African polygamy: Past and present

Highlights

• Colonial teachers reduce polygamy in the present.

• Colonial missions reduce polygamy in the present.

• Four natural experiments that expanded modern education did not reduce polygamy.

Abstract

I evaluate the impact of education on polygamy in Africa. Districts of French West Africa that received more colonial teachers and parts of sub-Saharan Africa that received Protestant or Catholic missions have lower polygamy rates in the present. I find no evidence of a causal effect of modern education on polygamy. Natural experiments that have expanded education in Nigeria, Zimbabwe, Sierra Leone and Kenya have not reduced polygamy. Colonial education and missionary education, then, have been more powerful sources of cultural change than the cases of modern schooling I consider.

Introduction

Polygamy and poverty are both widespread in Africa.¹ Several mechanisms have been proposed linking polygamy to slow growth, including low savings rates (Tertilt, 2005), reduced investment in girls' human capital (Edlund and Lagerlöf, 2006), and diminished labor supply of unmarried men (Edlund and Lagerlöf, 2012). In the “polygamy belt” stretching from Senegal to Tanzania, it is common for more than one third of married women to be polygamous (Jacoby, 1995). Excepting Haiti, polygamy is less common in other developing countries. In all other Demographic and Health Surveys (DHSs), at least 92% of married women are reported to be monogamous. This is despite a striking decline in the prevalence of polygamy in Africa over the last half century. In Benin, more than 60% of women in the sample used for this study who were married in 1970 are polygamists, while the figure for those married in 2000 is under 40%. This is also true of Burkina Faso, Guinea, and Senegal. Several other countries in the data have experienced similar erosions of polygamy. This is an evolution of marriage markets as dramatic as the rise in divorce in the United States or the decline of arranged marriage in Japan over the same period.

I use DHS data on 494,157 women from 34 countries to ask two questions. First, does increasing education reduce polygamy in Africa? Second, does the answer to this question differ between colonial and modern education? Theories of polygamy exist that suggest it will disappear as education becomes more prevalent (Gould et al., 2008). Similarly, female education and monogamy both correlate with general measures of female empowerment (Doepke et al., 2012); educating women may empower them to pursue more favorable marital outcomes.

I find that colonial education reduces polygamy. I use historic data from Huillery (2009) and Nunn (2014a) to show that schooling investments decades ago predict lower polygamy rates today. Within French West Africa, I show that former colonial districts that had more teachers per capita during the early colonial period display lower rates of polygamy today. This correlation is robust to the inclusion of several district characteristics as controls and to basing inference off comparisons of adjacent districts within narrow geographic neighborhoods. There is only limited evidence that observable characteristics of districts that predict greater receipt of colonial teachers also predict lower rates of polygamy. Similarly, parts of present-day Africa that are in closer proximity to colonial missions have lower rates of polygamy in the present. This is robust to several controls, increasingly stringent geographic fixed effects, different measures of proximity to missions, and comparison with immediately adjacent areas. There is no evidence that missions were located in areas that had less polygamy before colonial rule.

In the modern period, there exists a clear negative correlation between a woman's years of education and whether she is a polygamist. To test if this is causal, I exploit four natural experiments that have increased female education in Nigeria (Osili and Long, 2008), Zimbabwe (Agüero and Bharadwaj, 2014), Sierra Leone (Cannonier and Mocan, 2012), and Kenya (Chicoine, 2012). The Nigerian Universal Primary Education (UPE) program exposed specific cohorts of children from certain regions of the country to additional primary schooling. While Osili and Long (2008) use a difference-in-difference approach to show that this reduced fertility by age 25, I find no effect on whether women exposed to the program married polygamously. This is robust to state trends and to discarding Lagos from the analysis. Instrumental variables estimates using the intervention do not suggest a negative effect of a woman's education on her later polygamy.

In Zimbabwe, the end of white rule made higher levels of education available to black women. Using a regression discontinuity that compares cohorts just young enough to be treated by this change with their older peers, Agüero and Bharadwaj (2014) show that this improved their knowledge about HIV. By contrast, I find no effect on polygamy. This result is robust to changing the width of the window around the cutoff age and discarding cohorts just around the cutoff. Instrumental variables estimates are insignificant and, at their most negative, still suggest a causal effect that is smaller than the raw correlation observed in the data.

In Sierra Leone, a Free Primary Education (FPE) program benefited certain cohorts and varied in intensity over space. Cannonier and Mocan (2012) use a difference-in-difference approach to show that the program changed women's attitudes towards health and violence towards women. I find only weak evidence that the program affected polygamy. While my point estimates are consistently negative, they are insignificant and small in most specifications. Some robustness exercises suggest a significant negative effect of the program, but only if variation over space in program intensity is not used for identification. Otherwise, results are robust to alternative specifications of the control group, controlling for violence during the civil war, alternative coding of the treatment group, and restricting the sample to non-migrants.

In Kenya, a reform of the education system during the 1980s lengthened primary school, leading to an average increase in schooling attainment for affected cohorts. Identifying treatment effects using nonlinearities in exposure across cohorts, Chicoine (2012) finds that the reform led to reduced fertility and delayed marriage. I find no effects on polygamy. Instrumental variables estimates do not suggest a negative effect of years of schooling on polygamy. This is robust to restricting the sample of included cohorts, and there are similarly null effects for placebo cohorts unlikely to have been affected by the reform.

Together, these results suggest that different mechanisms must explain the nonzero effects of historical education and the null effects of modern education. Changes due to historical education accrued through the accumulation of benefits and externalities over time and the pressures for cultural change found in both secular and mission schools. The effects of modern schooling have been muted by a narrower focus, heterogeneous impacts that lack reinforcement, and the low quality of contemporary schooling.

My results contribute to our knowledge of the determinants of ethnic institutions. Institutions such as pre-colonial states and land tenure matter for modern incomes (Goldstein and Udry, 2008, Michalopoulos and Papaioannou, 2013). Although an empirical literature has explained national institutions as products of influences such as settler mortality, population, trade, or suitability for specific crops, less is known about the origins of ethnic institutions. Like national institutions, these may have been shaped by geographic endowments such as ecologically-driven gains from trade, by population pressure, or by colonization (Acemoglu et al., 2014, Fenske, 2014, Osafo-Kwaako and Robinson, 2013). I add to this literature by testing hypotheses about the origin of one specific ethnic institution, and by identifying variables that influence its persistence and evolution.

My results also add to our understanding of family structures. Several recent contributions have explained marriage patterns using the gender division of labor created by influences such as the plow (Alesina et al., 2013), animal husbandry (Voigtländer and Voth, 2013), natural resource wealth (Ross, 2008), or deep tillage (Carranza, 2014). Other views link marital rules to risk-sharing arrangements (Rosenzweig and Stark, 1989). Part of this literature has drawn on economic models, history, and anthropology to discuss the causes of polygamy, stressing variables such as male inequality (Barber, 2008, Bergstrom, 1994, Betzig, 1982, Betzig, 1992, Betzig, 1995, Kanazawa and Still, 1999), the gender division of labor (Boserup, 1970, Jacoby, 1995), political economy (Anderson and Tollison, 1998, De la Croix and Mariani, 2012, Lagerlöf, 2010), the slave trade (Dalton and Leung, 2014, Edlund and Ku, 2011, Thornton, 1983), the gender ratio (Becker, 1974, White and Burton, 1988), land quality (Korn, 2000), animal husbandry (Adshade and Kaiser, 2008), son-preference (Milazzo, 2014), abundance of land (Goody, 1973, Goody, 1976), assortative matching (Siow, 2006), and fertility preferences (Grossbard-Shechtman, 1986). Another branch of this literature has examined the implications of polygamy for outcomes such as cooperation (Akresh et al., 2011) or economic growth (Tertilt, 2005). While many models of polygamy suggest that it benefits women, its prevalence across countries is negatively correlated with other measures of female empowerment (Doepke et al., 2012). In this paper, I uncover a dramatic transition in the continent's marriage markets, and assess one plausible explanation for this change.

Finally, I add to a literature that tests the effects of education on female empowerment. The policy changes I exploit have each been used in previous studies to test whether educating women shapes their later fertility, health, and beliefs (Agüero and Ramachandran, 2010, Bhalotra and Clarke, 2013, Cannonier and Mocan, 2012, Chicoine, 2012, Osili and Long, 2008). Other studies have used similar natural experiments or randomized control trials to test whether educating women improves outcomes for them. In some cases, providing women with education or skills can have large beneficial effects, even compared to the impacts on men (Bandiera et al., 2014, Blattman et al., 2014). Many of these studies, however, have found effects that are weak or heterogeneous (Duflo, 2012, Friedman et al., 2011). This heterogeneity requires explanation. Although polygamy is a marital outcome for millions of African women, I am not aware of any study that has attempted to identify a causal effect of education on polygamy, in either experimental data or through the use of a natural experiment.

The question of historical persistence cuts across all three of these literatures. A large literature has recently demonstrated that several past events, investments, and institutions have had effects that remain visible today (Bruhn and Gallego, 2012, Dell, 2010, Nunn, 2014b). My results provide an unusual example in which not only does history matter, but in which it matters more than policy initiatives in the present day.

I introduce the multiple data sources that I use and describe broad patterns of polygamy and education in Section 2. I demonstrate the persistent effects of colonial education in Section 3 and test for effects of modern education in Section 4. I discuss the mechanisms explaining these effects in Section 5. In Section 6, I conclude.