of FGC, the practice was far from being The World Bank Economic Review 89 eradicated among girls born after the introduction of the law. Furthermore, and since FGC is mostly a private event in Senegal (Shell-Duncan et al. 2013), parents are probably uncertain about whether other Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 parents in their community cut their daughters. The vast majority of the girls born after the introduction of the law remained unmarried at the time of the survey, which prevents the estimation of the effect of FGC on marriage-market outcomes. Given the impossibility of testing the effects of the law on marriage-market outcomes, an alternative strategy to test this mechanism is examining the effect of the law on educational investments for households for which FGC does not play an important role in the marriage market. If parents increase educational investments to compensate for worse marriage-market outcomes among daughters who have not undergone FGC, we would not observe any effect of the law on education for the subsample of households where FGC is not linked to marriage markets. Although the 2010–2016 rounds of the Senegalese DHS lack information on the perceived benefits of FGC, we investigate this by exploring the heterogeneous effects of FGC on education for different samples that differ in terms of the value of FGC in the marriage market. More specifically, we explore whether the effect of FGC on education is different in rural and urban areas, where FGC has a different value in the marriage market. Because the value of FGC in the marriage market is higher in rural areas, a larger effect of FGC on educational outcomes in rural settings would be consistent with the marriage-market hypothesis. But why is the value of FGC in marriage markets higher in rural than in urban areas? Mackie (2000) argues that urbanization enlarges marriage opportunities, reducing the negative consequences in the marriage market of not having undergone FGC. Similarly, Orubuloye et al. (2000) conclude that the higher value of FGC in rural marriage markets relative to urban marriage markets is explained by the fact that social conventions are less enforceable in urban areas. Also in this line, Kandala and Komba (2015) show that the practice of FGC in Senegal is a social norm that is more embedded in the culture around marriage in rural settings. Table 6 shows the results of this analysis. The coefficients reported in columns 1–4 indicate that expo- sure to the law had a large effect on the prevalence of FGC both in rural and urban areas. However, while the effect of FGC on the probability of never having attended school is strong in rural areas, the magnitude of the impact is smaller and statistically indistinguishable from 0 at conventional confidence levels in ur- ban areas. On the other hand, we do not observe any effect of FGC on the probability of receiving four or more years of education, for either rural or urban households. Taken together, these results are consistent with the hypothesis that the effect of FGC on educational investments is driven by parents responding to the law by abandoning FGC and investing instead in the education of their daughters, who are not subjected to FGC, to minimize potential losses in the marriage market. While suggestive, these results cannot be interpreted as conclusive evidence for this mechanism. More research is needed, using surveys with information on both FGC and bride price to conclusively test this question. While we are not aware of the existence of appropriate data to test this hypothesis, we formalize the marriage-market mechanism proposed in this section in supplementary online appendix S2, providing a theoretical framework for the study of the effect of FGC on the marriage market in future studies. The marriage-market mechanism would be consistent with the small impact on the intensive margin of education found in our analysis if one or two years of education already affect bride price and marriage- market outcomes. Is this a feasible hypothesis? Hotte and Lambert (2020) find that, conditional on many socioeconomic characteristics, some primary education is associated with a higher bride price relative to no education. Ashraf et al. (2020) find similar results in Zambia and Indonesia. We complement these analyses, exploring the adjusted correlation between one or two years of education and marriage-market outcomes. While we do not have information on bride price, the results reported in table S3.8 show that one or two years of education is associated with improved marriage-market outcomes. Overall, the ex- isting evidence and our suggestive analysis indicate that one or two years of education might have an effect on marriage-market outcomes and bride price. However, why do parents not invest even more in 90 García-Hombrados and Salgado Table 6. Impact of Female Genital Cutting (FGC) on Education: Heterogenous Effects Urban sample Rural sample Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 (1) (2) (3) (4) (5) (6) FS: prevalence RF: never in SS: never in FS: prevalence RF: never in SS: never in FGC (0/1) school school FGC (0/1) school school Panel A: never in school (0/1) Intensity × PostLaw −0.312*** 0.003 — −0.213*** −0.102*** — (0.070) (0.027) (0.048) (0.023) FGC — — −0.010 — — 0.480*** (0.085) (0.109) F 19.81 — — 19.75 — — DHS cluster FE Yes Yes Yes Yes Yes Yes Mean dep. var. 0.244 0.193 0.193 0.349 0.474 0.474 N 9,434 9,434 9,434 17,083 17,083 17,083 No. ethnic groups 6 6 6 6 6 6 FS: prevalence RF: years of SS: years of FS: prevalence RF: years of SS: years of FGC (0/1) educ≥4 educ≥4 FGC (0/1) educ≥4 educ≥4 Panel B: years of education≥4 (0/1) Intensity × PostLaw −0.198*** −0.017 — −0.124*** 0.015 — (0.060) (0.035) — (0.028) (0.029) FGC — 0.084 — −0.122 (0.177) (0.229) F 10.88 — — 19.51 — — DHS cluster FE Yes Yes Yes Yes Yes Yes Mean dep. var. 0.274 0.711 0.711 0.396 0.412 0.412 N 6,827 6,827 6,827 11,118 11,118 11,118 No. ethnic groups 6 6 6 6 6 6 Source: Own calculations using rounds 2010, 2012, 2014, 2015 and 2016 of the Senegalese Demographic and Health Surveys (DHS). Note: Analytical sample restricted to the ethnic groups with parallel trends in terms of FGC among women born before the introduction of the law. All the regressions include as control variables a vector of year of birth × survey round dummies, a vector of ethnic group dummies, and DHS cluster of residence fixed effects. Standard errors clustered at the ethnic group × year of birth level using wild bootstrapping are reported in parentheses. Sample weights are not used in the analysis because they are not designed for representativeness of the analytical sample. *** p < 0.01, ** p < 0.05, * p < 0.1. the education of their daughters? While more research is needed to provide a conclusive answer, poten- tial reasons include a higher opportunity cost of education among older girls or diminishing returns to education in the marriage market. 7. Conclusions We show that the introduction of ban on FGC in Senegal reduced the prevalence of FGC and improved the extensive margin of educational attainment of women and girls. Although the law was not sufficient to eradicate this practice, it significantly contributed to reducing it by approximately 25 percent. Our results provide evidence supporting the introduction of anti-FGC bans in the countries where it is widespread but still not regulated. The Senegalese example of sanctioning a few media covered high-profile cases suggest an effective implementation strategy in settings where FGC is widespread and the institutional capacity to enforce the law is limited. The results of the paper also show that bans on this practice could have beneficial consequences in terms of human capital accumulation among women, contributing also to close the gender gap in education. On the other hand, we find little effect of the law on the intensive margin The World Bank Economic Review 91 of education, which is consistent with the beneficial effects on bride price of receiving some education, documented by Ashraf et al. (2020) and Hotte and Lambert (2020). Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 The stronger effects of the law on education in areas where FGC is less appreciated in the marriage market are compatible with the hypothesis developed in anthropological studies that some parents per- ceive FGC and educational investments as pre-marital investments and they open the debate on whether interventions aiming to lower the value of FGC in the marriage market or to change the relative price of alternative pre-marital investments could be effective strategies to reduce the prevalence of FGC. Addi- tional research examining empirically the causal effects of FGC on the marriage market in Senegal would be needed to test comprehensively the substitution hypothesis. In supplementary online appendix S2, we construct a theoretical model that relies on anthropological studies where FGC works as a pre-marital investment that can guide future analysis. Lack of suitable data and the low age of girls born after the introduction of the law prevented the estimation in this study of the effects of FGC on marriage-market outcomes. 8. Data Availability The data underlying this article are available in the article and in its online supplementary material. References Acemoglu, D., and M. O. Jackson. 2017. “Social Norms and the Enforcement of Laws.” Journal of the European Economic Association 15(2): 245–95. Allen, C., K. Allen, N. Davies, A. Hurn, D. Marshall, Y. Middlewick, and E. Niena et al. 2015. “Country Profile: FGM in Senegal.” Technical report, 28 Too Many. Ashraf, N., N. Bau, N. Nunn, and A. Voena. 2020. “Bride Price and Female Education.” Journal of Political Economy 128(2): 591–641. Becker, A. 2019. “On the Economic Origins of Restrictions on Women’s Sexuality.” CESifo Working Paper Series 7770, CESifo. Bellemare, M. F., L. Novak, and T. L. Steinmetz. 2015. “All in the Family: Explaining the Persistence of Female Genital Cutting in West Africa.” Journal of Development Economics 116: 252–65. Berg, R. C., V. Underland, J. Odgaard-Jensen, A. Fretheim, and G. E. Vist. 2014. “Effects of Female Gen- ital Cutting on Physical Health Outcomes: A Systematic Review and Meta-Analysis.” BMJ Open 4(11): e006316. Boyden, J., A. Pankhurst, and Y. Tafere. 2013. Harmful Traditional Practices and Child Protection: Contested Un- derstandings and Practices of Female Child Marriage and Circumcision in Ethiopia (Publisher’s version). Young Lives. Cameron, A. C., J. B. Gelbach, and D. L. Miller. 2008. “Bootstrap-Based Improvements for Inference with Clustered Errors.” Review of Economics and Statistics 90(3): 414–27. Cetorelli, V., B. Wilson, E. Batyra, and E. Coast. 2020. “Female Genital Mutilation/Cutting in Mali and Mauritania: Understanding Trends and Evaluating Policies.” Studies in Family Planning 51(1): 51–69. Crisman, B., S. Dykstra, C. Kenny, and M. O’Donnell. 2016. “The Impact of Legislation on the Hazard of Female Gen- ital Mutilation/Cutting: Regression Discontinuity Evidence from Burkina Faso”. Center for Global Development, Working Papers 432. De Cao, E., and G. La Mattina. 2019. “Does Maternal Education Decrease Female Genital Cutting?” AEA Papers and Proceedings, 109: 100–04. De Cao, E., and C. Lutz. 2018. “Sensitive Survey Questions: Measuring Attitudes Regarding Female Genital Cutting through a List Experiment.” Oxford Bulletin of Economics and Statistics 80(5): 871–92. Diabate, I., and S. Mesplé-Somps. 2019. “Female Genital Mutilation and Migration in Mali: Do Return Migrants Transfer Social Norms?” Journal of Population Economics 32(4): 1125–70. Efferson, C., S. Vogt, A. Elhadi, H. E. F. Ahmed, and E. Fehr. 2015. “Female Genital Cutting Is Not a Social Coordi- nation Norm.” Science 349(6255): 1446–47. 92 García-Hombrados and Salgado Efferson, C., S. Vogt, and E. Fehr. 2020. “The Promise and the Peril of Using Social Influence to Reverse Harmful Traditions.” Nature Human Behaviour 4(1): 55–68. Downloaded from https://academic.oup.com/wber/article/37/1/74/6761947 by International Monetary Fund / World Bank - IMF user on 11 September 2023 Field, E., and A. Ambrus. 2008. “Early Marriage, Age of Menarche, and Female Schooling Attainment in Bangladesh.” Journal of Political Economy 116(5): 881–930. Garcia-Hombrados, J. 2022. “Child marriage and infant mortality: causal evidence from Ethiopia.” Journal of Popu- lation Economics 35: 1163–1223. Grantham-McGregor, S., Y. B. Cheung, S. Cueto, P. Glewwe, L. Richter, B. Strupp, and I. C. D. S. Group et al.. 2007. “Developmental Potential in the First 5 Years for Children in Developing Countries.” Lancet 369(9555): 60–70. Harari, M. 2019. “Women’s Inheritance Rights and Bargaining Power: Evidence from Kenya.” Economic Development and Cultural Change 68(1): 189–238. Rozenn, H., and S. Lambert. 2020. “Marriage Payments and Wife’s Welfare: All You Need is Love,” PSE Working Papers halshs-02611667, HAL. Kandala, N.-B., and P. N. Komba. 2015. “Geographic Variation of Female Genital Mutilation and Legal Enforcement in Sub-Saharan Africa: A Case Study of Senegal.” American Journal of Tropical Medicine and Hygiene 92(4): 838. Karumbi, J., D. Gathara, and J. Muteshi. 2017. “Exploring the Associations between FGM/C and Child Marriage: A Review of the Evidence.” Evidence to End FGMC Programme Consortium Report. New York: Population Council. Kolawole, A., and V. Anke. 2010. “A Review of Determinants of Female Genital Mutilation in Nigeria II.” Journal of Medicine and Medical Sciences 1(11): 510–15. La Ferrara, E., and A. Milazzo. 2017. “Customary Norms, Inheritance, and Human Capital: Evidence from a Reform of the Matrilineal System in Ghana.” American Economic Journal: Applied Economics 9(4): 166–85. Mackie, G. 1996. “Ending Footbinding and Infibulation: A Convention Account.” American Sociological Review 61(6): 999–1017. ———. 2000. “Female Genital Cutting: The Beginning of the End.” In B. Shell-Duncan & Y. Hernlund (Eds.), Female Circumcision in Africa: Culture, Controversy, and Change (pp. 253–82). Boulder, CO: Lynne Rienner. Michalopoulos, S., L. Putterman, and D. N. Weil. 2019. “The Influence of Ancestral Lifeways on Individual Economic Outcomes in Sub-Saharan Africa.” Journal of the European Economic Association 17(4): 1186–1231. Novak, L. 2020. “Persistent Norms and Tipping Points: The Case of Female Genital Cutting.” Journal of Economic Behavior & Organization 177(C): 433–74. Nunn, N. 2012. “Culture and the Historical Process.” Economic History of Developing Regions 27(sup-1): 108–26. Orubuloye, I., P. Caldwell, and J. C. Caldwell. 2000. “Female Circumcision Among the Yoruba of Southwest Nigeria: The Beginning of Change.” Female Circumcision in Africa. The Australian National University, Canberra, 73–94. Poyker, M. 2021. “Regime Stability and the Persistence of Traditional Practices.” Review of Economics and Statistics: 1–45. Shell-Duncan, B., K. Wander, Y. Hernlund, and A. Moreau. 2013. “Legislating Change? Responses to Criminalizing Female Genital Cutting in Senegal.” Law & Society Review 47(4): 803–35. United Nations Children’s Fund. 2016. “Female Genital Mutilation/Cutting: A Global Concern UNICEF.” New York. Vogt, S., N. A. Mohmmed Zaid, H. El Fadil Ahmed, E. Fehr, and C. Efferson. 2016. “Changing Cultural Attitudes towards Female Genital Cutting.” Nature 538(7626): 506–9. Wagner, N. 2015. “Female Genital Cutting and Long-Term Health Consequences–Nationally Representative Esti- mates across 13 Countries.” Journal of Development Studies 51(3): 226–46. Wahhaj, Z. 2015. “A Theory of Child Marriage.” Studies in Economics 1520, School of Economics, University of Kent. Yoder, P. S., and S Wang. 2013. Female Genital Cutting: The Interpretation of Recent DHS Data. DHS Comparative Reports No. 33. Calverton, Maryland, USA: ICF International.