Productivity, Rank, and Returns in Polygamy

Tertilt (2005) stated that more than 10 % of marriages in 28 countries of sub-Saharan Africa are polygamous and that more than one-half of males in Cameroon are married to more than one woman, for example. Polygamy is mostly associated with Muslim ethnic groups (Elbedour et al. 2002), and according to the Koran, a man may have a maximum of four wives, whom he has to support and to treat equally (Boserup 1970). This coincides with geographical variation of polygamy and its more common occurrence in the West than in the East of the African continent, which Dalton and Leung (2014) explained with historical patterns of slave trade.

In accordance with Uusimaa et al. (2007), a maternal nuclear family denotes a mother and her biological children. In this setting, this term refers exclusively to polygamous wives and their biological children.

The Oromo are a traditionally semi-nomadic pastoralist ethnic group mostly found in the South of the country, and 53 % of the polygamous wives in the present sample state Oromo as their ethnicity. The remainder are Gedeo (18 %), Kembata (6 %), Gamo (5 %), Gurage (4 %), and Other (14 %).

Women may not change type here. This is a reasonable assumption because for partner choice the productivity at the time of the wedding is important, and becoming older or less able to contribute to family income over the course of the marriage is therefore not critical.

Co-wives are reported to live autonomously with little cooperation in many African ethnic groups (Boserup 1970; Kazianga and Klonner 2009), including the Oromo in Ethiopia (Gibson and Mace 2007). Akresh et al. (2016) also reported co-wives to live autonomously with their children in their own nuclear units with little cooperation among co-wives, at least with respect to daily activities. Their findings did provide evidence, however, for co-wives cooperating with each other with respect to cultivating plots in which the household head is not involved. Cooperation between wives, especially when the household head is not involved, is not influential for the selection into rank and therefore is not incorporated in this study.

Bars and underbars denote the concerned variables and functions for high- and low-productivity wives, respectively, throughout this section.

See Becker (1973) for a thorough discussion of the gain generated by marriage.

A noncooperative equilibrium within marriage, as suggested by Lundberg and Pollak (1993), is not considered here.

Productivity is very likely to be positively correlated with bargaining power, thereby reinforcing the effect of productivity on controlled shares of family income. Because of this reinforcement of the effect and the aim of isolating the direct effect of productivity on shares of family income controlled by the different types of spouses, the relationship between productivity and bargaining power is ignored. Were this relationship included, highly productive spouses would receive an even larger share of family income relative to less-productive spouses as a result of both better outside options and higher relative bargaining power. For the same reason, the resources invested by the household head are assumed to be identical across wives. If men allocated their resources efficiently and invested more resources the more productive a wife was, the conclusion of the shares of income being controlled by wives being dependent on productivity would be reinforced, and the effect of productivity would therefore be less obvious.

This assumption is reasonable given that being an unmarried woman is not socially accepted or economically sustainable in many rural settings of developing economies, especially among Muslim ethnic groups in which polygamy mostly occurs (Elbedour et al. 2002).

The probability of a husband attracting a second wife is identical among men and is considered in detail in the following section. Because there are more women than men, as assumed here, and because women’s returns when single are very low, it is not a profitable strategy for women to remain single and marry a single man in the second period given that they risk remaining unmarried in the long run, however. Relaxing the restrictions on the number of periods and the number of wives per husband could make this a profitable strategy, but it also implies that women cannot anticipate the number of co-wives. Again, waiting is not profitable because of the risk of remaining single. Similarly, first wives are not able to demand compensation for their husbands entering a second marital union: negotiation of a compensation before their marriage would leave the woman unmarried in the first period as a result of the surplus of women, and negotiation of more than the controlled share of family income determined earlier is not possible in this setting because her outside option and bargaining power remain identical.

If the fractions of joint output z _N1 and therefore (1 − z _N1) are not assumed to be fixed for the second period (even though a second marriage lowers joint output Y ₁ because of lower male investment), women entering marriage in the first period would also prefer a low-productivity co-wife.

Whereas parents often arrange the first marriage, the second one often occurs as a love marriage. This difference should not play a critical role for the conclusions derived from this section, however. If the second wife were highly productive, she would be likely to have become a first wife of another man, rather than being available as a second wife. Furthermore, if love played a role in granting her a privileged status within the family with respect to the allocation of the family’s resources, the empirical finding motivating this study would not exist.

Fafchamps and Quisumbing (2005) stated that most marriages in Ethiopia are arranged and that economic factors appear to be one of the main determinants of partner choice because of the evidence of assortative matching with respect to wealth and human capital. Payments of bride prices are not directly modeled here but could be included as a separate payment in addition to female income. Because men are identical here, bride prices should reflect female productivity only and therefore should be higher for high-productivity women than for low-productivity women. However, they will not influence the conclusions drawn from the framework (1) if they are smaller than the marginal product of a woman in a marital union so that the incomes of both spouses still increase with marriage, and (2) if the difference in bride prices between different types of women does not exceed the difference in utility derived by the man from marriage, which would go against the motivation of bride prices to begin with. If the difference in bride prices were higher than the difference in male payoffs from marriage, men would prefer low-productivity wives because they would derive more net utility from them, which would conflict with the definition of the two types of women examined here to begin with.

This assumption is reasonable in this setting due to incidences of civil wars in which men constitute the majority of casualties due to their higher exposure to this specific danger and due to men being more likely to leave rural areas in search of employment opportunities. Furthermore, Gibson and Mace (2007) mentioned that women are often forced into polygamous marriages due to a surplus of women in the marriage market, which is also given as a motivation for polygamous marriage by Becker (1973), besides large inequalities among men (Becker 1974).

The implications of relaxing Assumption 2 are discussed in Appendix 1.

See Fafchamps and Quisumbing (2005) for a detailed description of the study area, the sampling strategy, and the survey design of Round 4, particularly regarding the module on the marital history of the household head and his spouse(s).

Unfortunately, the amount of usable data in response to some questions is limited—for example, with respect to the spouses’ educational attainment, which would be an ideal proxy for female productivity as introduced in the framework earlier. Besides the low response rate, one explanation for the fact that there are hardly any usable data on spouses’ education is the rural location of these households, in which most adults have not received any formal education.

Measurement error is apparent if the duration of the marriage and the year of the wedding do not add up to either year in the Ethiopian calendar in which the interviews for the fourth round of the survey were conducted, for example.

The main results are robust to using only the subsample of wives for whom the ranking does not need to be imputed on the basis of age.

Only polygamous wives are included in the sample. As shown in Table 1, the number of third and fourth wives is very small, so the mean-comparison tests are performed between first and second wives only. The fact that there are fewer second than first wives is due to missing/misreported data on variables used in the estimation and observations being dropped for this reason.

The very low minimum ages at marriage need not be cases of misreporting or recording errors. Especially in rural Ethiopia, a promise of a young girl’s parents for her to marry a man in the future may be counted as a “marriage” for the purpose of this question, even if the actual wedding may have only occurred much later.

Parents poor includes cases in which the parents were perceived to be “poor” or “very poor,” and parents rich includes parents that were perceived to be “rich” or “very rich,” given the rare mentioning of either extreme case.

Boserup (1970) mentioned that polygamy often leads to high bride prices and therefore to families marrying off their daughters at a young age. In addition, Meekers (1992) found that marriages are arranged at a very young age to ensure that girls have not engaged in extramarital sexual behavior.

Grouping junior wives is necessary here because of the very limited number of third and fourth wives and in accordance with some of the literature (Mammen 2009; Timæus and Reynar 1998; Strauss and Mehra 1990). To verify that first wives exhibit different characteristics than second wives at the time of the wedding, I replicate Table 2 and Table 4 in Appendix 2 with a subsample in which I exclude wives who are more junior than the second wife. The results (not presented here but available upon request) show that although the loss of observations puts strain on the sample, the results are robust to using this subsample.

Questions on other characteristics of the spouses measuring human capital (such as formal education, farming experience, bride prices, or the value of assets brought into marriage) exhibit drastically low response rates in the present sample and may therefore not be used as explanatory variables.

The results are robust to including region rather than district indicator variables.

Because the sample exclusively consists of polygamously married women, the results for being a junior wife as the dependent variable are simply the mirror image of the ones presented here.

Inclusion of binary variables indicating whether the husband perceived the parents of his spouse to be richer or of equal wealth as his natal family does not qualitatively alter the results. Furthermore, neither variable yields statistically significant coefficients. Results are not presented but are available from the author upon request. This absence of an effect is interesting given the findings of Fafchamps and Quisumbing (2005), who found evidence for assortative matching with the same data set. However, these authors looked at assortative matching in the marriage market with respect to physical and human capital of spouses, rather than of their natal families, and they did not specifically explore polygamous unions—especially not wife order.

For example, the exclusion restriction may be violated as the wealth of the mother’s parents may directly impact a child’s educational outcomes, especially when they are associated with monetary costs. Furthermore, the age of the mother at the time of the wedding may have a direct effect on her children’s educational outcomes rather than solely through the channel of affecting her rank among wives. To be specific, women who marry at a younger age may be more traditional and less educated, or may be more modern, and therefore place less or more emphasis, respectively, on their children’s education than women who marry at a later age.

The results are robust to estimating a linear probability model for school enrollment being the dependent variable.

Birth order, being the sibling of the firstborn or of the first son within the household, and the number of full siblings at school age instead of the total number of children in the household do not exhibit a statistically significant impact on children’s educational outcomes. Results including each of these control variables separately support the main results; these results are not presented but are available from the author upon request.

The results are supported, and I find no evidence for selection into school enrollment in a Heckman selection model, the results of which are not presented because the model is not well-specified.