Has Child Marriage Declined in sub‐Saharan Africa? An Analysis of Trends in 31 Countries

Child marriage, defined by the United Nations as marriage before the age of 18, disproportionately affects girls and hinders progress toward development and public health goals. The magnitude of the threat child marriage poses to global development is highlighted by the fact that its elimination is one of the specific targets for achieving goal number five of the UN's Sustainable Development Goals: gender equality and the empowerment of all women and girls. Women who marry as children have fewer years of schooling than those who marry as adults, potentially leading to lower labor force participation and poorer long-term economic opportunities for themselves and their families (Jejeebhoy 1995; Field and Ambrus 2008; Parsons et al. 2015). Child marriage is also harmful to girls’ health. Married girls begin childbearing earlier, give birth at shorter intervals, and report having more unwanted pregnancies than their peers who marry as adults (Raj et al. 2009). All of these outcomes are associated with obstetric complications, which are the leading cause of death among young women in low- and middle-income countries (Patton et al. 2009). Young married women also have higher rates of HIV infection than their unmarried, sexually active peers (Glynn et al. 2001), likely as a result both of marrying older men who are already infected and of having limited agency to negotiate condom use (Bracher, Santow, and Watkins 2003; Kelly et al. 2003; Clark 2004). Very few studies differentiate the effects of marriage among very young girls from those among older adolescents. However, marriage may have more severe consequences for girls at the younger end of the age spectrum, say below the age of 15, relative to girls between 15 and 17 years of age. The educational opportunities for very young girls and their ability to develop social networks outside of their families may be curtailed to a greater degree simply as a result of their age. Raj (2010) found that girls in India who married before age 14 were less likely to have any formal schooling than those married between ages 16–17 and that the odds of experiencing spousal violence increased as age at marriage declined. In another study from South Asia, Godha, Hotchkiss, and Gage (2013) reported that women married before age 14 were less likely to use contraception prior to their first pregnancy and had more unintended pregnancies relative to girls married between 15 and 17 years of age, suggesting that control over sexual activity and reproductive decisions is even further diminished at younger ages. Moreover, the relative physiological immaturity of younger girls may put them at greater risk of acquiring sexually transmitted infections and experiencing obstetric complications (Moss et al. 1991; Nove et al. 2014). Age at marriage is rising throughout sub-Saharan Africa. The singulate mean age at marriage is now greater than 18 in the vast majority of countries in the region and in all of the countries included in our analysis (United Nations, Department of Economic and Social Affairs, Population Division 2015). Even so, sub-Saharan Africa has the highest rates of child marriage in the world, and previous studies have estimated that more than half of girls marry before age 18 in many countries in the region (Singh and Samara 1996; Mensch, Singh, and Casterline 2006). Measuring trends in the prevalence of child marriage over time is important for understanding where the practice is most common and for evaluating the effectiveness of efforts to eliminate it. However, measuring age at marriage in sub-Saharan Africa is difficult. Unlike Western marriages, which are often unambiguously dated by a ceremony, the signing of legal documents, and civil registration, marriage in sub-Saharan Africa is often described as a process consisting of multiple stages including legitimized sexual relations, cohabitation, and ceremonies. The process can be lengthy and the various stages occur in different sequences across ethnic and social groups (van de Walle and Meekers 1994; Locoh 1994; Arnaldo 2004). If several events are required to solidify a union, it may be unclear when the union was formalized. The Demographic and Health Surveys (DHS) were developed to measure fertility rates in a comparable manner across developing countries. The surveys estimate the total number of women at risk of childbearing and include broad questions intended to identify stable unions that may result in children. All female respondents are asked to report their current marital status and the month and year they were first married or began cohabiting with a partner. Those who report that they are married or living with a partner are considered in union. These questions result in imperfect data on age at marriage. Where union formation consists of multiple stages, it is unclear which point in the process corresponds to the reported month and year of marriage. Despite these limitations, the DHS are the most widely available nationally representative source of information on age at marriage in sub-Saharan Africa and throughout the developing world. The DHS program's focus on international comparability has facilitated multiple studies that compare change in age at marriage across developing regions as well as numerous studies that examine trends specific to sub-Saharan Africa (Singh and Samara 1996; Harwood-Lejeune 2001; Westoff 2003; Jensen and Thornton 2003; Garenne 2004; Mensch, Singh, and Casterline 2006; Mensch, Grant, and Blanc 2006; Shapiro and Gebreselassie 2013). Most of these studies have used a single DHS wave to estimate the proportion of women interviewed at different ages who reported being married before a specified age. For example, Mensch, Singh, and Casterline (2006) used the most recent survey wave in 27 sub-Saharan African countries to calculate the proportion of women interviewed between ages 20–24 and 40–44 who reported being married before the age of 18. A few studies have presented estimates of trends in age at marriage across birth cohorts rather than age groups. Jensen and Thornton (2003) reported trends in age at marriage over birth cohorts from 1950 to 1970 using a single DHS wave. Westoff (2003) and Garenne (2004) pooled data from World Fertility Surveys and DHS waves within countries to estimate trends over birth cohorts between 1925 and 1979. Only a handful of studies have focused on trends in child marriage (Singh and Samara 1996; Jensen and Thornton 2003; Mensch, Singh, and Casterline 2006; Mensch, Grant, and Blanc 2006). Measuring change in age at marriage by comparing reports from women of different ages leads to potential measurement error. Older DHS respondents report less complete information on age at marriage, meaning that their data are more frequently imputed than data for younger women (Gage 1995). Older women are also known to report that events occurred closer to the time of the survey than they actually did, a bias referred to as forward displacement and well recognized in studies using DHS data (Blanc and Rutenberg 1990; Gage 1995; Mensch, Grant, and Blanc 2006). If forward displacement of age at marriage affects DHS data, we would expect measures of change over time based on the comparison of women of different ages to systematically underestimate the magnitude of change. For example, if women aged 40–44 years tend to report that their first union took place closer to the date of the interview than it actually did, this would result in fewer women in this age group being classified as married before the age of 18 than actually were. This would lead to erroneously small estimates of change over time when compared with reports from women aged 20–24. Moreover, child marriage is strongly associated with poverty and maternal mortality at the national level (Raj and Boehmer 2013). Concerns regarding the validity of age-based comparisons for estimating change over time deepen if women who married as girls are more likely to die from maternal or poverty-related causes earlier than their peers who married as adults. This differential probability of survival would also lead to underestimates of change in the prevalence of child marriage over time. Recent work by Neal and Hosegood (2015) provides further reason to avoid using women of different ages to estimate trends. The authors documented significant inconsistencies in the reporting of reproductive events among women born in the same cohort but interviewed at different ages. They estimated the prevalence of marriage before age 15 among a sample of women born in the same cohort who were interviewed between ages 15 and 19 and again, five years later, when they were between ages 20 and 24. The estimated prevalence of marriage before the age of 15 differed markedly between these age groups. In the analysis that follows, we measured trends in the prevalence of marriage before age 15 and before age 18 in 31 countries over a 25-year period. We pooled data from all available DHS waves within each country to measure change over the same time period across all countries. We attempted to avoid biases resulting from the forward displacement of events and selective survival by using a sample of women who were interviewed at the same age. We then compared our results with estimates of change obtained by comparing women who were interviewed at different ages using a single DHS wave. For our primary analyses we used DHS data from 31 sub-Saharan African countries that had conducted at least two standard survey waves. These analyses rely on retrospective reports of age at first marriage provided by female respondents. We began by pooling data from all available survey waves within each country and then limited our sample to women who were born between 1965 and 1994 and were between ages 20 and 24 at the time they were interviewed. Sample sizes by country are listed in Table 1. Including only women interviewed between ages 20 and 24 avoids censoring of women below age 18 and allows us to minimize potential biases resulting from forward displacement of age at marriage and selective survival. We estimated the proportion of women who reported being married before 15 and 18 years of age over consecutive five-year birth cohorts using logistic regression. We regressed dummy variables for each birth cohort on a binary indicator of marriage before either age 15 or 18 and then used the coefficients associated with each dummy variable to predict the probability of marriage before that age for each birth cohort. We graphed these estimates to illustrate trends over a 25-year period from 1965–69 to 1990–94. We obtained estimates for each birth cohort in countries that conducted surveys at approximately five-year intervals between the late 1980s and the 2010s. In countries that had longer intervals between survey waves, we were only able to obtain estimates for some of the birth cohorts. Notably, birth cohorts do not correspond exactly to DHS waves; in many countries women born within the same cohort were interviewed in different survey waves. Some of the five-year cohorts do not include women born in every year. To ensure that our estimates are reasonably representative of the entire five-year cohort, we report results only for cohorts in which at least three birth years were represented by a minimum of 100 women. To quantify the magnitude of change over a period of 15–20 years, we also calculated prevalence differences by subtracting the proportion of women born between 1970 and 1974 who reported being married before ages 15 and 18 from the proportion of women born between 1985 and 1989 or 1990 and 1994, depending on data availability. In countries for which were able to estimate the proportion in both 1985–89 and 1990–94, we used the latter to measure change over the longer time period. We conducted four types of sensitivity analyses to test the robustness of our measurement approach. First, as mentioned above, some of the five-year cohorts are represented by three or four individual birth years. To examine whether our estimates are sensitive to the fact that some birth years are not represented, we also measured trends over three-year cohorts in which all birth years were represented (analyses not shown). Second, we assessed the sensitivity of our results to the choice of age range at the time of interview by measuring trends using retrospective reports from women between 25 and 29 years of age. Third, we compared our estimates of change in the prevalence of marriage before age 18 based on the analysis of pooled DHS data with estimates from the more common approach, which measures change by comparing women of different ages interviewed in a single DHS wave. For the latter analysis, we compared the proportion of women who reported being married before age 18 among those interviewed between ages 20–24 and between ages 40–44. This comparison was limited to a sample of seven countries in which we were able to estimate the prevalence of child marriage among women born between 1970 and 1974 and between 1990 and 1994 and that had conducted a DHS between 2013 and 2015 to ensure estimation over a similar time span. Most of the women aged 40–44 and interviewed in 2014 would have been born between 1970 and 1974; those aged 20–24 would have been born between 1990 and 1994. Finally, to assess the robustness of our primary results, we compared measures of trends based on retrospective reports of age at first marriage with measures based on the current marital status of girls under the age of 18. For these analyses, we took advantage of the repeated cross-sectional nature of the DHS and national censuses. We obtained census data from 14 countries that conducted at least two national censuses between 1985 and 2015 from the Integrated Public Use Microdata Series (IPUMS) (Minnesota Population Center 2015). We calculated the proportion of ever-married girls aged 15–17 interviewed in consecutive DHS surveys and recorded in each census using the same regression approach described above. However, we controlled for age in these analyses because estimates may be sensitive to changes in the distribution of age across survey and census waves. We also used national censuses to measure trends in the proportion of ever-married girls aged 12–14 included in consecutive censuses. We were unable to use DHS data for this younger age group because household rosters only record the marital status of persons aged 15 years and older. Although estimates of trends based on current marital status avoid reliance on retrospective reports, they introduce a new set of limitations. First, there is a large problem with censoring, as some women interviewed between ages 15 and 17 may have gone on to marry before their 18th birthday. This means that the estimated levels of child marriage obtained from cross-sectional measures are expected to be lower than those estimated in our primary analysis. Second, census data are reported by the head of household, introducing an additional source of measurement error. The person responding to the census questions may provide less accurate information on the age and marital status of the girls within the household than the girls themselves would. Lastly, censuses are conducted less frequently than household surveys, which means that we have fewer data points and cannot characterize trends as well as we can with DHS data. All of the estimates based on DHS data are weighted using de-normalized sampling weights following guidelines for the use of pooled data included in the DHS Sampling and Household Listing Manual (ICF International 2012). The de-normalization procedure requires estimates of the target population in each country at the time of each survey. We obtained estimates of the population of women aged 15–49 years in each survey year in each country from World Population Prospects 2015 (United Nations 2015). Estimates based on census data are weighted using the country- and sample-specific person weights provided by IPUMS. Trends in the prevalence of child marriage over time are presented by country and geographic region in Figures 1-3. There is striking heterogeneity between and within regions. West Africa has some of the highest rates of child marriage in the world. More than half of women born in the most recent cohort were married before age 18 in four of the 12 countries in this region: Burkina Faso, Guinea, Mali, and proportion in The prevalence of child marriage below 25 in only two countries in West and Child marriage is less common in where was the only country in which half of women born between 1985 and 1989 reported being married before 18. than of girls born in the same cohort were married as children in from and African countries are but prevalence estimates within the birth cohort, from a of in to in Trends in the prevalence of marriage before 15 and 18 years of age in West as reported by DHS respondents aged 20–24 and years Trends in the prevalence of marriage before 15 and 18 years of age in as reported by DHS respondents aged 20–24 and years Trends in the prevalence of marriage before 15 and 18 years of age in and as reported by DHS respondents aged 20–24 and years There is of a in the prevalence of marriage before age 18 in many but the of to have among recent birth cohorts in and and rates to be rising in In the prevalence of child marriage has only in and Burkina Faso, and of but over time. the we or of a in the proportion of girls married before age 18 in and presented in Table these The the change in the of women married before 15 and 18 years of age over a period of 15–20 years. Only countries for which we were able to estimate the prevalence among women born in and either 1985–89 or are The prevalence of marriage before age 18 in only 12 of these 20 countries and increased by points in over a period between and Figures also trends in the prevalence of marriage before the age of More than 20 of girls born between 1990 and 1994 were married before age 15 in Mali, and This proportion in In fewer than of girls born in the same cohort were married before age 15 in and These that the majority of child occur between ages 15 and Although marriage before age 15 is less there is also less of a in the prevalence of marriage among these very young girls. the 20 countries listed in Table we found of a significant in marriage before age 15 in only There has been in child marriage in and marriage among very young girls has increased by to points over this time period in Mali, and a result of the small changes in marriage before age 15 compared to marriage before age 18, the proportion of child among very young girls has increased in countries as Burkina Faso, Mali, and For example, the prevalence of marriage below the age of 18 has in over a 25-year the prevalence of marriage below age 15 in that country has women born between 1970 and 1974 and married as 18 were married before age this proportion to 25 among women born between 1990 and 1994. estimates are not sensitive to the of the birth cohort used and are to the choice of age In most estimates based on retrospective reports of age at first marriage from women aged 20–24 and born in the same cohort do not In the few countries in which we do differences between age as Mali, and trends over time are in the same is that the level of child marriage based on reports from women aged is higher than the level based on reports from women aged 20–24 in and This not correspond to our that forward displacement bias would result in estimates from older women. Although the in age at the time of interview is the between the two groups that in these countries our estimates of the proportion of women married as children may be We not an in the prevalence of child marriage, before the age of We trends in and to ensure that the are not to changes in the DHS sample over time not shown). All of the in were represented in each of the DHS waves used in this analysis, and the prevalence of marriage before ages 15 and 18 to have increased in some of them between 1970 and The in the prevalence of very marriage in occurred between women born in and those born in further change has been among more recent birth cohorts. The with the highest estimated prevalence of child marriage in the was not represented in the birth cohort. However, our results are not sensitive to this we found that the prevalence of marriage before age 15 increased even all data from from our analysis. in the of in over time us from trends. Estimates of the proportion of ever-married girls between ages 12–14 and 15–17 across survey and census waves are in Figures In most countries trends based on cross-sectional measures the same as trends based on retrospective reporting of age at first marriage. is important to than in some estimates from the most recent survey or census trends among women born in more recent years than those included in our primary estimates in Figures 1-3. This is the in and where estimates of trends in the prevalence of child marriage depending on whether we measure changes over birth cohorts or over year. For example, based on our primary analysis we that there had been change in the prevalence of marriage before age 18 in over a period birth cohorts from to However, cross-sectional estimates based on DHS and census data in a in the proportion of ever-married girls aged The most recent DHS in was conducted in and girls aged 15–17 interviewed in that survey were born between and birth years that are not represented in the estimates in The same is of census girls aged 15–17 recorded in the census would have been born between 1990 and This that it have been to a very recent in marriage before 18 years of age in if we have our primary Notably, census data in that the prevalence of marriage among girls aged 12–14 has not even among more recent birth cohorts. of ever-married girls aged 12–14 and 15–17 years included in DHS and census waves in West Africa of ever-married girls aged 12–14 and 15–17 years included in DHS and census waves in Africa of ever-married girls aged 12–14 and 15–17 years included in DHS and census waves in and Africa estimates from censuses in and in the proportion of ever-married girls aged 12–14 that are not in estimates of marriage before age 15 based on DHS data. In these three girls aged 12–14 recorded in the most recent census were born in years recent to be represented in our primary DHS the level of between estimates in Figures and census data in this that marriage among very young girls be on the in those countries as However, these results be with as the time period under between measurement and estimates of change are sensitive to these This is in where we a similar Estimates based on retrospective reports of age at first marriage from women aged 20–24 that the prevalence of marriage before age 15 by points in between 1970 and 1994 Table However, using census data we estimated that the proportion of ever-married girls aged 12–14 increased by points between the censuses conducted in and in of birth results from census data can be as the in the proportion of ever-married girls aged 12–14 born over an approximately period between 1974 and further using DHS data to measure the change in the prevalence of marriage before age 15 reported by women who were interviewed at ages 20–24 and born between and to birth cohorts for girls aged 12–14 recorded in the two we found a Finally, we compared our estimates of change in the proportion of women married before 18 years of age to estimates obtained by comparing women interviewed at different ages using a single DHS wave. from the two measurement are as in Table Estimates of change in the prevalence of child marriage over a period differed by less than points in all of the countries In and we estimated in the prevalence of child marriage when comparing women from different birth cohorts. However, in and the birth cohort comparison measures of change than estimates based on the comparison of age groups from a single DHS survey. These results of forward displacement bias across this sample of countries. before age 18 has less common throughout of sub-Saharan but more than one of girls to marry before in well over half of the countries in this progress toward child marriage has not been in some of these and the prevalence has for 20 years in seven of including and which have the highest prevalence of child marriage in the In to the rising mean age at marriage throughout sub-Saharan these that age at marriage is not rising within individual countries. Some of the population marry at very young to marriage ages. in the prevalence of child marriage have been among girls aged 15–17 years. More than half of the countries we significant progress toward the prevalence of marriage among girls younger than 15 This of progress to marriage among the and most girls. This is in Burkina and where marriage before age 18 has by and 17 there has been change in the prevalence of marriage before age 15 over the same period. Even in countries where have they have been 20 countries and were the only two in which the prevalence of marriage before age 15 by more than The prevalence of marriage among very young girls has in Mali, and We attempted to avoid known of bias in the measurement of age at marriage by comparing women who were interviewed at the same age across birth cohorts. However, of marriage have likely over the time period in to social educational and labor force opportunities for as well as the or of and marriage is that women born and interviewed in more recent years greater to report that they were older at the time of their marriage than they actually potentially leading to changes in the degree of measurement over time that bias our if women born in recent cohorts are more likely to report that they were 18 or older at the time of marriage, then we would expect our results to in the prevalence of child marriage. We few of this in our results, among very young girls. levels of child marriage throughout of sub-Saharan Africa efforts to the 25 of the 31 countries in our analysis had set a minimum legal age for marriage at 18 years or older (United Nations Guinea, the of and were and had legal between 15 and 17 years of age. All 31 countries to the minimum in the of from or or (United Nations Center Although these an important for the of our estimates that they are to eliminate the the numerous to the minimum age, even of is to eliminate child marriage. A found that for girls’ was one of the few that have been to child marriage and of the studies included in that were conducted in sub-Saharan Africa. in and found that the of schooling by or on the of child marriage et al. 2006; et al. and et al. et al. 2015). A in that provided girls with and economic to in and facilitated the associated with child marriage was the only study to differentiate between effects on marriage among very young girls and effects among older adolescents. The proportion of girls between ages and 14 who were married by approximately points a the had on the marriage of girls between ages 15 and 19 and 2009). There is a to before child marriage is and Sustainable Development can be is to the social of child marriage within sub-Saharan Africa and across the In we more marriage among the girls and the social that it in the of a rising in the age at marriage throughout most of sub-Saharan Africa. The of progress over a period in some countries the for at marriage. There is based on data from sub-Saharan Africa that for girls’ can child marriage. the described in these studies may results that efforts have been unable to but this a on the of and to and educational opportunities for girls has the potential to marriage and likely progress toward a of additional development but be to ensure that all girls have to these

Hali tarjima qilinmagan