Two stylized facts about poverty in Africa motivate this article: female-headed households tend to be poorer, and poverty has been falling in the aggregate since the 1990s. These facts raise two questions. First, how have female-headed households fared? Second, what role have they played in Africa’s impressive recent aggregate growth and poverty reduction? Using data covering the entire region, we reexamine the current prevalence and characteristics of female-headed households and ask whether their prevalence has been rising, what factors have been associated with such changes since the mid-1990s, and whether poverty has fallen equiproportionately for male- and female-headed households. Lower female headship is associated with higher gross domestic product. However, other subtle transformations occurring across Africa—changes in marriage behavior, family formation, health, and education—are positively related to female headship, resulting in a growing share of female-headed households. This shift has been happening alongside declining aggregate poverty incidence. However, rather than being left behind, female-headed households have generally seen faster poverty reduction. As a whole, this group has contributed substantially to the reduction in poverty despite their smaller share in the population.
Among the geographic regions of the world, sub-Saharan Africa has had the highest incidence of poverty since around 1990. Furthermore, progress against poverty has long been slower than in other regions. This has changed in the new millennium. Sub-Saharan Africa has enjoyed a sustained period of economic growth as well as robust poverty reduction since the late 1990s. In addition to its high incidence of poverty, Africa has, in the past, been singled out for the significant share of households headed by women. The literature to date has generally suggested that female-headed households tend to be poorer.
These observations raise the question of what has been happening to the prevalence of female-headed households (FHHs) and their living standards during the recent period of sustained growth and declining poverty. Hypotheses can be advanced either way. On the one hand, high poverty can constrain the prospects of escaping poverty, such as when caught in a poverty trap. On the other hand, poor people may face relatively high economic returns to the new opportunities unleashed by growth. Gender-differentiated traditional networks may advance one group’s ability to take advantage of new opportunities, as has been argued for India (Munshi and Rosenzweig 2006). The poor are also likely to have benefited disproportionately from the expansion of social protection in the region (World Bank 2012). To complicate matters, the group of people living in FHHs may be fundamentally changing over time. In light of the ambiguity surrounding this issue, we seek to answer two broad questions. First, has female headship increased or decreased during the recent period of aggregate reduction in poverty? Second, is poverty also falling for FHHs, or are they being left behind?
Our study provides new evidence on FHHs for sub-Saharan Africa (Africa, hereafter) drawing on the microdata from all suitable national household surveys. All the available Demographic and Health Surveys (DHSs) for Africa are used to describe the current prevalence and characteristics of FHHs and to examine changes in their prevalence over the last 25 or so years covered by the data. We also explore the macro, population, and demographic factors that may correlate with the prevalence of FHHs, as well as changes in that prevalence. With these supplementary data, we investigate associations with rising education levels; changing circumstances and exogenous shocks, such as economic, health, and conflict crises; and the breakup of traditional systems of patriarchal marriage and family norms.
Using a set of high-quality household consumption surveys for 24 countries with comparable surveys for at least two dates, our study goes on to examine how FHHs have fared during the recent improvements in living standards, providing a breakdown of the total change in poverty into that contributed by male-headed households (MHHs) versus FHHs. To examine potential heterogeneity among FHHs, the analysis is extended to different types of FHHs.
We demonstrate that the share of households (and of the population living in households) headed by women is rising over time across the continent. Yet, with controls for other factors, higher gross domestic product (GDP) is associated with lower female headship. The apparent paradox that female headship has been rising during a period of economic growth is explained by the fact that other factors are changing across Africa. Changes in demographic and population characteristics, social norms, education, and the nature of the family all appear to be encouraging female headship. We also try to reconcile this finding with the recent aggregate reduction in poverty. Poverty is found to have declined for both MHHs and FHHs, but in most countries for which the data are adequate, poverty has been falling faster for FHHs as a whole. A decomposition of the change in poverty further indicates that rather than putting a break on poverty reduction, FHHs are contributing nearly as much, despite their smaller share in the population.
A number of studies have now convincingly established that Africa has seen falling aggregate poverty incidence (Beegle et al. 2016; Chen and Ravallion 2013). How do FHHs and the population living in FHHs figure into this story?
Much has been written and claimed about FHHs since the dawning realization going back to the 1970s that a rising share of households were headed by women in developed countries as well as the recognition—and worries—that similar trends were emerging in the developing world.1 Observing that FHHs in the West were most prevalent among poor and minority groups, concern arose about the transmission of poverty over time on the argument that poor women would be ill-equipped to care for their children on their own, and unless these women were aided by public policies, poverty would be perpetuated. Further claims are that FHHs are the poorest of the poor or that poverty has been feminized. Against these claims, the literature has also recognized the diversity among FHHs and rejected blunt generalizations (Chant 2008; Varley 1996).
Several studies have addressed whether and why the prevalence of female headship has altered globally over time, uncovering considerable variation across and within regions. Using 43 DHSs in the 1990s, Bongaarts (2001) found that the proportion of FHHs ranged from 13 % in the Near East and North Africa to 24 % in Latin America, with Asia (16 %) and Africa (22 %) in between. Ayad et al. (1997) exposed large variation within regions, especially in Africa, where the fraction of FHHs during 1990–1996 was less than 10 % in Burkina Faso, Mali, and Niger, and more than 30 % in Ghana, Kenya, Namibia, and Zimbabwe. To our knowledge, there have been no updates since 2000.
In industrialized countries, significant increases in female headship from the 1970s have been associated with demographic and socioeconomic changes, including expansion of women’s rights, such as legal rights to divorce, child custody, and housing; increased women’s labor force participation; increases in both unmarried fertility and its social acceptability; higher female longevity; and aging populations (Chant 1997; Moghadam 2005). Also contributing to upward trends in most regions are factors such as erosion of the extended family system, increases in age at marriage (see Jensen and Thornton 2003; Mensch et al. 2005), and reductions in spousal age gaps. In Africa, labor-related migration has been dominated by men seeking work in urban areas and leaving behind female heads (Buvinic and Rao Gupta 1997). Additional explanations for the high prevalence in Africa center on HIV infection and AIDS deaths; violent civil conflicts that have generated family dislocation and widows (Buvinic et al. 2013); a culture emphasizing lineage more than conjugal ties, and descent systems evolved from matrilineal kinship; and changes in women’s legal access to property, land, and labor markets.
The literature is infused by an assumption of increasing prevalence in FHHs across regions, yet relatively little empirical evidence exists on trends beyond the 1990s. Arias and Palloni (1999) even contested this view for Latin America over the 1970–1990 period, finding declining or stationary prevalence in most countries and arguing that women’s propensity to head a household has not risen as much as has the prevalence of surviving widows and divorcees.
Intimations of rising female headship in many parts of the developed and developing world have heightened fears of large future increases in poverty, given the well-documented economic disadvantages faced by women (Buvinic and Rao Gupta 1997; Chant 1997, 2008). A considerable number of studies, often focusing on a single country, have investigated whether households headed by men or by women are poorer. Common practice has been to make simple comparisons of mean household per capita consumption or poverty measures, or to regress log consumption per capita on controls and a dummy variable for the head’s gender.
Most observers have concluded that FHHs are poorer than MHHs (Buvinic and Rao Gupta 1997; Chant 1997, 2008), but the evidence is far from conclusive, and little consensus emerges in the literature. Buvinic and Rao Gupta (1997) reviewed 61 studies covering countries in Africa, Asia, Latin America, and the Caribbean, concluding that on balance, the evidence supports the hypothesis of higher poverty among FHHs. Lampietti and Stalker (2000) reviewed 58 World Bank Poverty Assessments carried out since 1994 and concluded that FHHs are, on average, poorer in 43 % of countries examined. Using data for Africa, Asia, and Central America, Quisumbing et al. (2001) found FHHs to be poorer in only 2 of 10 countries.
The evidence on the effects for children of living in a FHH in Africa also appears inconclusive. Some studies have found negative consequences for child well-being and mortality (Clark and Hamplová 2013). Against this, other evidence has suggested that female heads channel a higher fraction of their more limited resources for their children’s human capital. Many studies, in fact, have found that children—particularly girls—in FHHs have better schooling and nutritional outcomes (Buvinic and Rao Gupta 1997; Chant 1997; Kennedy and Haddad 1994; Lloyd and Blanc 1996).
A number of reasons account for the lack of agreement, including inconsistent definitions of headship, considerable diversity among FHHs and differences in how well-being is measured, and in particular, how the distinctive demographics of FHHs are taken into account.
No universally accepted definition of headship exists. Household-based surveys have collected information on heads as a natural way to anchor relationships among household members and thus provide a framework for describing family structure. Surveys typically rely on self-reported headship status. For example, the DHS interviewer’s manual states, “the person who is identified as the head of the household has to be someone who usually lives in the household. This person may be acknowledged as the head on the basis of age (older), sex (generally, but not necessarily, male), economic status (main provider), or some other reason. It is up to the respondent to define who heads the household” (ICF International 2012:26).
Dissatisfied with survey definitions and concerned about inconsistencies across countries, socioeconomic groups, and time, many analysts have imposed their own definitions. Equating headship with responsibility for the household’s economic well-being has been popular, and multiple versions have been attempted (Buvinic and Rao Gupta 1997; Lloyd and Gage-Brandon 1993; Rosenhouse 1989). The literature also identifies male-absent households and distinguishes between de jure FHHs (no live-in male partner or economic support from one) and de facto FHHs (with noncoresident husbands associated with labor migration or polygamous unions). The absent male partner in de facto households is assumed to maintain a large role in household decision-making and to contribute remittances. As the literature notes, these definitions impose strong assumptions for which surveys typically do not allow corroboration. Holding responsibility for a household’s economic well-being or contributing income does not necessarily lead to headship assignment or vice versa (see Buvinic and Rao Gupta 1997; Handa 1994; Lloyd and Gage-Brandon 1993; Posel 2001; Rosenhouse 1989). And one might ask why being the main breadwinner is more salient to headship than looking after household well-being in other ways. In this article, we prefer not to make such judgments and rely on self-reported headship.
Female heads are thus a diverse group, including widows; divorced, separated, and abandoned women; married women with a nonresident (polygynous or migrant) husband; single women; and mothers. Distinctions are largely based on the motives that led them to be heads (see Chant 1997, 2008; Handa 1994; Joshi 2005; Kennedy and Haddad 1994; and Klasen et al. 2015). Female headship can also be transitory (Buvinic and Rao Gupta 1997; Clark and Hamplová 2013; Joshi 2005).
Although all FHHs are unlikely to be worse off than MHHs, certain types of women are frequently found to head relatively disadvantaged households. In Africa, widow-headed households have been identified as significantly impoverished in Uganda (Appleton 1996), Zimbabwe (Horrell and Krishnan 2007), and Mali (van de Walle 2013). Divorce is common, and although young women typically remarry, there is considerable evidence of hardship for divorcees and their children (Clark and Brauner-Otto 2015; Clark and Hamplová 2013). FHHs who receive transfers from a male member are consistently found to be as well-off (in terms of consumption or income) as MHHs—and substantially better off than other FHHs (Buvinic and Rao Gupta 1997; Horrell and Krishnan 2007; Lampietti and Stalker 2000). Those who do not can be among the poorest (Kennedy and Haddad 1994).
The lack of consensus concerning the relative well-being of FHHs is also partly due to the use of noncomparable or inconsistent measures of living standards and/or benchmarks for judging deprivation. The sensitivity of results to methods is reasonably well recognized in the more academic literature (Haddad et al. 1996; Louat et al. 1993; Quisumbing et al. 2001). With respect to comparisons between MHHs and FHHs, a major factor concerns how adjustments are made for household size and composition. Poverty comparisons are sensitive to this choice. Failure to consider that FHHs are typically smaller overstates poverty among them (Quisumbing et al. 2001). There may also be economies of scale in consumption. Not accounting for the latter—a more common mistake—typically results in an understatement of poverty in FHHs. Allowing for scale economies can reverse conclusions about whether FHHs are richer or poorer (Drèze and Srinivasan 1997; van de Walle 2013).
FHHs tend to have higher dependency ratios and shares of children (see the section, Frequency and Characteristics of Female-Headed Households in Africa). Differences in demographic composition and in the consumption needs of adults and children can be accounted for by using adult equivalent scale-adjusted poverty measures. However, this approach implies knowing the consumption needs of different household members—usually based on actual consumption data in household surveys—which may differ significantly across areas and countries, and thus may not accurately reflect biological needs (Quisumbing et al. 2001).
We investigate changes in the prevalence of FHHs in Africa and whether FHHs have enjoyed a similar pace of poverty reduction as MHHs. We are not aware of any past work on the latter question taking an Africa-wide perspective.2 We test the sensitivity of our main results to distinguishing between different types of FHHs and to allowing for scale economies in consumption. Aforementioned issues may be somewhat less worrying when the focus is on changes instead of levels. Furthermore, in asking these questions, we do not assume that female headship is exogenous. Our objective is not to establish the causal effect of headship but rather to take stock of the correlations found in the data in a methodologically consistent way that is sensitive to the measurement issues across countries.
We use two types of surveys: the full series of DHSs for Africa over the last 25 years, and the World Bank’s PovcalNet database of harmonized household surveys for Africa. The DHSs, which cover approximately 89 % of Africa’s population, are used to briefly describe the prevalence and characteristics of FHHs relative to MHHs and to investigate Africa-wide changes in prevalence. The latter can be more effectively examined with DHSs, which contain demographic characteristics that prove key to tracking changes over time.
The DHSs also have the advantage of administering the same questionnaires (altered to fit local particularities) across countries. When using them, we define the household exclusive of nonresident visitors, and we define adults as aged 15 and older. The dependency ratio is given by the number of household members younger than 15 and older than 65 to that of members aged 15 to 64. We refer both to the share of FHHs and to the share of the population living in FHHs. When examining changes in these shares in the upcoming section, Changes in the Prevalence of FHHs Over Time, we must draw some explanatory variables from non-DHS sources. Given possible concerns with correlated measurement error, we also use alternative sources for the urban population share and female labor force participation rate.3 The variables and sources are described in Online Resource 1.
To complement the DHS-based analysis and examine changes in consumption poverty over the last 20 years or so, we use the consumption survey database compiled expressly for the purpose of making sound comparisons across African countries. In particular, we explore changes in the headcount index of poverty calculated based on household per capita consumption expenditures, for MHHs and FHHs separately, using common measures and methods across countries. The consumption data are converted to country consumer price index–adjusted 2005 purchasing power parity (PPP) dollars, and we use the international poverty line of $1.25 at PPP (Ravallion et al. 2009).4 We also test the sensitivity of the results to dividing household consumption by the square root of household size to allow for the generally smaller size of FHHs and economies of scale in consumption.5
We have a total of 24 countries—accounting for approximately 80 % of Africa’s current population—with at least two surveys that have been deemed fully comparable and can be used to measure changes in poverty.6 In three cases, there are two spells of comparable surveys for the same country. We thus examine a total of 27 spells. The resulting poverty measures are then used to examine changes in poverty and implement a decomposition of the contribution of FHHs and MHHs to the overall changes in each country. For the analysis using the consumption surveys, we define adults as aged 18 and older.
Rather than impose our own definition of when a household is female-headed, we use headship self-reports. As noted earlier, this approach makes the assumption that headship is defined similarly across countries, socioeconomic strata, and time. Structure is imposed by separating FHHs into those with a resident adult male member and those without one, given that this may be associated with variations in well-being. The expectation is that marital status of the head is another important factor to take into account (Clark and Hamplová 2013; Lambert et al. 2017). Parts of the analysis are therefore also conducted for the aforementioned subgroups and by whether the head is married. Another limitation of the data is that they are cross sectional and do not allow for the possibly transitory nature of female headship.
Frequency and Characteristics of Female-Headed Households in Africa
We begin with an overview of the frequency and attributes of FHHs and emphasize how sharply some attributes differ from those of MHHs. Table 1 presents statistics on the mean prevalence of FHHs by country (total, urban, and rural) as well as regional aggregates, using the latest available DHSs for 35 countries.7 Africa-wide, 26 % of all households are headed by women, comprising 21 % of Africa’s population. A pronounced variance across countries and regions is apparent. West Africa exhibits the lowest prevalence with one of five households headed by a woman and accounting for 15 % of the population. This variance partly reflects the continuing practice of polygamy, together with high widow and divorcée remarriage rates that continue to be widespread in majority Muslim countries (Lardoux and van de Walle 2003; Tabutin et al. 2004). Southern Africa has the highest rate, at 43 % of both FHHs and population living in FHHs. With the exception of Southern Africa, FHHs are more common in urban areas. It is a plausible conjecture that female headship is higher among matrilineal ethnic groups, but testing this would require considerable new data work.
Table 2 presents the mean characteristics of MHHs, FHHs, and FHHs with and without a male adult. The differences in means between the household groupings with a female relative to those with a male head are all statistically significant at the 1 % level or better. On average, FHHs have older heads (reflecting the many widowed heads) with fewer years of education (4.1 vs. 5.6). They tend to be smaller (3.9 vs. 5.1) and have higher dependency ratios (1.2 vs. 1.0). Mirroring the latter, female heads are 27 times more likely to live in households in which they are the only adult living with one or more children. In contrast, almost three-quarters of MHHs are composed of two or more adults and children, compared with only 44 % of FHHs. FHHs are also more likely to be single adult households (16 % vs. 10 %).
Another striking difference relates to the head’s marital status (Table 2). The vast majority of male heads are married (88 %), but this is true of only one-third of female heads. The others are primarily widows (40 %) and divorcées (17 %). The large gender disparity in marital status reflects in part large spousal age gaps, the practice of polygyny, and higher male remarriage rates following widowhood and divorce (de Walque and Kline 2012; Tabutin et al. 2004) as well as the ravages of conflict (Buvinic et al. 2013). To provide a different perspective on marital status, one can examine the share of all adults of a given status who are heads by gender. Overall, 62 % of adult men are heads compared with 18 % of women. With the exception of single women, married women have the lowest probability of being heads (at 10 %), and widows have the highest (71 %), followed by divorcées (56 %).
As noted earlier, heterogeneity exists among FHHs. The 26 % of FHHs can be disaggregated into 10 % with an adult male and 16 % without. In all regions, a preponderance of FHHs contains no adult male. Among the latter, 38 % of heads are widowed, 31 % are married, 19 % are divorced, and the rest are single (Table 2). Relative to FHHs with a male adult, those without tend to have younger, more-educated heads and fewer members, but higher dependency ratios at 1.5 versus 0.9. Male adult members are most often sons (60 %).
Using the DHS wealth index as a proxy for household economic status and controlling for the head’s age, we find that a larger share of FHHs than MHHs are classified in their country’s bottom wealth quintile, with the difference rising with age.8 This finding holds for all regions except West Africa.
A general pattern emerges in which FHHs are a quite heterogeneous group, some of whom are clearly disadvantaged in a number of ways.
Changes in the Prevalence of FHHs Over Time
How has the prevalence of FHHs and of the population living in FHHs evolved over the last 20 years, and what are the correlates of any revealed change? Fig. 1 depicts overall trends across individual countries by macro region.9 The share of a country’s FHHs (panel a) and the share of the population living in them (panel b) are plotted for each country’s earliest DHS on the horizontal axis against that for its latest survey. Points above the 45 degree line of equality indicate a rise.
The general picture is of an increase in the prevalence both of FHHs and of the population living in FHHs for most countries. Declines in both are evidenced for Ghana, Chad, Congo, and Lesotho, although the change is small for the last three.10 Disaggregating FHHs into those with and without an adult male also reveals a rising overall trend for both household types. The decline in Lesotho is apparently due to fewer FHHs with a male adult, while the marked change for Ghana arises from a drop in the share with no adult male, which swamps a smaller increase in FHHs with a male member. In general, the patterns of change are quite different across the two household types (see Online Resource 2, Fig. S2.1).
Figure 2 provides nonparametric plots of the probability that a woman aged 15 or older heads a household controlling for her age and disaggregating across the regional country groupings, for the earliest and latest surveys. Distinct level differences are evident across the macro regions. However, in all regions and across the age distribution, adult women clearly are significantly more likely to be heads over time.
What factors are associated with rising prevalence? Turning to regression analysis, we examine the shares both of the population living in FHHs and of FHHs over the last 25 years or so by comparing DHSs across countries and years. Thus, the observations are country/year DHS-based. Because the share is bounded, we use its logit transformation to ensure normally distributed errors. The dependent variable is the natural log of the odds ratio—that is, ln(S / (1 – S)), where S is the fraction of the population living in FHHs (or of FHHs)—for each DHS, giving 98 total observations (98 surveys for 34 countries; Table S1.3, Online Resource 1). Ordinary least squares (OLS) regressions then examine how country-specific time-varying regressors are related to differences across countries and years. We considered using an estimator that allows for latent country-specific effects in the error term, which could be either fixed or random effects. Using a country fixed-effects model would have the advantage (as opposed to a random-effects model) of reducing estimation bias resulting from omitted time-invariant country factors. Unfortunately, given a relatively small sample, of which less than two-thirds of the countries have three or more observations over time and eight have only one, our power to identify effects within country would be limited. Furthermore, many factors of interest are unlikely to substantially vary within countries over the relatively short period observed. Much of the source of variance of interest will be across countries.11
Standard errors are robust and clustered at the country level, allowing for correlated residuals within countries. We thus allow for a general structure of correlated residuals within clusters. However, we do not have sufficient time series observations to credibly identify a first-order (or any order) serial correlation parameter for the errors.
Although closely related and both rising over the period, the shares of FHHs and of the population living in FHHs are distinct variables that can be differentially affected by factors such as changes in fertility, mortality, education, residential, and marital patterns. However, the two sets of regressions produce remarkably similar coefficients, slightly different in magnitude but qualitatively identical. As a result, and in the spirit of brevity, our discussion focuses only on the share of the population living in FHHs. The regressions (and variance decomposition) for the share of FHHs are presented in Tables S2.1 and S2.2 (Online Resource 1).
Column 1 of Table 3 presents the underlying Africa-wide time trend between 1990 and 2013 based on a regression of the log of the odds ratio on time. The statistically significant trend is equal to 0.4 of a percentage point annual increase in the share of population in FHHs when evaluated at the mean sample share. The next regression additionally controls for factors or their proxies that the literature emphasizes as determinants of the preponderance of FHHs: log GDP per capita (based on year 2005 PPP at constant year 2011 international dollars); the agricultural share of GDP as a proxy for local employment opportunities and incentives for migration; HIV prevalence (%); the urban and Muslim population shares (%); whether the country experienced serious conflict over the last 10 years; and the female labor force participation rate.12 All variables are time-varying and country-specific.
None of the covariates account for the trend. However, HIV prevalence and the Muslim share—with positive and negative signs, respectively—are statistically significant correlates (column 2). Our full specification in column 3 adds average years of schooling, life expectancy, and average age of adults all separately by gender, as well as women’s mean age at first marriage.
Strikingly, the time trend vanishes upon inclusion of these demographic and human capital population attributes. On closer inspection, it appears to be specifically the combination of rising age at first marriage and higher education levels that eliminates the positive trend in the dependent variable. Given the high correlation between these two traits, it is difficult to say which is more salient. Higher age at marriage allows more schooling for girls but also exerts an important independent association with the share living in FHHs. It can be expected to have a disparate trajectory and evolve through time and across countries differently than the other variables. Indeed, the newly entered demographic and education variables are all highly correlated both across genders and among themselves. However, they appear to be independently related to the population share of FHHs since they retain individual significance.
The separate explanatory power of each factor can be better seen in column 4 of Table 3, which presents our preferred data-consistent specification based on tests of the estimated coefficients, as a restricted version of the full model in column 3. We cannot separate the effects of male and female education (F(1,33) = 0.03, p = .859). Together, neither is statistically significant given how correlated they are; yet, each is significant when entered singly. One interpretation is that it is the rise in overall schooling levels that matters and (as confirmed in the data) that the male-female gap has been changing less than the levels for both. Our restricted version in column 4 thus controls for the overall average although we could equally well control for female or male schooling instead. Evaluated at dependent variable mean, an extra year of schooling is associated with a 3 percentage point increase in the share of population in FHHs.
On average, a one-year rise in women’s age at first marriage is related to a 2.5 percentage point increase in the population share in FHHs—an association almost as large as that of an extra year of schooling.
As with education, a test that the coefficients on male and female life expectancy are the same cannot be rejected (F(1,33) = 0.01, p = .913). We replace these with overall life expectancy, which is significant at the 1 % level. Again, this is an arbitrary choice given that the individual measures are each significant on their own and not dissimilar in magnitude. This independent variable’s positive association―equal to 0.5 of a percentage point boost in the share of population living in FHHs per extra year―reflects the high prevalence of widowed heads among older women relative to older men.
Finally, a test of the homogeneity restriction on age passes (F(1,33) = 0.04, p = .845), indicating that it is the gender age gap that matters. We replace the gender-specific measures by the gap measured as women’s minus men’s average age, which measures differential age distributions reflecting differential age-related mortality. An additional year added to the gap is associated with a 2 percentage point increase in the population share in FHHs.
Among the correlates previously entered in column 2, log GDP per capita (significant only at the 10 % level) and conflict emerge as statistically significant after the population and human capital characteristics are added (columns 3 and 4). A simple correlation indicates that higher-income countries tend to have larger proportions of population living in FHHs. However, as shown in Table 3, after other factors are taken into account, higher GDP is found to be associated with a reduction in the population share in FHHs. This finding is presumably due partly to lower work-related migration by men, associated with a growing local economy. But the magnitude of the correlation is small: a 5 % increase in GDP is correlated with an estimated 0.3 percentage point reduction in the FHH population share. The female labor force participation rate plays no independent role. A recent conflict, by contrast, is associated with a 3 percentage point rise in a country’s FHH population share and is statistically significant at the 1 % level or lower.
The HIV prevalence rate retains a significant and sizable positive association. At mean points, for each percentage point increase in HIV prevalence, we find a 0.7 of a percentage point increase in the dependent variable. In recent years, HIV prevalence rates have tended to be higher for women across African countries (UNAIDS et al. 2004). However, the positive effect of HIV can be explained by the fact that there are still vastly more MHHs than FHHs. When a male head perishes, a spouse or other adult female member is available to assume headship more often than men can assume headship when a female head succumbs to AIDS. On balance then, HIV prevalence will tend to move hand in hand with the prevalence of FHHs.
The Muslim share has no explanatory power in either the complete model or the preferred restricted form, presumably because of the high correlation between majority Muslim countries and country attributes, such as age at first marriage and years of education. Religion per se does not appear to be a decisive factor.
The next two sets of regressions in Table 3 repeat the analysis for the subsamples of FHHs with and without a male adult. We find similarities in the models but also marked differences, underlining the need to treat these groups separately. Again focusing on the restricted model, we first note similar trend increases, accounted for by demographics and education. Independently of these, a 1 % increase in the Muslim population is associated with a reduction of 0.07 of a percentage point in the population share in FHHs with no male. This finding possibly reflects the continued practice of polygyny together with social and cultural norms associated with the Muslim religion as practiced in much of Africa that encourage women who suffer marital dissolutions to remarry quickly or be absorbed into MHHs unless they have a male protector, such as an adolescent son (hence the positive effect on FHHs with a male, column 7) (Lambert et al. 2017; Tabutin et al. 2004). The other difference for this group is the insignificance of age at first marriage as a correlated factor. Urbanization, presumably through effects on social norms and the acceptance of female headship, is positively correlated with the prevalence of FHHs with a male adult. However, the association is small, at 0.1 of a percentage point increase for each extra percentage of the population living in urban areas.
Interestingly, women’s education is related to the population in FHHs with male adults (with an additional year associated with a 1.8 percentage point increase), and men’s is related to the share in FHHs without a male (a 1.4 percentage point increase). We interpret the latter to indicate that higher male education leads to work-related male migration with women left behind as heads. The former may signal a number of influences: more-educated women are more autonomous and empowered to assert headship over a son or grandson. Together with higher age at marriage, expanded education will also tend to lengthen the coresidence of adolescents with parents and possibly more so in FHHs that have been found to emphasize schooling and girl’s education, as noted earlier.13
Table 4 presents decompositions of the explained variance of our restricted regressions.14 Overall, the regressions account for 72 % to 80 % of the variance in the population shares living in FHHs across Africa. The decomposition is similar for the overall share and the share living in FHHs with an adult male. Years of education, the gender age gap, and age at first marriage each contribute approximately 20 % and together account for approximately 60 % of the explained variance (59.2 % and 61.3 % for the overall share and the share with an adult male, respectively). HIV prevalence systematically accounts for some 12 % across all three groups. The Muslim share adds 9.5 % to the explained variance of the overall dependent variable and 7.0 % to that of FHHs with a male adult. For both, the association is positive, although its role in the regressions was not statistically different from zero. Its contribution rises to 26 % for the variance of FHHs with no male, but here the correlation is negative, indicating that the share of population living in FHHs without an adult male is significantly lower in preponderantly Muslim countries.
Recent Changes in Poverty by Household Headship
We compute poverty headcount indices for the entire population and separately for the population living in FHHs and MHHs at two dates for 24 countries. Table 5 lists the countries, survey dates, and years between surveys. Columns 3 and 4 give the headcount indices for the earlier and later survey dates. The annualized changes in poverty are presented first for the entire population and then for the population distinguished by whether they have a male or female head. In 19 of 27 country periods―more than 70 %―poverty fell. The same outcome characterizes MHHs in all the same countries; for FHHs, however, poverty fell in an additional three countries, or 22 of 27 cases.
Figure 3 plots annualized absolute rates of change in the headcount index for MHHs on the y-axis against the same for FHHs on the x-axis. (The underlying numbers are given in columns 6 and 7, Table 5.) An examination of Fig. 3 makes clear that overall, in countries that have seen a spell of poverty decline, poverty has been falling at a faster pace among FHHs.15 Where poverty has risen, households headed by women have experienced a less pronounced rise, with one exception: Kenya, 1994–1997. In 20 cases, FHHs appear to have fared better or no worse than MHHs. However, tests of whether these differences are statistically significant indicate that in 11 cases, FHHs had a significantly better poverty performance, and MHHs did so in 2 cases; however, in the rest, no difference is apparent (Table 5). We thus conclude that in 25 of 27 cases, FHHs fared better or no worse than MHHs.
What has the poverty reduction record revealed above meant for the overall contribution to falling poverty rates of FHHs versus MHHs? The decomposition in Table 6 shows that after the smaller share of the population belonging to FHHs is allowed for, MHHs contributed more to the overall changes in the majority of countries. When we aggregate across all countries and weight by the country’s total population share, FHHs account for 27 % of the overall reduction in poverty. Almost all the rest is attributed to poverty reduction among MHHs; the changing share of FHHs makes a negligible contribution. In three countries, FHHs accounted for the majority of the reduction in poverty (Namibia, Nigeria, and South Africa). In the seven spells for which overall poverty increased, the contribution of FHHs to the increase was less than that of MHHs.
Perhaps living standards have evolved differently for different types of FHHs. Some may have systematically done better due to their distinctive demographics or household structure. We test the sensitivity of our findings to heterogeneity among FHHs by calculating poverty measures for those with and without a male adult; and separately for those with and without a currently married head, and again by whether they also contain a male adult. FHHs headed by a widow versus those who are not are also examined. Finally, we repeat the exercise for FHHs and MHHs, using consumption divided by the square root of household size to allow for scale economies in the consumption of smaller households. This approach allows us to see whether our conclusion of faster falling poverty for FHHs is robust to assumptions on how household size affects consumption well-being. These results are discussed, but only those for the four-way disaggregation of FHHs are shown.
Annualized changes in poverty for households with married and unmarried female heads, including a male adult or not, are shown in the last four columns of Table 5.17 A simple comparison of annual rates of poverty change shows that in all cases, MHHs were outpaced by at least one among the FHH types. Among the latter, however, we find much heterogeneity in performance with no overall pattern across countries or macro regions. FHHs with a married head and a male adult performed best in eight country/year spells, followed by households with a married head but no resident male (seven cases) and those with an unmarried head and a male (seven cases), and finally by households with an unmarried head and no male (three cases). Although a FHH category always fared better than MHHs, others among the FHH types often fared worse. Thus, a very mixed picture emerges in how different types of FHHs fared across countries with no obvious patterns across countries and periods.
Disaggregating FHHs into those headed by a widowed or nonwidowed head provides a similar conclusion. In only two cases did MHHs fare significantly better with respect to annual changes in poverty. But in the rest, widow-headed households experienced the highest annual rate of poverty reduction (or lowest poverty increase) in 11 cases (with the difference being statistically significant in 6 cases), while households headed by nonwidows did so in 10 cases (4 of which are significantly different from zero).18
The results are also robust to using poverty measures that allow for scale economies in consumption.19 In this case, FHHs outperformed MHHs in progress against poverty in 16 of 27 country spells. The differences, however, are statistically significant only in 12 cases—9 in favor of FHHs, and 3 in favor of MHHs. Here again, in the majority of cases (24 of 27), FHHs fared better or no worse than MHHs.
Poverty is falling overall, the prevalence of FHHs is rising, and on balance, FHHs have seen faster poverty reduction. Did the share of the poor accounted for by FHHs also fall over this period? Our findings indicate a mixed picture (Table 7 and Fig. S2.3, Online Resource 1). The share of the poor living in FHHs increased in 14 spells but declined in the remaining 13—an almost equal number of spells. A similarly mixed pattern is evident in each of Africa’s macro regions.
Living standards have risen, and poverty has fallen considerably across sub-Saharan Africa since the late 1990s. There have been concerns that some groups with above-average poverty rates may be left behind. To the best of our knowledge, our article is the first to focus specifically on female-headed households and to ask how this group fared during this period. Drawing on the microdata from virtually all available national surveys for the region, we demonstrate that the shares of households headed by women as well as of the population living in them increased over time across the continent. However, higher prevalence is consistent with the finding that poverty has declined. We find that higher GDP itself is associated with lower rates of female headship. However, other changes have occurred across Africa. Changes in demographic and population characteristics—particularly, women’s age at first marriage and education—are positively correlated with female headship.
We tried to reconcile this finding with the recent aggregate reduction in poverty. Poverty declined for both household groups, but in most countries for which the data are adequate, it fell faster for FHHs as a whole. Allowing for the diversity among FHHs reveals marked differences in poverty changes among them but no systematic patterns. The living standards of the various types of FHHs followed dissimilar paths across countries and periods, with no one type consistently outperforming the others. One category of FHH fared well in one country or period, while another fared best elsewhere. A decomposition of the change in poverty indicates that rather than putting a break on poverty reduction, FHHs contributed appreciably to the overall decline in poverty despite their smaller overall share in the population.
Why has poverty fallen faster for FHHs? Perhaps poor FHHs have faced relatively high economic returns to the new opportunities unleashed by growth; or perhaps they have benefited disproportionately from the expansion of social protection in the region; or perhaps the group of people living in FHHs has fundamentally changed over time. This new stylized fact about poverty in Africa warrants further research.
The authors are grateful to Elena Bardasi, Mayra Buvinic, Luc Christiaensen, Louise Grogan, Sylvie Lambert, Martin Ravallion, Adam Wagstaff, the journal’s three referees, and seminar participants at the World Bank, the University of Fribourg, Switzerland, and the 2015 ECINEQ Conference for useful comments. They also thank the World Bank Research Support Budget for its funding support. These are the views of the authors and need not reflect those of the World Bank and its affiliated organizations.
Begun with the seminal paper by Buvinic et al. (1978).
Posel and Rogan (2012) explored the issue for South Africa for 1997–2006.
Regressions using the DHS-based alternatives give similar results.
At the time of writing, it was unclear what the equivalent poverty line would be when using the newly released 2011 PPP. However, different PPPs do not affect comparisons within country, the focus here.
We adjust the poverty line to $2.80/day (equivalent to $1.25 when switching to the square root scale) using a pivot point of N = 5 (Ravallion 2015).
Comparability was defined based on the consumption module and survey design: same survey months, similar survey design, diary/recall consistency, and national representativeness.
Similar FHH prevalence rates are obtained using the consumption surveys.
The wealth index is a composite measure of a household’s living standards generated using principal components analysis using data on the ownership of assets and amenities. It places households on a continuous scale of relative wealth and into wealth quintiles.
Table S1.2 in Online Resource 1 provides the list of countries by macro region. The sample is restricted to the 26 countries with at least two surveys and uses the earliest and latest years (52 surveys total).
Chad (1996–2004), Congo (2005–2011), and Lesotho (2004–2009) have fewer years between surveys than other countries (Table S1.2, Online Resource 1).
When we run the model with country fixed effects, we observe two main findings. First, conditional on all controls, only years of education and age at first marriage retain a significant association with the dependent variable. We conclude that, as expected, the fixed-effects estimator has little power for identifying the effect of variables that vary more across than within countries over time. Second, the trend remains significant and unaccounted for by the regressors. Thus, although the fixed effects pick up omitted factors and explain within-country variance, this is insufficient to account for changes over time and is not the result of the dropped observations: the trend remains fully explained by the OLS run without the eight countries.
Online Resource 1 provides data definitions and sources.
Our findings are not inconsistent with evidence that higher education and age at first marriage are associated with lower divorce prevalence (Clark and Brauner-Otto 2015). These and other factors may simultaneously reduce divorce rates but put upward pressure on the formation of FHHs through other processes, such as fewer remarriages following a dissolution.
Computed using the REGO command in STATA (Huettner and Sunder 2012). REGO decomposes the R2 into the contribution of the regressors into Shapley (for individual regressors) or Owen values (for within-group regressors). We restrict the decomposition to Shapley values.
Logging the changes and plotting the proportional changes gives a similar picture.
This is a type of decomposition proposed by Ravallion and Huppi (1991).
The corresponding decomposition is not presented because it provides little additional information given the small group sizes.
The number of country spells is sometimes lower because of missing marital status for some countries.