BACKGROUND:

Achieving gender equality in education is an important development goal. We tested the hypothesis that the gender gap in adolescent education is accentuated by illnesses among young children in the household.

METHODS:

Using Demographic and Health Surveys on 41 821 households in 38 low- and middle-income countries, we used linear regression to estimate the difference in the probability adolescent girls and boys were in school, and how this gap responded to illness episodes among children <5 years old. To test the hypothesis that investments in child health are related to the gender gap in education, we assessed the relationship between the gender gap and national immunization coverage.

RESULTS:

In our sample of 120 708 adolescent boys and girls residing in 38 countries, girls were 5.08% less likely to attend school than boys in the absence of a recent illness among young children within the same household (95% confidence interval [CI], 5.50%–4.65%). This gap increased to 7.77% (95% CI, 8.24%–7.30%) and 8.53% (95% CI, 9.32%–7.74%) if the household reported 1 and 2 or more illness episodes, respectively. The gender gap in schooling in response to illness was larger in households with a working mother. Increases in child vaccination rates were associated with a closing of the gender gap in schooling (correlation coefficient = 0.34, P = .02).

CONCLUSIONS:

Illnesses among children strongly predict a widening of the gender gap in education. Investments in early childhood health may have important effects on schooling attainment for adolescent girls.

What’s Known on This Subject:

In most developing countries, girls are less likely to complete secondary school. Much attention is given to the importance of gender inequalities, yet there is a lack of insight into mechanisms to close the gender gap while improving child health.

What This Study Adds:

We illuminated the pathways connecting child health and gender inequalities. The gender gap in adolescent education is accentuated by illnesses among young children in the household, because adolescent girls in developing countries are often tasked with child care and chores.

Gender equity in education is fundamental for achieving the sustainable development goals (SDGs), which are shared goals agreed on by member states of the United Nations.1,2 SDGs 3, 4, and 5 call for healthy lives at all ages, inclusive and quality education, and gender equality and the empowerment of women, respectively. These SDGs recognize the intersecting roles of child health, education, and gender equity in promoting a better furture.3 Although disparities in primary education have decreased substantially over the last 2 decades, the gender gap in completion of secondary education persists. In Sub-Saharan Africa and South Asia, boys are 1.55 times more likely to complete secondary education than girls.4,5 Thus, many girls will miss out on the benefits of a secondary school education, including the reduced risk of HIV acquisition, delayed sexual debut, fewer pregnancies during the teen years, higher wages and employment later in life, and improved educational attainment for their children.4,6,8 The potential gains from increased participation of women in the labor force (mostly in low- and middle-income countries [LMIC]) to global economic output over the next 10 years are estimated at $12 trillion.9,10 Identifying actionable contributors to gender disparities in secondary education is therefore of broad relevance to health and development.

Despite its importance, causes for the gap in school attendance between adolescent girls and boys residing in LMIC are poorly characterized. In this article, we examine the role of child illness in explaining the gap in adolescent schooling. Our conceptual framework suggests that the gender differences in household responsibilities result in the substitution of older girl children for maternal care of a younger sibling who is ill. If older girls are preferentially tasked with household chores from an early age, then the illness of a younger child in the household will raise the likelihood that adolescent girls will stay out of school to care for the sick child (and not raise the likelihood for adolescent boys).11,12 The deprivation of educational opportunities contributes to less-educated women and poorer outcomes for their children, which presents a vicious cycle of gender inequality (Fig 1).

FIGURE 1

Proposed mechanism linking childhood illness to gender gap in secondary education. The figure depicts the conceptual relationship between early childhood illness, the adolescent gender gap in schooling, and long-run health and socioeconomic outcomes. The diagram depicts a vicious cycle of gender inequality that can be broken at various points by interventions.

FIGURE 1

Proposed mechanism linking childhood illness to gender gap in secondary education. The figure depicts the conceptual relationship between early childhood illness, the adolescent gender gap in schooling, and long-run health and socioeconomic outcomes. The diagram depicts a vicious cycle of gender inequality that can be broken at various points by interventions.

Time-use survey data support the hypothesis that adolescent girls often participate in time-consuming household chores at the expense of their education.13 Moreover, multiple epidemiologic surveys also demonstrate that young children are ill frequently in LMIC.3,14,16 We provide in our analysis an empirical linkage between childhood illness and girls’ education. Our primary goal is to assess the magnitude of the pathway leading from childhood illness to a widening in the gap in schooling between adolescent boys and girls.

Characterizing such linkages could revise estimates of the returns to investments in child health in developing countries and aid in the design of complementary policies that capitalize on improvements in child health to retain adolescent girls in school, thus creating a virtuous cycle of improved health and development.3,16 

We employed individual-level data from the Demographic and Health Surveys (DHS) to examine the gap in the likelihood that adolescent boys and adolescent girls will be in school in relationship to illnesses among younger siblings. The DHS are nationally representative household surveys that provide information on the health of populations. In a second analysis, we examined whether the gap in school attendance would increase when the mother works outside the home because of the enhanced need for child care in the home. In a third analysis, we examined national data on the gender gap in adolescent education in association with national vaccination rates. We hypothesized that countries with high rates of childhood vaccination will experience lower rates of young child illness, thereby decreasing the need for adolescent girls to devote time to caring for sick children.

The primary data source of this analysis is the DHS.15 In this analysis, we used 44 surveys conducted in 38 countries between 1999 and 2013. DHS typically include between 4 and 10 000 households in which the educational status of all household members is enumerated. For our primary analysis, which is the analysis of differential school attendance by gender to child illness episodes, we used information from all households with (1) at least 1 child <5 years of age, (2) at least 1 adolescent girl, (3) at least 1 adolescent boy, (4) information about current school attendance for all adolescent boys and girls, and (5) information about illnesses among the household’s children <5 years old within 2 weeks preceding the survey. All households also had information on whether the mother works outside the home. Maternal work outside the home was identified by asking the following 2 questions: “Aside from your own housework, are you currently working?” and “Do you usually work at home or away from home?” The response rate for all the critical data elements (household census, child illness, and education status) was over 95% in the surveys used. We defined “adolescent” in this study as ages between 11 and 17 years old because this age group has been shown to be old enough to care for a young child but young enough to attend school in the 7 to 12 grade range.17 In the sensitivity analyses presented in Supplemental Fig 5, we varied the youngest age range from 10 to 12 and the oldest from 16 to 18. Our figures illustrate that the findings are stable within this range of ages. For each adolescent, we used the DHS question “Is the household member still in school?” as the principal indicator of school attendance.

In each DHS, mothers were asked about signs of recent illnesses of young children. We used queries about the presence of diarrhea, fever, or cough in the 2 weeks preceding the survey to indicate illness (see Supplemental Information for variable definitions and for wording of survey questions). The presence of any of these signs indicated an episode of illness, but we counted no more than 1 episode for any single child even if their mother reported multiple signs. More than 1 episode of illness in the household may occur, however, if >1 child had one or more signs of illness in the 2 weeks preceding the survey. Household records with missing information for the included variables were excluded from the analysis (see Supplemental Table 4 for details).

Finally, to assess the relationship between the gender gap and vaccination coverage, we used the DHS-prepared statistics platform (STATCompiler; DHS, Rockville, MD) for the vaccination rates obtained during each survey.16 We present the findings by using the proportion of children who were fully immunized (ie, children who received all 8 basic vaccines, including polio, diphtheria, pertussis, tetanus, and measles).

Relationship Between School Attendance and Young Sibling Illness

Households that frequently experience episodes of illness for children <5 years old may be different from other households in many ways that we can and cannot observe. The observed differences are readily apparent in the data (Table 1): households with a young child who was ill in the 2 weeks preceding the survey (∼50% of our sample) tend to have fewer assets and lower levels of maternal literacy. Hence, comparing how illness of young children affects the schooling opportunities for older girls across households will be confounded by unobserved factors that may affect both child illnesses and education of adolescent girls (eg, residence in remote areas may make securing clean water a challenge and may also make it difficult to reach school, accentuating a gender gap).

TABLE 1

Child and Household Characteristics by Gender and Illness Episodes

Illness EpisodesFull Sample
012BoysGirls
N 59 624 47 239 13 845 61 201 59 507 
Has electricity (% of households) 43.8 36.8 23.6 38.1 39.3 
Has car (% of households) 7.1 6.7 5.2 6.5 6.9 
Has telephone (% of households) 11.1 9.6 6.1 9.7 10.3 
Urban city (% urban) 38.1 35.4 25.3 35.0 36.1 
Mothers’ literacy levels (% illiterate) 34.3 49.4 65.6 44.9 43.0 
Highest education levels (% of teens)      
 No education 14.7 21.0 34.9 17.1 21.9 
 Primary 61.3 59.9 53.8 62.3 57.5 
 Secondary or higher 23.9 19.0 11.3 20.6 20.6 
Illness EpisodesFull Sample
012BoysGirls
N 59 624 47 239 13 845 61 201 59 507 
Has electricity (% of households) 43.8 36.8 23.6 38.1 39.3 
Has car (% of households) 7.1 6.7 5.2 6.5 6.9 
Has telephone (% of households) 11.1 9.6 6.1 9.7 10.3 
Urban city (% urban) 38.1 35.4 25.3 35.0 36.1 
Mothers’ literacy levels (% illiterate) 34.3 49.4 65.6 44.9 43.0 
Highest education levels (% of teens)      
 No education 14.7 21.0 34.9 17.1 21.9 
 Primary 61.3 59.9 53.8 62.3 57.5 
 Secondary or higher 23.9 19.0 11.3 20.6 20.6 

The table illustrates frequencies in the full analytic sample of 120 708 adolescent boys and girls of household and education characteristics by number of illness episodes and gender. All differences across the number of illnesses are significant at P < .001 (χ2). Household characteristics are different among households with 0, 1, or 2 illness episodes. These observed (and unobserved, fixed) differences are controlled for with household fixed effects.

To circumvent such biases, we instead looked within households and measured the differential probability of attending school for adolescent girls versus boys in relationship to the same illness episodes.17 This allowed us to isolate the role of factors that affect the schooling of boys and girls differentially within households and that interact with illness such as gender preferences for who should care for the sick child. Thus, we at least partially controlled for variables that persist over time within households, such as maternal education, household wealth, and cultural attitudes. Specifically, we regressed educational outcomes on gender; childhood illness episodes interacted with gender and indicator variables for each household in the sample (household fixed effects). For example, consider 2 households, each with 1 adolescent boy who is in school and 1 adolescent girl who is not in school. In the first household, no young children are ill, and in the second household 1 child is ill. Our models test the hypotheses that, within these households, the probability that the girl is out of school is greater than the probability that the boy is out of school, and additionally whether this probability is higher in households with a sick young child. The marginal effect of interest is then the interaction of childhood illness episodes with (older) girl gender. Please see Supplemental Information for the exact econometric specification.

Influence of Maternal Work on School Attendance

We next examine the influence of maternal work outside the home on the gender gap in school attendance associated with younger sibling illness. To do so, we estimated the same specification described immediately above and written out formally in Supplemental Information in 2 samples: 1 in which the mother works outside the home and 1 in which she does not.

In Supplemental Fig 7 and Supplemental Fig 8, we included 2 important extensions of the baseline specification. First, we separated the sample by rural and urban locations. Second, we used alternative DHS outcome variables, including years of schooling and whether the child has attended school in the current year. In Supplemental Fig 9, we included only households with older boy children as a “placebo test.” In Supplemental Table 3, we confirmed that the effects of birth order and gender are separate by including interactions in both with childhood illness.

National Vaccination Rates and Gender Gap in Schooling

We examined national estimates of the schooling gender gap and their association with national vaccination coverage rates in the 38 countries. The schooling gap for each survey was defined as the weighted mean of the schooling gap in each survey for 0, 1, and >1 illness. We then obtained vaccination rates for the same country-year survey, defined as receipt of the complete series of polio vaccinations; diphtheria, tetanus, and pertussis vaccinations; and measles-containing vaccinations, for a total of 8 vaccinations.16 We also conducted subanalyses of the relationship between vaccination coverage and the schooling gender gap disaggregated by income group and by vaccine type. These findings are presented in Supplemental Fig 6.

The sponsor of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The corresponding author had full access to all the data in the study and had final responsibility for the decision to submit for publication. The study was determined as exempt by the Stanford Institutional Review Board.

Our data consisted of 120 708 adolescents residing in 41 821 households in 38 countries. Data were obtained from 47 surveys conducted between 1999 and 2013. The sample included all households, women, adolescents, and children that met our inclusion criteria. By using the criteria that the household have at least 1 child <5 years old and at least 2 older children (at least 1 boy and 1 girl), we limited the sample among the universe of DHS to slightly over 116 000 households in 63 countries. Once we required that the survey include information on schooling and gender and illness episodes, the number of households available for analysis was ∼42 000 (or 36% of all potential households in the DHS). We provide in Table 2 an assessment of the number of households with incomplete information, conditional on the country being in our sample. By using this measure, excluded households generally account for <1% of the within-country sample. As shown in Supplemental Table 4, there are no differences in observable characteristics between country samples with and without these data fields. We present in Table 1 the distribution of illness episodes and selected social characteristics. Approximately 50% of the included households reported no episodes of illness, whereas 39% reported 1 episode and 11.5% reported 2 episodes during the 2-week period preceding the survey. There were also significant differences in social variables (including car and telephone ownership and indicators of maternal education) between the households reporting 0, 1, or 2 illness episodes. We present in Table 2 the adolescent school attendance data, illness episodes, and vaccination coverage for each country included in the analyses. During the study period, 67.1% of all adolescent boys and 61.2% of all adolescent girls were in school. By country income group, the gender gap in adolescent education was 8.4% in low-income countries, 3.6% in lower-middle–income countries, and 2.0% in upper-middle–income countries. Over time, the gap decreased from 6.7% between 1999 and 2005 to 4.2% between 2006 and 2013 (however, the set of countries analyzed differed between the early and later time periods).

TABLE 2

Survey Characteristics

CountryYearGirlsBoysHouseholdsVaccinatedCompleteness
NIn School (%)NIn School (%)NIll <5 Yr Old Children(%)(%)
0 (%)1 (%)2+ (%)
Armenia 2000 69 0.87 65 0.846 55 0.709 0.291 71.4 100 
Benin 2006 1830 0.544 2003 0.699 1322 0.577 0.351 0.072 47.1 99.4 
Bolivia 2003 1367 0.5 1375 0.517 1077 0.521 0.391 0.088 50.4 99.7 
Bolivia 2008 1019 0.839 1061 0.88 829 0.559 0.375 0.066 76.7 100 
Burkina Faso 2003 2155 0.227 2431 0.3 1441 0.42 0.477 0.103 43.9 99.7 
Cambodia 2000 1306 0.501 1287 0.656 1010 0.598 0.353 0.049 39.9 99.7 
Cameroon 2004 1394 0.742 1435 0.842 938 0.452 0.431 0.117 48.2 99.5 
Chad 2004 774 0.469 851 0.644 554 0.433 0.457 0.11 11.3 98.7 
Colombia 2000 1683 0.769 1717 0.738 1319 0.556 0.377 0.067 62.4 99.9 
Colombia 2004 407 0.73 411 0.723 322 0.584 0.376 0.04 63.9 99.8 
Colombia 2009 1924 0.836 1899 0.83 1511 0.541 0.396 0.062 67.7 99 
Congo 2005 820 0.843 858 0.88 593 0.558 0.39 0.052 52.1 99.7 
Dominican Republic 2002 1081 0.907 1119 0.884 869 0.631 0.299 0.07 34.9 99.3 
Dominican Republic 2007 1060 0.885 1079 0.87 863 0.672 0.294 0.034 58 97.5 
Egypt 2000 1906 0.633 1921 0.79 1389 0.665 0.289 0.047 92.2 100 
Ethiopia 2000 1367 0.342 1404 0.501 1094 0.536 0.374 0.09 14.3 100 
Ghana 2003 485 0.647 532 0.714 384 0.625 0.336 0.039 69.4 99.9 
Guinea 2005 981 0.499 1076 0.667 697 0.534 0.396 0.07 37.2 99.5 
Haiti 2000 916 0.728 951 0.766 706 0.367 0.459 0.174 33.5 97 
Honduras 2011 1906 0.491 1965 0.43 1438 0.531 0.392 0.076 84.5 99.8 
Kenya 2003 738 0.767 785 0.857 570 0.43 0.446 0.125 51.8 99.5 
Lesotho 2004 750 0.829 742 0.686 599 0.646 0.326 0.028 67.8 99.7 
Liberia 2013 886 0.885 929 0.919 701 0.494 0.427 0.08 54.8 93.8 
Madagascar 2003 607 0.601 641 0.647 468 0.675 0.278 0.047 52.9 99.6 
Malawi 2000 1486 0.812 1536 0.838 1174 0.386 0.51 0.104 70.1 99.7 
Malawi 2004 833 0.794 843 0.826 652 0.403 0.491 0.106 64.4 99.8 
Mali 2001 4603 0.288 4683 0.403 986 0.124 0.489 0.387 28.7 99.5 
Mali 2006 1485 0.415 1538 0.521 1045 0.654 0.271 0.076 48.2 99.3 
Morocco 2003 1109 0.484 1125 0.636 841 0.653 0.303 0.044 89.1 99.9 
Mozambique 2003 1364 0.718 1392 0.802 1017 0.559 0.383 0.057 63.3 99.7 
Namibia 2000 710 0.872 709 0.869 498 0.592 0.355 0.052 64.8 99 
Nepal 2001 848 0.459 798 0.709 646 0.455 0.454 0.091 65.6 100 
Nicaragua 2001 1440 0.605 1455 0.553 1027 0.579 0.333 0.088 70.1 99.8 
Niger 2006 1325 0.362 1393 0.462 922 0.47 0.407 0.124 29 98.7 
Nigeria 2003 687 0.683 751 0.724 517 0.47 0.42 0.11 12.9 99.8 
Peru 2000 1939 0.793 1974 0.849 1565 0.546 0.38 0.074 64 99.8 
Peru 2003 2207 0.838 2260 0.86 1819 0.549 0.409 0.042 64.7 99.8 
Philippines 2003 980 0.835 999 0.74 756 0.591 0.335 0.074 69.8 99.8 
Rwanda 2000 1207 0.308 1189 0.304 907 0.526 0.385 0.089 76 99.7 
Rwanda 2005 1015 0.718 1030 0.75 818 0.601 0.308 0.09 75.2 99.6 
Senegal 2005 3052 0.48 3170 0.54 1799 0.403 0.473 0.124 58.7 99.4 
Tanzania 2004 1282 0.708 1290 0.745 952 0.575 0.334 0.091 71.1 99.7 
Tanzania 2009 1263 0.714 1240 0.76 933 0.663 0.284 0.053 75.2 99.3 
Turkey 2003 557 0.442 545 0.772 396 0.389 0.46 0.152 45.7 99.9 
Uganda 2000 978 0.849 998 0.88 689 0.447 0.389 0.164 36.7 99.6 
Zambia 2001 951 0.685 1003 0.763 716 0.397 0.486 0.117 70 99.4 
Zimbabwe 1999 755 0.783 743 0.816 535 0.609 0.338 0.052 64 99.9 
CountryYearGirlsBoysHouseholdsVaccinatedCompleteness
NIn School (%)NIn School (%)NIll <5 Yr Old Children(%)(%)
0 (%)1 (%)2+ (%)
Armenia 2000 69 0.87 65 0.846 55 0.709 0.291 71.4 100 
Benin 2006 1830 0.544 2003 0.699 1322 0.577 0.351 0.072 47.1 99.4 
Bolivia 2003 1367 0.5 1375 0.517 1077 0.521 0.391 0.088 50.4 99.7 
Bolivia 2008 1019 0.839 1061 0.88 829 0.559 0.375 0.066 76.7 100 
Burkina Faso 2003 2155 0.227 2431 0.3 1441 0.42 0.477 0.103 43.9 99.7 
Cambodia 2000 1306 0.501 1287 0.656 1010 0.598 0.353 0.049 39.9 99.7 
Cameroon 2004 1394 0.742 1435 0.842 938 0.452 0.431 0.117 48.2 99.5 
Chad 2004 774 0.469 851 0.644 554 0.433 0.457 0.11 11.3 98.7 
Colombia 2000 1683 0.769 1717 0.738 1319 0.556 0.377 0.067 62.4 99.9 
Colombia 2004 407 0.73 411 0.723 322 0.584 0.376 0.04 63.9 99.8 
Colombia 2009 1924 0.836 1899 0.83 1511 0.541 0.396 0.062 67.7 99 
Congo 2005 820 0.843 858 0.88 593 0.558 0.39 0.052 52.1 99.7 
Dominican Republic 2002 1081 0.907 1119 0.884 869 0.631 0.299 0.07 34.9 99.3 
Dominican Republic 2007 1060 0.885 1079 0.87 863 0.672 0.294 0.034 58 97.5 
Egypt 2000 1906 0.633 1921 0.79 1389 0.665 0.289 0.047 92.2 100 
Ethiopia 2000 1367 0.342 1404 0.501 1094 0.536 0.374 0.09 14.3 100 
Ghana 2003 485 0.647 532 0.714 384 0.625 0.336 0.039 69.4 99.9 
Guinea 2005 981 0.499 1076 0.667 697 0.534 0.396 0.07 37.2 99.5 
Haiti 2000 916 0.728 951 0.766 706 0.367 0.459 0.174 33.5 97 
Honduras 2011 1906 0.491 1965 0.43 1438 0.531 0.392 0.076 84.5 99.8 
Kenya 2003 738 0.767 785 0.857 570 0.43 0.446 0.125 51.8 99.5 
Lesotho 2004 750 0.829 742 0.686 599 0.646 0.326 0.028 67.8 99.7 
Liberia 2013 886 0.885 929 0.919 701 0.494 0.427 0.08 54.8 93.8 
Madagascar 2003 607 0.601 641 0.647 468 0.675 0.278 0.047 52.9 99.6 
Malawi 2000 1486 0.812 1536 0.838 1174 0.386 0.51 0.104 70.1 99.7 
Malawi 2004 833 0.794 843 0.826 652 0.403 0.491 0.106 64.4 99.8 
Mali 2001 4603 0.288 4683 0.403 986 0.124 0.489 0.387 28.7 99.5 
Mali 2006 1485 0.415 1538 0.521 1045 0.654 0.271 0.076 48.2 99.3 
Morocco 2003 1109 0.484 1125 0.636 841 0.653 0.303 0.044 89.1 99.9 
Mozambique 2003 1364 0.718 1392 0.802 1017 0.559 0.383 0.057 63.3 99.7 
Namibia 2000 710 0.872 709 0.869 498 0.592 0.355 0.052 64.8 99 
Nepal 2001 848 0.459 798 0.709 646 0.455 0.454 0.091 65.6 100 
Nicaragua 2001 1440 0.605 1455 0.553 1027 0.579 0.333 0.088 70.1 99.8 
Niger 2006 1325 0.362 1393 0.462 922 0.47 0.407 0.124 29 98.7 
Nigeria 2003 687 0.683 751 0.724 517 0.47 0.42 0.11 12.9 99.8 
Peru 2000 1939 0.793 1974 0.849 1565 0.546 0.38 0.074 64 99.8 
Peru 2003 2207 0.838 2260 0.86 1819 0.549 0.409 0.042 64.7 99.8 
Philippines 2003 980 0.835 999 0.74 756 0.591 0.335 0.074 69.8 99.8 
Rwanda 2000 1207 0.308 1189 0.304 907 0.526 0.385 0.089 76 99.7 
Rwanda 2005 1015 0.718 1030 0.75 818 0.601 0.308 0.09 75.2 99.6 
Senegal 2005 3052 0.48 3170 0.54 1799 0.403 0.473 0.124 58.7 99.4 
Tanzania 2004 1282 0.708 1290 0.745 952 0.575 0.334 0.091 71.1 99.7 
Tanzania 2009 1263 0.714 1240 0.76 933 0.663 0.284 0.053 75.2 99.3 
Turkey 2003 557 0.442 545 0.772 396 0.389 0.46 0.152 45.7 99.9 
Uganda 2000 978 0.849 998 0.88 689 0.447 0.389 0.164 36.7 99.6 
Zambia 2001 951 0.685 1003 0.763 716 0.397 0.486 0.117 70 99.4 
Zimbabwe 1999 755 0.783 743 0.816 535 0.609 0.338 0.052 64 99.9 

We find a significant gender gap in education between adolescent boys and girls when there are sick children at home (Fig 2). When no young children are ill, adolescent girls are on average 5.08% less likely to attend school than adolescent boys within the same household (95% confidence interval [CI], 5.50%–4.65%). This gap increases in magnitude to 7.77% (95% CI, 7.30%–8.24%) if the household reports 1 illness episode among children <5 years old and 8.53% (95% CI, 9.32%–7.74%) if there are ≥2 illness episodes.

FIGURE 2

The gender gap in education of older siblings as a function of illness episodes of children <5 years old in the household. The average marginal effect (eg, percentage point reduction in school attendance for girls versus boys) and a 95% CI are shown on the graph. The leftmost column depicts and reports the effect of a gender gap in education if there were no children <5 years old with illness episodes in the last 2 weeks reported by the household. The middle column depicts the same for households that have 1 illness episode in the preceding 2 weeks, and the rightmost column presents the estimate for those households with ≥2 illness episodes.

FIGURE 2

The gender gap in education of older siblings as a function of illness episodes of children <5 years old in the household. The average marginal effect (eg, percentage point reduction in school attendance for girls versus boys) and a 95% CI are shown on the graph. The leftmost column depicts and reports the effect of a gender gap in education if there were no children <5 years old with illness episodes in the last 2 weeks reported by the household. The middle column depicts the same for households that have 1 illness episode in the preceding 2 weeks, and the rightmost column presents the estimate for those households with ≥2 illness episodes.

In households where mothers are working, girls are 6.1% less likely to attend school than older boys within the same household (95% CI, 6.69%–5.60%); Fig 3). This gap increases in magnitude to 8.7% (95% CI, 9.27%–8.08%) if the household reports 1 illness episode and 10.06% (95% CI, 11.06%–9.05%) if there are ≥2 illness episodes. Comparable estimates for households in which the mother does not work outside the home are 3.13% (95% CI, 3.84%–2.42%), 6.21% (95% CI, 6.98%–5.44%), and 6.03% (95% CI, 7.30%–4.76%), respectively.

FIGURE 3

The gender gap in education of older siblings as a function of illness episodes of children <5 years old in the household by maternal labor force participation. The figure depicts the gender gap in education as a function of children <5 years old with illness. The dark bars represent the results of the average marginal effect for households whose mothers work outside the home, and the lighter bars represent the effect for mothers who do not work outside the home. The leftmost column depicts and reports the effect of a gender gap in education if there were no children <5 years old with illness episodes in the last 2 weeks reported by the household. The middle column depicts the same for households that have 1 illness episode in the preceding 2 weeks, and the rightmost column presents the estimate for those households with ≥2 illness episodes.

FIGURE 3

The gender gap in education of older siblings as a function of illness episodes of children <5 years old in the household by maternal labor force participation. The figure depicts the gender gap in education as a function of children <5 years old with illness. The dark bars represent the results of the average marginal effect for households whose mothers work outside the home, and the lighter bars represent the effect for mothers who do not work outside the home. The leftmost column depicts and reports the effect of a gender gap in education if there were no children <5 years old with illness episodes in the last 2 weeks reported by the household. The middle column depicts the same for households that have 1 illness episode in the preceding 2 weeks, and the rightmost column presents the estimate for those households with ≥2 illness episodes.

In our analysis of vaccination rates relative to the education gap, we found a statistically significant and strong negative correlation between the vaccination rates of children <5 years old and the gender gap in education (eg, the higher the vaccination rate, the smaller the gender gap in education; correlation coefficient = 0.34, P = .02; Fig 4). The adolescent gender gap in education approaches zero with coverage rates exceeding ∼70% for all 8 vaccines. We performed several specification checks to ensure that our results are robust. These include varying the age of the included older children, using alternative educational outcomes, and splitting the sample by rural and urban location. We describe these supplementary changes in detail. In Supplemental Fig 5, we varied the age thresholds for older children in the following different ways: 10 to 16, 10 to 17, 10 to 18, 11 to 16, 12 to 16, 12 to 17, and 12 to 18. Our results are not sensitive to varying the thresholds.

FIGURE 4

Country-level gender gap and vaccination rates. The figure demonstrates the correlation between the gender gap in adolescent schooling in response to households with children <5 years old with illness at the country-level and the receipt of all 8 basic vaccinations. The adolescent education gender gap is derived by using a fixed-effects estimator as described in the text.

FIGURE 4

Country-level gender gap and vaccination rates. The figure demonstrates the correlation between the gender gap in adolescent schooling in response to households with children <5 years old with illness at the country-level and the receipt of all 8 basic vaccinations. The adolescent education gender gap is derived by using a fixed-effects estimator as described in the text.

The association between maternal work location and the gender gap related to illness episodes is driven by the low-income countries in our sample (correlation coefficient = 0.48, P = .02), which are far below middle-income countries in average vaccination coverage; in middle-income countries, the gender gap is narrow irrespective of the vaccination rate, with vaccination rates unrelated to the gender gap in education (correlation coefficient = 0.10, P = .66; see Supplemental Fig 6 for detailed presentation of this analysis). In addition, the gradients in illness episodes and educational gender gaps were comparable in urban and rural settings (see Supplemental Fig 7 for detailed presentation of this analysis). We also found that in households that had only older boy children (11–17 years old), there was no educational gap between older and younger boy children (Supplemental Fig 9). Finally, we explored the role of birth order in influencing the gender gap findings. By interacting the gender of the child, the number of illnesses, and an indicator for being the oldest sibling, we found that the oldest sibling is less likely to be in school when compared with younger siblings, and this gap is greater if the older siblings are girls. There is no gradient with increasing illnesses for boys irrespective of birth order (Supplemental Table 3). Our analyses that used alternative educational variables, such as whether the child attended any school in the past year or the number of years of education, did not alter the findings (see Supplemental Fig 8 for detailed presentation of this analysis).

Our findings in this study suggest that adolescent girls’ schooling is sensitive to the health of younger children in the household. Adolescent girls were found to be less likely to attend school than adolescent boys in the same household, and that this gender gap was exacerbated by the domestic demands associated with illness among young children in the household. However, the impact of young child illness on school attendance was not observed in households with only adolescent boys. These effects were substantial and suggest that the gender gap in adolescent school attendance increased by >50% when young children in the household became ill. The estimates are even larger when mothers worked outside the home. These results were generally supported by our findings that education gender gaps were smaller in countries with relatively high vaccination rates, which may be associated with lower young child illness. This result was strongest in low-income countries and is consistent with the suggestion that in such settings, the opportunities associated with the education of girls are likely to compete with gender-specific, domestic roles of adolescent girls.

A strength of our analysis is that we examined within-household differences in education attendance in the presence of young child illness, which mitigates concerns associated with the nonrandom allocation of child illness across households. Any analysis relating childhood illnesses to education patterns must account for the likelihood that the household burden of childhood disease will be related to how the education of adolescent girls is valued or the material resources available to support it. The scale and comprehensive nature of the DHS permitted the analysis of gender gaps within households, an approach that diminishes the potential for confounding by unobservable household factors.

Our findings emphasize the complex determinants of the gender gap in adolescent school attendance.8,13 That episodes of young child illness in the household can affect adolescent girl education participation suggest that policies directed at enhancing girl’s education should address not only the availability of educational opportunities but also the prevalence of competing domestic tasks that fall more heavily on girls. Policies that strengthen family and community support for challenges such as sick child care will likely prove essential, particularly as women move increasingly into the work force outside the home.

There are several limitations to our analysis. The household-level analysis used only illnesses reported over the 2 weeks preceding the survey. This focus on this recent period reduces recall bias but may lead to an underestimate of the overall effect of illness on the adolescent gender gap in education. In addition, our sample necessarily includes households with at least 3 children, which may not fully reflect the experiences of households with fewer children, although fertility rates are generally >3 in low-income countries.18 Although the models employed in this study remove many confounding factors, there may remain unobserved variables that are related to adverse child health events that interact with older sibling gender. However, similar results have been replicated by using natural experiments that impact child health, such as national immunization campaigns.11 Researchers for future studies should attempt to disentangle when girls are the most vulnerable and what potential policies are most effective at assisting parents in educating their daughters. These policies may vary by local context depending, for example, on local labor markets and educational infrastructure. In addition, the country-level regressions associating the gender gap in education with vaccination rates should be interpreted cautiously because they cannot document causality; other social determinants across countries could be driving both higher vaccination rates and a lower children <5 years old illness-associated gender gap in education.

Our analysis suggests that gender-based household tasks represent an important and underrecognized driver of the gap in schooling between adolescent boys and girls. This finding was based on the analysis of comprehensive surveys conducted in a variety of low- and middle-income settings and is consistent with observational studies demonstrating that girls are preferentially tasked to care for young children.11,12 Given the long-term benefits of secondary schooling for women’s health and economic outcomes, our study underscores the potential societal returns of policies that improve child health and support the ability of low-income families to meet the dual challenges of early child-rearing and sustained, high-quality adolescent education for girls in complex and highly dynamic societies.

     
  • SDG

    sustainable development goal

  •  
  • LMIC

    low- and middle-income countries

  •  
  • DHS

    Demographic and Health Surveys

  •  
  • CI

    confidence interval

Drs Alsan and Bendavid conceived the work and conducted the analysis along with Ms Xing; Dr Darmstadt contributed to the conceptualization of factors impacting gender gaps and contributed to the manuscript preparation; Dr Wise helped with manuscript preparation and improved the clarity of the presentation; and all authors approved the final manuscript as submitted.

FUNDING: This work was supported in part by grant funding from the US National Institutes of Health Eunice Kennedy Shriver National Institute of Child Health and Human Development (1K01HD084709-01; PI Alsan) as well as from the Doris Duke Charitable Foundation (Bendavid) and Centre on the Demography and Economics of Health and Aging (Bendavid). Funded by the National Institutes of Health (NIH).

COMPANION PAPER: A companion to this article can be found online at www.pediatrics.org/cgi/doi/10.1542/peds.2017-1322.

We thank Sarah Henry at Stanford University for her critical review of a draft of the manuscript.

1
United Nations
. Sustainable development goals - goal 5: achieve gender equality and empower all women and girls. 17 goals to transform our world. Available at: www.un.org/sustainabledevelopment/gender-equality/#. Accessed February 3, 2016
2
Gakidou
E
,
Cowling
K
,
Lozano
R
,
Murray
CJ
.
Increased educational attainment and its effect on child mortality in 175 countries between 1970 and 2009: a systematic analysis.
Lancet
.
2010
;
376
(
9745
):
959
974
[PubMed]
3
ICF_International
. DHS stat compiler. Available at: www.statcompiler.com/. Accessed 2016
4
UNESCO
. Girls’ education - the facts. Available at: https://en.unesco.org/gem-report/sites/gem-report/files/girls-factsheet-en.pdf. Accessed 2013
5
The World Bank
. Girls’ education. Available at: www.worldbank.org/en/topic/education/brief/girls-education. Accessed 2014
6
Alsan
MM
,
Cutler
DM
.
Girls’ education and HIV risk: evidence from Uganda.
J Health Econ
.
2013
;
32
(
5
):
863
872
[PubMed]
7
De Neve
J-W
,
Fink
G
,
Subramanian
SV
,
Moyo
S
,
Bor
J
.
Secondary education and HIV infection in Botswana.
Lancet Glob Health
.
2016
;
4
(
1
):
e23
[PubMed]
8
MacArthur Foundation
. Girls’ secondary education in developing countries. Available at: https://www.macfound.org/programs/girlseducation/strategy/. Accessed February 3, 2016
9
The Economist
. Women and work - the power of parity. Available at: www.economist.com/news/finance-and-economics/21667949-world-would-be-much-richer-place-if-more-women-had-paying-jobs-power. Accessed February 2, 2016
10
Dobbs
R
,
Manyika
J
,
Chui
JWM
,
Lund
S
. The power of parity: how advancing women’s equality can add $12 trillion to global growth. Available at: https://www.mckinsey.com/global-themes/employment-and-growth/the-power-of-parity-advancing-womens-equality-in-the-united-states
11
Alsan
M.
The gender impact of young children’s health on human capital: evidence from Turkey. Available at: https://healthpolicy.fsi.stanford.edu/sites/default/files/tnic_2.pdf. Accessed 2016
12
Pitt
M
,
Rosenzweig
M
.
Estimating the intrahousehold incidence of illness: child health and gender-inequality in the allocation of time.
Int Econ Rev (Philadelphia)
.
1990
;
31
(
4
):
969
989
13
Lloyd
CB
,
Grant
M
,
Ritchie
A
.
Gender differences in time use among adolescents in developing countries: implications of rising school enrollment rates.
J Res Adolesc
.
2008
;
18
(
1
):
99
120
14
Kosek
M
,
Bern
C
,
Guerrant
RL
.
The global burden of diarrhoeal disease, as estimated from studies published between 1992 and 2000.
Bull World Health Organ
.
2003
;
81
(
3
):
197
204
15
ICF_International
. Demographic and Health Surveys (various). Available at: http://dhsprogram.com/data/. Accessed 2016
16
ICF_International
. Demographic and Health Surveys (various). http://dhsprogram.com/data/. Accessed April 16, 2016
17
WHO
. Maternal newborn child and adolescent health. Available at: www.who.int/maternal_child_adolescent/topics/adolescence/dev/en/. Accessed 2017
18
WHO
. Global health observatory data repository. Available at: apps.who.int. Accessed 2016

Competing Interests

POTENTIAL CONFLICT OF INTEREST: The authors have indicated they have no potential conflicts of interest to disclose.

FINANCIAL DISCLOSURE: The authors have indicated they have no financial relationships relevant to this article to disclose.

Supplementary data