We study the relationship between polygyny and HIV infection using nationally representative survey data with linked serostatus information from 20 African countries. Our results indicate that junior wives in polygynous unions are more likely to be HIV positive than spouses of monogamous men, but also that HIV prevalence is lower in populations with more polygyny. With these results in mind, we investigate four explanations for the contrasting individual- and ecological-level associations. These relate to (1) the adverse selection of HIV-positive women into polygynous unions, (2) the sexual network structure characteristic of polygyny, (3) the relatively low coital frequency in conjugal dyads of polygynous marriages (coital dilution), and (4) the restricted access to sexual partners for younger men in populations where polygynous men presumably monopolize the women in their community (monopolizing polygynists). We find evidence for some of these mechanisms, and together they support the proposition that polygynous marriage systems impede the spread of HIV. We relate these results to the debate about partnership concurrency as a primary behavioral driver for the fast propagation of HIV in some parts of sub-Saharan Africa.
Polygyny and HIV: A Review of Prior Evidence
Polygyny was frequently featured in early scientific debates on the epidemiology of HIV (Gausset 2001; Oppong and Kalipeni 2003), but it is largely absent from the contemporary literature. This may be due to the difficulty of reconciling claims about polygyny as a risk factor for HIV infection with data illustrating that HIV prevalence is particularly low in countries where polygyny is most common. Descriptions of polygyny as a harmful cultural practice are nonetheless pervasive in current popular and policy discourses. In Malawi, a bill has been proposed to ban polygyny (Nyasa Times2010), and the southern African Development Community labeled polygyny a “major contributing factor to HIV” (Lusaka Times2010). A third example comes from a UNESCO advocacy booklet for Zambia (Fig. 1); it features two women who contemplate the dangers of polygyny for the spread of HIV in their community. Resourceful as they are, they campaign in their village, and—all’s well that ends well—the elders eventually decide to do away with the backward practice of polygyny. The script in the booklet is one in which the awareness of a few women translates into the swift abandonment of polygyny. In practice, these policy initiatives—reminiscent of missionary efforts in the colonial period—have triggered hostile reactions because they contravene religious prescripts and challenge cultural practice (BBC 2010; CNN 2010).
Despite these claims and policy initiatives, there has been little or no systematic analysis of the implications of polygyny for the spread of HIV. An in-depth analysis of polygyny is also timely because it allows us to evaluate the proposition that concurrent sexual partnerships are an important driver of the HIV epidemic in sub-Saharan Africa (Epstein 2007; Halperin and Epstein 2004; Hudson 1993; Morris and Kretzschmar 1997; Watts and May 1992). Polygynous marriage is a special form of partnership concurrency in the sense that it is institutionalized, the partnerships are relatively long-term, and it is gender-asymmetric (only men have more than one partner). Polygyny is also common: in 33 African countries with recent data, the percentage of married women (ages 15–49) with co-wives ranged from 3% in Madagascar, where it is prohibited by law, to 53% in Guinea. The median value is about 23% (Reniers and Watkins 2010).
There are two important reasons why partnership concurrency could trigger large-scale HIV epidemics. First, a monogamous relationship traps the virus until it dissolves and new partnerships are formed. The sequential nature of relationships in (serial) monogamy thus acts as a buffer to reduce the epidemic spread (Moody 2002; Morris and Kretzschmar 1997). Second, an HIV-positive person’s viral load, and therefore his or her infectiousness, peaks in the first couple of months after seroconversion (Boily et al. 2009; Wawer et al. 2005). Someone who just acquired HIV is therefore more likely to pass the virus to someone else if he or she has a concurrent seronegative partner. Under serial monogamy, the gap between HIV acquisition and sexual intercourse with a new partner is longer and thus less likely to occur in the highly infectious window period. The interaction between primary infection and partnership concurrency is corroborated by simulation studies (Eaton et al. 2011; Goodreau et al. 2012).
An important, yet often misunderstood repercussion of partnership concurrency is that it only affects the probability of transmitting HIV, not the probability of acquiring HIV (Morris 2001). Assuming fixed partnership duration, coital frequency, and partner characteristics, a man with three concurrent partners over a span of 10 years will have the same risk of acquiring HIV as a man who has these partners in sequence. Should he acquire HIV, however, he can transmit the virus to only one other person if he has his partners in sequence and was infected by his second partner, but he can transmit HIV to both other partners under the scenario of partnership concurrency (provided that the two other partners are HIV negative). In other words, the concurrency hypothesis predicts a positive correlation between the index case’s concurrency and the HIV status of his partners (but not his own HIV status). This last point has important methodological implications because it means that individual-level and ego-centered studies of HIV risk factors cannot detect individual-level concurrency effects, and these are precisely the most commonly available data sources.
Aside from its merits and the new insights it produced, the concurrency hypothesis merely revolves around one of several network effects that are relevant for the spread of a sexually transmitted infection (STI). As we will argue below, the practice of partnership concurrency is associated with very specific personal, partnership, and sexual network attributes that could offset potentially adverse effects of concurrency. Actors in concurrent partnerships may also make volitional behavioral adjustments to compensate for the elevated risk of infection. In addition, any assessment of the concurrency hypothesis should acknowledge that it is primarily supported by simulation studies. To date, the empirical support for the concurrency hypothesis is not very strong (Lurie and Rosenthal 2010; Sawers and Stillwaggon 2010), and that stems in part from the methodological challenges in measuring partnership concurrency and its effects (Morris 2010).
An empirical study of the relationship between polygyny and HIV infection thus offers a new perspective on a research question that has otherwise been quite elusive. We will treat polygyny as an individual-level risk factor, but also invoke polygyny as a cultural system with indirect implications for the spread of an STI. In that sense, our approach echoes that of social scientists who have studied the relationship between polygyny and fertility (e.g., Ezeh 1997; Pebley and Mbugua 1989; Pison 1986). The analogy with the fertility literature does not stop there. HIV and fertility share several proximate determinants (e.g., the age at sexual debut and coital frequency), and we will refer to those for developing our hypotheses.
We use data from 20 surveys with individually linked HIV serostatus information and use a multilevel model to reassess the neutral or negative ecological association between polygyny and HIV reported in Reniers and Watkins (2010). We confirm the findings from the earlier study but also identify a positive individual-level correlation for junior wives of polygynous men. Coincidentally, a similar discrepancy between the individual- and ecological-level relationships (albeit reversed) also exists for polygyny and fertility.1 The multilevel analysis is followed by an investigation into the mechanisms that could account for what we conveniently dub benign concurrency. These mechanisms relate to (1) the adverse selection of HIV-positive women into polygynous unions, (2) the sexual network structure characteristic of polygyny, (3) the relatively low coital frequency in conjugal dyads of polygynous unions (coital dilution), and (4) the restricted access to partners for younger men in populations where polygynous men presumably monopolize the women in their community (monopolizing polygynists). We explain the rationale for each of these hypothesized mechanisms in greater detail below.
The sustenance of polygynous marriage systems via the remarriage and disproportionate recruitment of divorcees and widows as junior wives into polygynous unions was described well before the advent of the widespread HIV epidemic (Goldman and Pebley 1989; Lesthaeghe et al. 1989; Pison 1986; Van de Walle 1968). Timæus and Reynar (1998) presented an analysis of more recent data. This is particularly relevant in populations with generalized HIV epidemics because these women often have elevated HIV prevalence levels and could bring HIV into the household (De Walque and Kline 2012; Reniers 2008; Reniers and Tfaily 2008). A comparable selection effect has been identified in the fertility literature: levels of female subfecundity or infecundity are often higher among women with polygynous husbands because presumed infecundity is a motivation for divorce, and divorcees are more likely to end up in polygynous unions. Alternatively, the presumed infecundity of his spouse may motivate the husband to add a wife to his household (Pebley and Mbugua 1989; Timæus and Reynar 1998).
Sexual Network Structure
The sexual network structure produced by polygyny is one of gender-asymmetric partnership concurrency: men can have multiple wives, whereas women are in principle monogamous. Unlike gender-symmetric concurrency, polygyny limits the size of temporally connected sexual network components to the number of wives of polygynous men. The component size distribution is important because large components are believed to be conducive for the spread of an STI (Bearman et al. 2004; Morris and Kretzschmar 1997).2 This is the idealized form of polygyny, however, and it is important to recognize that any protective effect of the sexual network structure will dilute as soon as women in polygynous unions have affairs on the side. In that scenario, women with outside liaisons act as bridges between otherwise disjoint network components of monogamous or polygynous men. From a networks point of view, and assuming that partnership concurrency is common among men, a key determining factor of the epidemic potential in populations with polygyny is thus the level of female rather than male extramarital partnerships.
Some observers have pointed out that extramarital liaisons of (junior) wives of polygynous men are tolerated or even expected: tolerated as long as the couple acts with the necessary discretion, and expected if the patriarch cannot satisfy the desires of all his wives (Delius and Glaser 2004).3 Caldwell et al. (1991) made similar observations for the Yoruba in Nigeria. For men, we may expect a higher prevalence of extramarital partnerships in populations with polygyny because the institution of polygyny itself supposedly endorses the notion that men require more than one woman for sexual satisfaction (Caldwell et al. 1993).4 Nonmarital partnerships can also be part of their quest for an additional spouse. The empirical evidence for an elevated prevalence of outside partnerships in polygynous unions is neither overwhelming nor consistent, but points in that direction (Carael et al. 2001; Isiugo-Abanihe 1994; Mitsunaga et al. 2005; Nnko et al. 2004; Reniers and Tfaily 2008). Bearing in mind that self-reports of nonmarital sexual partners may not be trustworthy, we evaluate whether these are more commonly reported by men and women in polygynous unions compared with those in monogamous unions.
The hypothesis concerning the relatively low coital frequency in polygynous unions is also derived from the fertility literature. Unlike a monogamous husband, a polygynous man divides his time between two or more women, which tends to reduce the frequency of sexual intercourse with each of his wives. In addition, the age difference between spouses is often larger in polygynous than in monogamous unions (Barbieri and Hertrich 2005). This age difference will also have repercussions for the coital frequency because male potency tends to decline with age (Westoff 1974),5 but also because a large spousal age difference can be an obstruction to a strong emotional bond. The relatively low coital frequency in conjugal dyads of polygynous unions is claimed to have a fertility-inhibiting effect, but the evidence is mixed (Barrett 1971; Garenne and van de Walle 1989; Josephson 2002; Musham 1956; Pebley and Mbugua 1989).
In HIV research, the number of sex partners has often attracted more attention than the number of sex acts, but the latter has been hypothesized to have more immediate repercussions for the transmission of HIV (Blower and Boe 1993). Worth noting is that the transmission dynamics are not necessarily the same for STIs with high probabilities of transmission per sex act and a short duration of infectiousness (e.g., gonorrhea) and STIs with a relatively low probability of transmission and a long duration of infectiousness (e.g., HIV) (Garnett 2002). In the first case, the number of sex partners is more important than the frequency of intercourse with each of those partners, and rapid partnership turnover is likely to maximize the epidemic potential.6 In contrast, the number of sex acts per partner is more salient for an STI with low but long-duration infectiousness. As a first step in the evaluation of a coital dilution effect, we compare the reported frequency of sexual intercourse of women in monogamous and polygynous unions.
Another aspect of polygyny with possible consequences for the spread of HIV relates to young men’s access to sexual partners. Polygynous men usually have one or more wives who are considerably younger than themselves, and they may squeeze the younger men out of the market for sexual partners and thus reduce their exposure to HIV. Many have speculated about increased levels of male celibacy in populations with polygyny, ranging from Malthus (1989) in the early nineteenth century to more contemporary observers (Goody 1973). Interestingly, it is also a concurrency-compensating factor that is built into the simulation work of Morris and Kretzschmar (2000): in order to compare partnership concurrency with serial monogamy, the number of partnerships in the simulated population is held constant, and an increase in concurrency thus produces more isolated nodes in the sexual network. When gender-asymmetric concurrency is modeled, the men in particular end up as isolates.
These depictions of monopolizing polygynists are challenged by a simple demographic analysis that demonstrates that a system of plurality of wives is facilitated by a relatively low age at first marriage for women and considerable age differences between spouses. In populations with a relatively young age structure, that will produce a surplus of marriageable women. The polygynous marriage system is further sustained by the relatively rapid remarriage of widows and divorcees and the low rates of permanent celibacy (Dorjahn 1959; Goldman and Pebley 1989; Pison 1986; Van de Walle 1968). The focus of the demographic analysis has been marriage partners and not sex partners per se, but it is worth noting that the hypothesized mechanism may collapse if the age at sexual debut differs between men and women, and age-disparate partnerships are common. We thus examine how two manifestations of access to sexual partners—namely, the male age at sexual debut and the reported coital frequency of young men—covary with the prevalence of polygyny.
Data and Methods
We use data from 20 African Demographic and Health Surveys (DHS) and HIV/AIDS Indicator Surveys (AIS) with individually linked survey and HIV serostatus information for men and women in their reproductive ages. In alphabetical order, these are Burkina Faso (2003), Cameroon (2004), Democratic Republic of Congo (2007), Ethiopia (2005), Ghana (2003), Guinea (2005), Ivory Coast (2005), Kenya (2003), Lesotho (2004), Liberia (2007), Malawi (2004), Mali (2006), Niger (2006), Rwanda (2005), Senegal (2005), Sierra Leone (2008), Swaziland (2006–2007), Tanzania (2003), Zambia (2007), and Zimbabwe (2005–2006). Data, survey instruments, and other documentation can be retrieved from the Measure DHS website (http://www.measuredhs.com). In Reniers and Watkins (2010), we provide more detail about the sample, including country-wise descriptive statistics for many of the variables that we use in this study.
Two disadvantages of the DHS and AIS for this analysis are the lack of detail on marriage and partnerships and insufficient standardization of the questionnaires across surveys. Marriage durations for higher-order marriages and the outcome of previous marriages are, for example, reported only in a few surveys. The wife’s rank in a polygynous union is missing in a few surveys as well. Our analyses and conclusions are also restricted by the cross-sectional nature of the DHS. Consequently, we mostly use contemporaneous measures of polygyny and other covariates, whereas our primary outcome of interest—HIV prevalence—is the result of cumulative exposure over the 10 years prior to the survey. Important temporal changes in the prevalence of polygyny driven by HIV prevalence itself cannot be excluded, and strong causal claims are thus difficult to justify. 7 Another limitation is that the data do not permit us to identify men and women who have ever been in a polygynous union but were not so at the time of the interview.
The empirical analyses start with a multilevel analysis wherein we evaluate the association between polygyny and HIV at the individual and ecological or survey-cluster level. We present the results for a pooled sample of all countries and for three groups of countries stratified by HIV prevalence: low (<2%), medium (2%–10%), and high (>10%). The discussion of the multilevel model is followed by an inquiry into the mechanisms that could reconcile the contrasting findings about the relationship between polygyny and HIV at the individual and ecological level. We test the selection and coital dilution hypotheses, evaluate whether the asymmetric sexual network structure in populations that practice polygyny is breached by (women’s) infidelity, and compare the access to sexual partners among younger men in populations with varying levels of polygyny. We present individual-level analyses of these mechanisms in terms of survey-cluster fixed-effects models with a minimal set of other covariates. In a couple of instances in which we are interested in the coefficients for community-level attributes, we use multilevel logit and ordinary least squares regression models. The analyses of the hypothesized mechanisms are carried out for all countries separately. Survey weights (and, if appropriate, HIV seroprevalence survey weights) are used for computing summary statistics of the prevalence of polygyny, HIV, and their covariates. No weighting is used in the regression analyses.
We define the prevalence of polygyny as the average of the percentage of married men and the percentage of married women in polygynous unions. We average male- and female-centered definitions because they account for two characteristics of polygyny: namely, the incidence (the proportion of men with more than one wife) and the intensity of polygyny (the average number of wives per polygynist) (Van de Walle 1968).8 At the individual level, we work with a self-reported marital status variable that distinguishes between being single or never married, in a monogamous union (reference category), in a polygynous union (including the wife’s rank when appropriate), 9 and formerly married. We combine formal and cohabiting consensual unions. Other variables and indices are defined when they are introduced.
The Conundrum: A Discrepant Individual- and Ecological-Level Association Between Polygyny and HIV
In Fig. 2, we present the results from a hierarchical logit model with three levels (individual, survey cluster, and country) and with individual-level HIV status as the outcome. The set of coefficients at the top are odds ratios for a pooled analysis of all countries followed by estimates for groups of countries, stratified by HIV prevalence. The analyses were done separately for men and women, and we included age and an indicator for the type of the survey cluster (urban/rural) as statistical controls.
From the estimates in the panel on the left, we learn that living in a survey cluster with more polygyny is negatively correlated with HIV status for both men and women. The effect sizes are much larger for the pooled analysis than in the regressions stratified by HIV prevalence, and that is due to important between-country variation in the prevalence of polygyny and HIV (Reniers and Watkins 2010). Adding country fixed effects attenuates the coefficients, but they remain significant (see also Table 1).
Turning to the individual-level association between union type and HIV status, in the panel on the right, we identify three suggestive results. First, there is no relationship between polygyny status and HIV status for men. The odds ratio is, however, larger and approaches statistical significance in settings with high HIV prevalence. Second, first wives seem less likely to be HIV positive than wives of monogamous men. Finally, higher-order or junior wives of polygynous men are more often HIV positive than spouses of monogamous men. It is also worth noting that the correlation between polygyny status and HIV status for all women combined is either neutral or positive (not shown). Disregarding the wife’s rank, in other words, can lead to important errors of interpretation.
In sum, the association between polygyny and HIV is rather complex. The negative ecological correlation confirms the benign relationship between polygyny and HIV that we postulated earlier, and is in conflict with predictions of the concurrency hypothesis. This is probably the most captivating finding in Fig. 2 because it suggests that polygyny or other features of populations that practice polygyny inhibit the spread of HIV. In the remainder of the article, we explore some of the mechanisms that could account for that.
At the individual level, it is not entirely clear how concurrency could influence these associations either. We argue in the next section that the elevated HIV prevalence among junior wives is produced by the selection of HIV-positive women into polygynous unions. A pattern that is perhaps more suggestive of a concurrency effect are the increasing odds ratios for first wives of polygynous husbands as we move from low- to high-HIV-prevalence strata because they may be exposed, via their husbands, to the HIV virus present in junior wives. This is a mere speculation, however, because we are not well equipped to tease out individual-level concurrency effects with cross-sectional and ego-centered data.
Accounting for the Elevated HIV Prevalence Among Junior Wives of Polygynous Men: Adverse Selection
Our leading hypothesis for the elevated HIV prevalence in junior wives of polygynous men is that it is driven by the selection of HIV-positive women into polygynous unions. To directly evaluate such a selection effect, one should know the HIV status at the time of marriage. Because of the cross-sectional nature of the surveys, however, we can assess HIV status only at the time of the interview, and that may be influenced by the transmission of HIV within marriage. To work around this problem, we compare the odds that women in their first marriage and women at higher marriage orders (i.e., divorcees and widows) are married to a polygynous rather than a monogamous husband. We do so because both marriage order and the outcome of the previous marriage are correlated with HIV status, and they do not vary by marriage duration. The results of that analysis are presented in Fig. 3.
In the left panel of Fig. 3, we present the percentage of women who are in second or higher-order marriages (gray bars) and, among those, the prevalence of widows (white bars). Information about the outcome of the previous marriage is available for less than half the countries, which explains the high number of missing estimates. The countries themselves are arranged from low prevalence at the top to high prevalence at the bottom of the graph.
In the panel on the right, we present the odds of being in a polygynous versus monogamous union for women who are in their second or higher-order marriage compared with women in their first marriage. Without exception, women at higher marriage orders are more likely to be in polygynous unions. In a sample restricted to women who have been married more than once, widows appear more likely than divorcees to remarry a polygynous husband, but the effect is not consistent: its parameter estimate is positive for four countries, negative for another, and insignificant for the three remaining countries. These results nonetheless confirm our hypothesis that widows and divorcees are disproportionately recruited into polygynous unions, thus contributing to the relatively high rates of HIV prevalence in junior wives. Not shown in Fig. 3, but well described in Timæus and Reynar (1998), is that divorcees and widows often end up as second or third wives of polygynous men, a dynamic that is in part sustained via the inheritance of widows by one of their late husbands’ relatives.
The adverse selection hypothesis is supported by a supplementary analysis wherein we compare the odds of being HIV positive for first wives and junior wives of polygynous men with those of spouses of monogamous men (pooled analysis for all countries with information on wife’s rank). We condition on being married to an HIV-negative husband to avoid that the HIV status observed at the time of the survey is acquired from the current spouse.10 It does not, however, correct for the possible acquisition of HIV during marriage via other channels (e.g., an extramarital partner or unsafe injection). The results suggest that the odds of being HIV positive are 1.46 times higher (95% confidence interval (CI): 1.11–1.93) for junior wives of HIV-negative polygynous men than for spouses of HIV-negative monogamous men. This coefficient is adjusted for age, age squared, urban/rural place of residence, and country. The estimated odds ratio based on a survey-cluster fixed-effects model with a statistical control for age is 1.79 (95% CI: 1.24–2.57).11 None of the coefficients for first wives are statistically significant, suggesting that there are no important differences in the HIV status of first wives of polygynous men and spouses of monogamous husbands. These results thus deviate somewhat from the analysis presented in Fig. 2 (and Table 1), but they are based on a much more restrictive sample.
Accounting for the Lower HIV Prevalence in Populations with Polygyny
Sexual Network Structure
The adverse selection of HIV-positive women into polygynous unions may thus drive the elevated HIV prevalence in junior wives of polygynous men, but this selection effect alone is unlikely to account for the negative or neutral ecological association. To the contrary, selection alone might promote the epidemic spread because it creates a system of sexual mixing whereby HIV-positive women are disproportionately recruited into concurrent partnerships.12 Other, compensating mechanisms thus need to be explored, and one potential factor of importance is the sexual network structure characteristic of polygyny.
Polygynous marriage systems entail gender-asymmetric partnership concurrency, which is perhaps not as conducive for the spread of HIV as a system of gender-symmetric concurrency. This holds only to the extent that women in polygynous unions do not have extramarital partners, and that condition is evaluated in Fig. 4. In the left panel, we present the percentage of men and women reporting at least one nonmarital sexual partnership in the 12 months before the survey.13 On the right side, we compare the odds of reporting nonmarital partners by union type. In the majority of countries, men and women in polygynous unions report more nonmarital affairs than men and women in monogamous unions (these results are most consistent for women), and that produces leakage in an otherwise reasonably protective sexual network structure. This conclusion comes with two qualifiers. First, we need to keep in mind that the fraction reporting nonmarital sex is fairly low, and particularly so for women and in countries where the relationship between polygyny status and nonmarital sex is strongest: in 16 out of 20 countries, less than 5% of women reported having a nonmarital partner in the preceding 12 months. Liberia and Cameroon are the only two countries where the reported level of nonmarital partners exceeds 5% and where the association with the prevalence of polygyny is significant. Second, accurate reports of sexual behavior are notoriously elusive, and there is no reason to believe that these data are any different (Cleland et al. 2004; Curtis and Sutherland 2004; Langhaug et al. 2010; Phillips et al. 2010).
In sum, these results are not conclusive. If anything, they suggest that men and women in polygynous unions have more nonmarital sex partners, which would expose them individually as well as breach the presumed protective asymmetric sexual network structure.
We hypothesized that a polygynous husband divides his time and coital budget between his respective spouses, leading to a reduction in the number of sex acts with each of his wives. Large age differences between spouses, and the age of the husband itself, could further depress the frequency of intercourse. To seek confirmation, we evaluate whether self-reported sexual activity (both marital and nonmarital) in the week and month prior to the survey correlates with polygyny status. We exclude women from the analysis who are postpartum abstaining. If anything, this restriction will lead to upward bias in the coefficient estimates because the practice of polygyny is usually associated with longer durations of postpartum abstinence (Lesthaeghe et al. 1994).
In the left panel of Fig. 5, we present the prevalence of sexual activity in the week and month before the survey as reported by married women. In the right panel, we evaluate whether polygyny status is a predictor of recent sexual activity. Compared with women in monogamous unions, women in polygynous unions report lower coital frequency in all but one of the countries (Niger). Not shown in Fig. 5 is that coital frequency is usually slightly higher for junior than for senior wives. The fertility literature has explained this as a sexual preference of the husband for the newest addition to his household (Lardoux and van de Walle 2003). In populations where fertility confers status, junior wives with fewer children will also have a greater incentive to compete for their husband’s attention. The estimates presented in Fig. 5 include a control for the respondent’s age (here, women). Inclusion of an additional control for husband’s age hardly changes the coefficients. The age effects themselves are negative, albeit not always linear.
An intriguing regularity in Fig. 5 is that strength of the association increases as we move from countries with low HIV prevalence (top) to those with high prevalence (bottom). This suggests that the reduction in coital frequency is not just an issue of resource constraints on the part of the husband, but that it is also related to behavioral adjustments of conjugal partners in settings and unions wherein the risk of HIV transmission is higher (e.g., unions with inherited AIDS widows).14 This interpretation is supported by the more general finding that coital frequency recently declined in eastern and southern African populations but not in West African countries, where the HIV/AIDS epidemic never reached the same proportions (Westoff 2007). A competing explanation is that coital frequency is lower in countries with higher HIV prevalence because of AIDS-related morbidity.
Irrespective of the underlying reasons, the finding that the number of sex acts is lower in conjugal dyads of polygynous unions is important. In conjunction with the adverse selection discussed earlier, this implies that HIV-positive women are disproportionately recruited into relationships in which the sexual frequency—and therefore the likelihood of transmission—is lower. In other words, these two mechanisms (selection and coital dilution) contribute to the positive individual-level relationship (junior wives) and inhibit the spread of HIV at the population level.
The last proposed mechanism for the relatively low HIV prevalence rates in populations with elevated levels of polygyny is that young men in these populations have restricted access to sexual partners. We first evaluate that hypothesis via the age at first sex and assume that restricted access to sexual partners translates into a higher age at sexual debut. The unit of analysis is a survey cluster, and the univariate distributions of the median age at first sex by country are represented by the gray boxplots in the left panel of Fig. 6.15 On the right side, we show the coefficients from a regression of the median on the prevalence of polygyny (gray markers). These demonstrate that the median age at first sex is not significantly higher in clusters where polygyny is more common.
In a second analysis, we compare the frequency of intercourse reported by men aged 30 and younger in populations with varying degrees of polygyny. This is done by means of a multilevel logit model whereby the prevalence of polygyny is measured at the cluster level and sex in the last week and the last month are the individual-level outcomes of interest. The results, summarized in Fig. 7, do not suggest a consistent relationship between the prevalence of polygyny and the reported coital frequency of young men. In sum, neither of these analyses produces any evidence for the hypothesis that young men in populations with polygyny are deprived of sexual partners.
Instead, we find that the nature of the partnerships of young men in populations with polygyny is quite distinctive. We show this in Fig. 6 using the white boxplots and white markers for the coefficient estimates. These represent the distribution of the interval between the age at first sex and at first marriage (left panel) and its relationship with polygyny (right panel). The coefficients demonstrate that the gap between first sex and first marriage is smaller (statistically significant in just over half the countries) in populations or clusters with higher levels of polygyny. In other words, polygyny does not restrict the sexual activity level of young men but channels it into formal relationship types at an earlier age. This is important because it has been argued that a protracted period of premarital sexual activity in southern African populations contributes to the relatively high rates of HIV prevalence in that region (Bongaarts 2007).
Individual and Ecological Associations Revisited
In Table 1, we present the individual- and ecological-level associations between polygyny and HIV again, but this time with the inclusion of an extensive set of statistical controls. One set of controls represents attributes or behaviors that are known to be related to HIV infection. They comprise age, the type of place of residence (Carael 1997), education (Fortson 2008; Hargreaves et al. 2008), male circumcision (Auvert et al. 2005; Gray et al. 2007; Moses et al. 1990), the prevalence of other STIs (Fleming and Wasserheit 1999; Over and Piot 1991), religion (Gray 2004), and mobility or migration (Decosas et al. 1995). A second set of controls was chosen to represent, to the extent possible, the mechanisms that we discussed previously. These include the prevalence of nonmarital sex, the median age at first marriage, the interval between the age at sexual debut and the age at first marriage, and a measure of the selection of widows and divorcees into polygynous unions (dummy variable that identifies survey clusters where more than one-third of the remarried divorcees and widows are in a polygynous union). The type of place of residence is an indicator variable that contrasts urban with rural clusters. Our measures of the age at first marriage and sexual debut have been defined earlier, and education is measured in terms of the years of schooling. The other indices are operationalized as percentages. The measure of religion distinguishes between Islam and other religions, and the prevalence of STIs is based on self-reports of genital sores, ulcers, or discharge (categorized as 0%, 0%–10%, and >10%). Spatial mobility is measured as the percentage of interviewed men (the information is not available for women) who spent more than one month away from their homes. The measure of nonmarital sex is based on self-reports of married respondents. The prevalence of STIs, male mobility, and nonmarital sex all pertain to the year before the survey.
Models 1 and 2 are three-level models (individual, survey cluster, and country) with a minimal and extended set of control variables, respectively. Model 3 is a two-level analysis with country fixed effects. At the individual level, the marital status indicators have the coefficient estimates that we expected: formerly married men and women are most likely to be HIV positive, and never-married men are somewhat less likely to be HIV positive than currently married men in monogamous unions. For women, being single is positively correlated with HIV status, but the coefficient loses statistical significance after other control variables are introduced. There are no noticeable differences in the HIV prevalence of men in polygynous and monogamous unions. This is not the case for women. As was clear from Fig. 2, first wives are somewhat less likely to be HIV positive than monogamously married women, and junior wives are significantly more likely to be infected. Adding statistical controls does not attenuate these coefficients. The parameter estimates for junior wives support the adverse selection hypothesis postulated earlier. Further inquiry is necessary to distill the reason(s) for the relatively low HIV prevalence in first wives of polygynous men. Figure 2 also suggests that that relationship is less pronounced in populations with high HIV prevalence.
The survey-cluster measure of polygyny has a negative coefficient for both men and women. As we move from Model 1 to Model 3, the parameter estimates become smaller but remain significant. To the extent that cross-sectional data allow us to draw such conclusions, these results confirm what we expected all along: polygyny inhibits or slows down the spread of HIV. None of the other statistical controls reveal any surprising associations. The coefficients for age are indicative of a curvilinear pattern, and educational attainment is positively correlated with HIV status. HIV prevalence rates are higher in urban areas, male circumcision is protective, and the coefficients for STIs are generally positive. The measures of mobility and religion have been omitted from the final models because of a lack of statistical significance. Admittedly, our measure of mobility is not very refined, and a rigorous evaluation of the effect of religion should include a more fine-grained distinction between religions and denominations, as well as a measure of religious involvement (Trinitapoli and Regnerus 2006).
The other covariates in Table 1 pertain to some of the mechanisms that we explored for explaining the benign effect of polygyny. Late marriage is protective (particularly late marriage of women; see also Clark 2004), but its effect should be interpreted in conjunction with the coefficient for the length of the sexually active period before marriage. Together, they suggest that late marriage will contain the spread of HIV only if it is not accompanied by premarital sexual activity. In line with Bongaarts’ (2007) hypothesis, long duration of premarital sexual activity itself promotes the spread of HIV. As could be expected, HIV prevalence is also higher in settings with elevated levels of self-reported nonmarital sex among married men and women. We also tested the implications of the selection of divorcees and widows into polygynous unions. In survey clusters where this selection is relatively strong, HIV prevalence tends to be lower. For men, the coefficient is not statistically significant.
Junior wives of polygynous men are more likely to be HIV positive than the spouses of monogamous men, but HIV prevalence is lower in populations where the practice of polygyny is more common. This contrast in the relationship between polygyny and HIV at different levels of aggregation set the stage for an exploration of the mechanisms relating polygyny to HIV. We find that the disproportionate recruitment of divorcees and widows (HIV-positive women) into polygynous unions contributes to the positive individual-level association. These women often end up as second or third wives of polygynous men and may be acquired through practices such as levirate marriage and widow inheritance. Adverse selection alone is unlikely to account for the negative or neutral ecological correlation, however, and we therefore considered a number of other factors with complementary effects.
First, polygyny produces a sexual network with gender-asymmetric partnership concurrency, and that will restrict the size of the temporally connected network components compared with a network in which both men and women have concurrent partnerships (i.e., a sexual network with gender-symmetric concurrency). This gender asymmetry in the sexual network structure is believed to contain the epidemic potential, but such a conclusion comes with a couple of caveats. Simulations suggest, for example, that the protective effect of asymmetric concurrency is modest at best (Santhakumaran et al. 2010). In addition, we find that both men and women in polygynous unions have more nonmarital sex partners than those who are in monogamous marriages, and that erodes the protective character of the polygynous sexual network structure. Even though the prevalence of (reported) nonmarital partners is fairly low among women and in countries where the relationship with polygyny seems strongest, the evidence for the protective effect of the sexual network structure in populations with polygyny is not conclusive, and more theoretical and empirical work is needed to assess the full range of its implications for the spread of HIV.
The second and probably more promising explanation for the benign relationship between polygyny and HIV is the lower coital frequency in conjugal dyads of polygynous compared with monogamous unions. In conjunction with the adverse selection effect described earlier, polygyny thus produces a system of sexual mixing whereby those with the greatest risk of being HIV positive are absorbed into a regime of lower sexual intensity. Although this is still far from imposing a quarantine on HIV-positive individuals, it will contain the spread of HIV at the population level.
Third, polygyny is associated with a shorter sexually active period before marriage. Contrary to our suspicion that polygynous men monopolize the women in their community, the ages at sexual debut and the coital frequency of young men are comparable in both populations that practice polygyny and those that do not; what distinguishes populations with polygyny is that their sexual activity is channeled into marriage at a younger age. As a consequence, the sexually active interval prior to marriage is much shorter. Late marriage and a protracted episode of sexual activity before marriage—possibly involving partnership concurrency and high partnership turnover rates—have previously been proposed as factors contributing to the pervasiveness of HIV in southern African countries.
The list of contributing factors considered here is not exhaustive, and it is likely that polygyny is correlated with other network attributes or socially regulated behavior that matters for the spread of an STI. These could include the relatively large age difference between spouses and a prolonged period of postpartum sexual abstinence. Polygynous societies are often also characterized by high rates of marital dissolution and the easy remarriage of widows and divorcees (Pison 1986; Van de Walle 1990). The effects of these correlates are difficult to anticipate. The large age differences between spouses imply that women have sex with men of an age range in which HIV prevalence is usually higher (Gregson et al. 2002; Kelly et al. 2003), but cross-generational sex will also have repercussions for the coital frequency. Similarly, postpartum abstinence may lead to a reduction in the frequency of intercourse (it is by definition the case for women), but it has also been described as a risk factor because it induces men to seek partners elsewhere (Cleland et al. 1999; Glynn et al. 2001).
The easy transition in and out of marriage could have opposing consequences as well: it increases the total number of (marriage) partners over an individual’s lifetime, but it could as well contain the need or desire for informal partnerships between marriages. A factor that is more likely to be uniformly protective pertains to the interval between successive partnerships, or more precisely, the interval between the (potential) infection by the previous partner and first sex with the new partner. Partnership gaps of several weeks are desirable because they preclude that seropositive individuals have intercourse with a new partner during the highly infectious period shortly after seroconversion (Kraut-Becher and Aral 2003). Such gaps are probably longer in the case of two successive marriages as opposed to informal partnerships. This is particularly the case where social customs prescribe widows to observe a mourning period before remarrying.
Our data and methods do not allow us to go much beyond the identification of plausible mechanisms for the benign relationship between polygyny and HIV. Even though the precise contribution of each of these mechanisms is still outstanding, our study challenges the discourse that polygyny and widow inheritance are harmful cultural practices.16 One’s opinion might differ if precedence is given to individual over public health, but it is clear that our current understanding of the effects of partnership concurrency for the spread of STIs is not sufficiently specific to inform public health policies on these issues. Even more worrying is that public health advocacy targeting generic partnership concurrency could have counterproductive effects in populations where polygyny is common.
It is equally important to stress, however, that our results do not necessarily invalidate the existence of concurrency effects per se. First, we neither deny nor offer any proof against the existence of causal concurrency effects. In fact, some of the individual-level associations that we identified in Fig. 2 (e.g., the shrinking differences in HIV status between women in monogamous unions and first wives of polygynous men as we move from settings with low HIV prevalence to those with high prevalence) could be produced by partnership concurrency, but cross-sectional and ego-centered data such as the DHS are not the best resource to evaluate these and related hypotheses. Second, concurrency could be driving these results if communities with low levels of polygyny are characterized by higher levels of informal concurrency, and vice versa. Even if that were true, more work is needed to qualify and quantify different types of concurrency and their implications for the spread of HIV (Boily 2010; Gorbach et al. 2002; UNAIDS 2009). As is readily acknowledged by the authors themselves, the early modeling work of Morris and Kretzschmar (1997) is only “loosely based on empirically observed patterns.” The particular implications of dynamic and static, formal and informal, or short- and long-term concurrency are not well understood. Some advocates of the concurrency hypothesis have suggested that long-term overlapping partnerships are a distinctive feature of the sexual networks in African populations and are responsible for the exceptional magnitude of the epidemic in some of these countries (Epstein 2008). Yet, polygyny is the quintessential case of long-term concurrency, and our study suggests that it impedes rather than accelerates epidemic spread.
An important lesson from these analyses is that polygyny—including partnership concurrency more generally—is not an uncorrelated or exogenous characteristic of a sexual network (Aral 2010; Boily 2010; Kretzschmar et al. 2010). We have pointed out that polygyny is coupled with a very specific partnership mixing pattern. Polygyny is also associated with a reduction in the coital frequency within conjugal dyads and shorter spells of premarital sexual activity. All of these are sexual network attributes—in the case of coital dilution, perhaps a conscious behavioral adjustment—that are pertinent for the spread of STIs. In contrast, much of the simulation work to date has tried to isolate the causal effect of a single characteristic of the sexual network structure, namely partnership concurrency.17 Even if the abstraction of models presents a plausible and persuasive case that partnership concurrency amplifies the diffusion of HIV, other changes in sexual networks and behavior that are intimately related to concurrency may nullify or reverse its causal effect. In other words, a reductionist fallacy is lurking, and policy makers are advised to proceed with caution until we have gained a better understanding of partnership concurrency, its various manifestations, and all of its consequences.
We thank Jimi Adams, Ron Brookmeyer, Jeffrey Eaton, Stephane Helleringer, Mirjam Kretzschmar, Ron Lesthaeghe, Bruno Masquelier, Matthew Salganik, Ann Swidler, Susan Watkins, and the journal’s reviewers for stimulating discussions and comments on the manuscript. We thank Measure DHS for giving us access to the data sets. This study was made possible with support from an NICHD Center Grant to the Office of Population Research (R24HD047879) and an NICHD grant to the University of Colorado Population Center (R24 HD066613).
Women in polygynous unions have lower fertility than women in monogamous unions, but because polygyny maximizes the time that women spend in a union, population-level fertility rates tend to be positively correlated with the prevalence of polygyny (Pison 1986).
Morris and Kretzschmar (1997) modeled the spread of HIV in a sexual network with gender-symmetric concurrency but acknowledged the gender asymmetry produced by polygyny. In a later paper (2000), they relaxed the assumption of gender symmetry (to match empirical data for Uganda) with the expected dampening effect on component sizes and epidemic potential. The component size distribution is, however, not the only attribute of a polygynous sexual network with repercussions for the diffusion of an STI. Polygynous networks are, for example, characterized by star-shaped network components wherein polygynous men serve as the central nodes, whereas gender-symmetric partnership concurrency tends to produce chainlike network components. See Reniers and Watkins (2010) for an illustration of a simulated sexual network with gender-symmetric and gender-asymmetric concurrency.
In these cases, it is often understood that the husband retains the paternity rights to the children that might ensue (Delius and Glaser 2004).
See Delius and Glaser (2004) for a critique.
There is some discussion as to whether the age effect persists after marital duration is taken into account (Brewis and Meyer 2005).
See Welch et al. (1998) for a simulation study of gonorrhea transmission.
A previous assessment revealed no strong correlation between national-level HIV prevalence and the annual rate of change in the prevalence of polygyny (Reniers and Watkins 2010). Ideally, one would need to repeat such an analysis at the subnational level.
Analyses with measures of the incidence and intensity themselves point in the same direction. These are presented in Reniers and Tfaily (2010).
See Timæus and Reynar (1998) for a discussion of the correspondence between husbands and wives in the reporting of polygyny status.
This restriction also requires us to work with a subsample of linked spouses (i.e., the DHS couples-files), and the linkage of spouses is only possible if they co-reside.
These estimates could be affected by a mortality selection bias because first wives are likely to have been married longer than junior wives, and those who were HIV positive at the time of marriage could have died already. In the DHS, marriage durations are usually known for first marriages only, and this is difficult to account for. To reduce the potential for mortality selection bias, we repeated the analysis but restricted the sample to women under age 30 because AIDS mortality is not as common in young women. The results point in the same direction, but the coefficients are estimated with greater uncertainty: OR = 1.48 (95% CI: 1.01–2.15) for the logit model, and OR = 1.97 (95% CI: 1.11–3.49) for the survey-cluster fixed-effects model. Restricting the analysis to even younger women does not change the odds ratio estimates much, but they gradually lose statistical significance.
The magnitude of that selection effect may further depend on the distribution of the number of spouses per polygynous husband (i.e., the intensity of polygyny).
For men and women who married in the year prior to the survey, a nonmarital sexual partnership is not necessarily an extramarital or concurrent partnership because they may have had another partner prior to marrying the current spouse. Unfortunately, these cases cannot be excluded in the analyses because marriage duration is known for first marriages only.
In the subset of countries with information on the outcome of the previous marriage, widows report a lower frequency of intercourse, and that is particularly the case in Zambia and Zimbabwe, two of the countries with the highest HIV prevalence (not shown).
Our measures of the age at sexual debut and—in a later analysis— the age at first marriage are survey-cluster values of the median age. For a small number of clusters in which over 50% of respondents are self-reported virgins or single, the values have been imputed at the age of the oldest individual plus 1.
Widow inheritance, like polygyny, is often presented as an entrenched and harmful cultural practice that promotes the spread of HIV (Nyindo 2005; Okeyo and Allen 1994; UNAIDS 2006). Again, such claims come without much supporting evidence and, considering our results, appear premature.
Even this turns out to be difficult because changes in concurrency also induce other structural network changes (e.g., the number of isolated nodes or the mean number of partnerships non-isolates) (Boily 2010).