2014/ED/EFA/MRT/PI/14 Background paper prepared for the Education for All Global Monitoring Report 2013/4 Teaching and learning: Achieving quality for all Educational Attainment and Maternal Mortality Sonia Bhalotra with Damian Clarke 2013 This paper was commissioned by the Education for All Global Monitoring Report as background information to assist in drafting the 2013/4 report. It has not been edited by the team. The views and opinions expressed in this paper are those of the author(s) and should not be attributed to the EFA Global Monitoring Report or to UNESCO. The papers can be cited with the following reference: “Paper commissioned for the EFA Global Monitoring Report 2013/4, Teaching and learning: Achieving quality for all” For further information, please contact [email protected] Attainment and Maternal Mortality Sonia Bhalotra∗ Damian Clarke† September 1, 2013 Abstract We examine whether a causal relationship exists between maternal education and ma- ternal mortality. Despite considerable evidence in favour of a causal relationship between education and a range of other health behaviours and outcomes, and a significant gradi- ent between maternal mortality and education across time and countries, no comprehensive study exists to examine whether this relationship is—at least in part—causal. By forming a large panel of data consisting of 108 countries over 20 years, and by examining three natural experiments resulting in plausibly exogenous expansions in education, we present consider- able evidence that increases in maternal education causally reduce the likelihood of dying in child birth. The size of this relationship is considerable. Our preferred estimates suggest that a country moving from 0 to 1 years of education will reduce maternal mortality by 174 deaths per 100,000 births, while moving from 7 to 8 years results in a smaller, but still significant, 15 deaths per 100,000 births. 1 Introduction Life expectancy exhibits a consistently positive gradient in education that, unlike income gra- dients, tends not to diminish as education levels rise and is evident across and within rich and poor countries (Richards and Barry, 1998; Kunst and Mackenbach, 1994; Elo and Preston, 1996; Lleras-Muney, 2005; Meara et al., 2008). Maternal mortality is an important determinant of life ∗[email protected] †[email protected] 1expectancy for women in developing countries but there is not much evidence of the relationship between maternal mortality and education. This paper attempts to fill that gap. Every day, approximately 800 women die from preventable causes related to pregnancy and childbirth, 99% of all maternal deaths occur in developing countries and they are largely pre- ventable (WHO, 2012). The maternal mortality ratio in developing countries is 240 per 100,000 births versus 16 per 100,000 in developed countries. There are large disparities between and within countries, with a few countries having extremely high maternal mortality ratios of 1,000 or more per 100,000 live births; more than half of all maternal deaths occur in sub-Saharan Africa and almost one third occur in South Asia (WHO, 2012). The WHO identifies poverty, limited access to public services, cultural practices and lack of information as the main causes or sources of variation in maternal mortality ratios. Between 1990 and 2010, maternal mortality worldwide dropped by almost 50%. While this is impressive, certainly relative to trends in the preceding twenty years, the average rate of decline in the global maternal mortality ratio (the number of maternal deaths per 100,000 live births) of 3.1% per year over this period falls well below the annual decline of 5.5% required to achieve the Millennium Development Goals (MDG) for maternal health adopted by the international community in 2000. There are signs of increasing policy commitment to addressing maternal mortality: official development assistance for maternal and new-born health has risen relative, for instance, to funding for child health (Greco et al., 2008).1 However policy documents on maternal mortality seldom indicate education as a cause and the academic literature has very little to say about the relationship between maternal mortality and education. This said, there is a lively literature in economics that documents a positive corre- lation between education and other indicators of health. One possibility, relatively uninteresting for policy, is that this correlation is driven by time preference, people with lower discount rates tending to invest more in both forms of human capital. While this is likely so, the crop of studies that identify causal effects of education on health most often tend to associate this with edu- cated people engaging in healthier behaviours, for instance, smoking and drinking less and being more likely to seek prenatal care, adopt new drugs or adhere to treatment for disease (Goldman and Smith, 2010; Goldman and Lakdawalla, 2001; Currie and Moretti, 2003; Lichtenberg and Lleras-Muney, 2005; Cutler and Lleras-Muney, 2010; Glied and Lleras-Muney, 2008; Jensen and Lleras-Muney, 2012). This is consistent with the efficacy that education confers in acquiring and 1Overall donor disbursements increased from US$2,119 million in 2003 to $3,482 million in 2006; funding for child health increased by 63% and that for maternal and new-born health increased by 66%. In the 68 priority countries, child-related disbursements increased from a mean of $4 per child in 2003 to $7 per child in 2006; disbursements for maternal and neonatal health increased from $7 per live birth in 2003 to $12 per live birth in 2006. 2processing relevant information (Rosenzweig, 1995; Rosenzweig and Schultz, 1989; Cutler and Lleras-Muney, 2010). Education may also influence health through its influence on income and, thereby, the facility to purchase health inputs but the impacts of education on health tend to be larger than direct impacts of income, to hold conditional upon income, and to be similar for men and women, at least in richer countries (Lleras-Muney, 2005; Cutler and Lleras-Muney, 2010). The proximate causes of maternal mortality are pregnancy complications such as pre- eclamp- sia, bleeding, infections and unsafe abortion. To the extent that educated women are more likely to adopt simple and low-cost practices to maintain hygiene, are more able to react to symptoms such as bleeding or high blood pressure, more likely to assess the information on abortion and place of abortion, and more willing to accept treatment and birth attendance, education may plausibly cause declines in maternal mortality. Moreover, if these are the mechanisms by which it acts then education may have an impact on maternal mortality conditional upon the state of health services. The payoff to education tends to be higher in circumstances where learning is important, for instance, circumstances of structural and technological change (Foster and Rosen- zweig, 2004). Such change is characteristic of developing countries which, in the post-war period, have experienced massive technological change in the public health sphere (Cutler et al., 2006) as well as changes in the economic (and occupational) structure. Using panel data for 108 developing countries and new estimates of maternal mortality ratios for 1990-2010, we estimate the relationship between maternal mortality and educational attainment under the standard panel data identifying assumptions, unconditional and conditional upon indicators of income, public health provision and fertility. We find large and consistently significant impacts of education. Indeed, depending upon the specification, the (standardized) impacts of education on maternal mortality are equal to or larger than the impacts of skilled birth attendance, an instrument that the WHO currently focuses upon in devising programs to address maternal mortality. In contrast, within-country variation in GDP has no signigicant impact on maternal mortality conditional upon education. We supplement the cross-country analysis with case studies from Nigeria, Zimbabwe and Kenya. These are three countries that implemented substantial education reforms. We use cohort and region variation in education generated by the reform, that we argue is plausibly exogenous. To match this to maternal mortality rates, we construct cohort mortality rates for regions within these countries using the Demographic and Health Surveys. In two of the three case we identify a sizeable and negative impact of improvements in education on maternal mortality. In both Nigeria and Zimbabwe we find that the educational expansions due to policy experiments significantly reduce maternal death in affected cohorts, and find that these results are surprisingly close to those estimated in cross-country regressions. In Kenya—which experienced a schooling extension at the lower high school level—we do not find evidence of such a reduction, estimating that the effect on maternal 3mortality is not significantly different to zero. All in all, this paper generates new data on maternal mortality that is available for individuals within regions and states of countries, and it provides what would appear to be the first systematic consideration of the relationship between maternal mortality ratios and education. It opens up several avenues for further research. 2 Literature The medical literature suggests that maternal mortality is principally associated with labour, de- livery and the early postpartum period; largely occuring between the third trimester of pregnancy and the first post-partum week. The principal medical cause of these deaths is obstetric haem- orrhage, a largely preventable cause provided “access to timely and competent obstetric care” is available (Ronsmans and Graham, 2006). Along with haemorrhage, a number of other causes are reported, including sepsis, abortion, hypertensive disorders, obstructed labour, embolism and ectopic pregnancy2. Despite the fact that these complications account for the majority of deaths during labour, birth and the post-partum period, the prevalence of each varies by region (Khan et al., 2006). Whilst haemorrhage is the principal cause in developing regions, complications from caesarian sections and anaesthesia are more common in developed countries, and other region-specific disease burdens are noted (abortion in Latin America and the Caribbean, sepsis and HIV in Africa and anaemia in Asia). Much focus in the existing literature is on the difficulty of obtaining accurate statistics of maternal mortality, particularly in those areas where maternal mortality is highest(Ronsmans and Graham, 2006; McCarthy and Maine, 1992; McAlister and Baskett, 2006). The ability to compile credible data is confounded by the lack of consistent classification prior to the introduction of ICD-9 coding, the need to collect data from an array of data sources, misclassification of maternal death, significant under-reporting of maternal mortality, and missing and incomplete data(Yazbeck, 2012; Hogan et al., 2010). Early work3 in the medical literature highlighted the importance of addressing the ‘proximal causes’ or inputs of maternal mortality; namely the likelihood that a women becomes pregnant, 2Highly cited figures (AbouZahr et al., 1991) suggest figures of 25% due to haemorrhage, 20% due to indirect causes, 15% due to infection, 13% due to abortions, 12% due to eclampsia, 8% to obstructed labour and 8% to other direct causes. Khan et al. (2006) provide evidence broadly in agreement with these values however note that these estimates suffer from large confidence intervals. 3The Safe Motherhood Initiative of 1987 was put in place in response to the increasing recognition of this topic in developing countries. 4or the likelihood of complications (and the treatment of these complications) conditional upon her becoming pregnant (McCarthy and Maine, 1992; Goodburn and Campbell, 2001; Trussell and Pebley, 1984). Along with attendence and medical care for mothers during pregnancy and child birth, these studies suggest that altering maternal age, quantity of births, and spacing of births could have a considerable effect on rates of maternal mortality. More recent medical studies also suggest that socioeconomic factors are fundamental in the reduction of maternal mortality. Costello et al. (2004), for example, argue that the focus on primary care and skilled attendance is not a sufficient strategy, citing considerable reductions in maternal mortality fol- lowing community-based interventions4. McAlister and Baskett (2006) suggest that social—and specifically gender-specific factors such as female education—should predict maternal mortality, demonstrating cross-country correlation between these variables. Recent work from Ahmed et al. (2012) suggests that family planning inputs would allow for considerable reductions in maternal deaths world-wide, with the authors estimating that the equivalent of 44% of maternal deaths in 2008 were averted due to contraceptive use, and that meeting unmet demand for family planning could prevent a further 29%. 3 Methodology 3.1 Cross-Country Analysis In order to examine the effect of educational expansions on maternal mortality, we run country- level regressions of the following form: MMRit = αi + educitβ + Witγ + δt + εit, (1) where MMRit refers to the maternal mortality rate in country i and period t, educit to a series of variables measuring the percent of the population with a given level of education, and Wit refers to a vector of relevant country- and time-varying covariates. Given that we have both tem- poral and regional variation in the variables of interest, we employ a fixed-effects methodology which allows for us to control for all time-fixed country-specific factors which may confound the analysis in (1). The ability to eliminate these time invariant factors is particularly important in this analysis, given that many unobserved factors, such as the institutional and political aspects of a given country, are likely to be correlated with both maternal mortality and educational at- tainment. Whilst some of these variables are not strictly “time-invariant”, they evolve sufficiently 4Also see Manandhar et al. (2004) for the results of a randomized trial of community-based meetings. 5slowly that our analysis over a comparatively short time horizon will mean that a fixed-effect estimator will largely control for them. We cluster standard errors at the level of the country (allowing for the stochastic term uit to be correlated over time for each i.) We are able to estimate (1) for a number of different subgroups. Whilst we primarily focus on women of reproductive age, our data allows us to run a similar analysis for educational attainment of both gender groups and for various age subgroups created within the fertile age range. We also segregate countries by region- and income-level, and present group-specific estimates of the above specification. A full list of the educational variables which we use and a description of the time varying covariates Wit is provided in the data section of this document. 3.2 Natural Experiments In addition to (1) which we estimate with panel data, we focus on a number of country-specific cases in countries which have experienced isolated, policy-defined expansions in schooling achieve- ment. Such expansions allow for us to causally identify the effect of education on maternal mortality if we believe that we can identify similar subgroups of individuals who are differen- tially affected by these isolated shocks to education. In these country case studies, we construct comparable subgroups of women who entered school on either side of an important educational reform. We exploit variation in the intensity of treatment between regions and cohorts of birth to estimate the effect of educational expansions on maternal mortality. Prominent educational reforms discussed in the economics literature are listed below in table 1. Of the reforms listed here, we identify three which are suitable to use as natural experiments to identify the effect of education on maternal mortality. There are Nigeria, Zimbabwe, and Kenya. Of the other three cases listed, they were either not examined as they occurred too recently to observe the outcome of interest (maternal mortality) or because microdata for maternal mortality outcomes was not available for the country in question. For each of the three natural experiments examined, we briefly describe the nature of the reform, and the identification strategy used in estimation. 3.2.1 Nigeria 1976 The Universal Primary Education (UPE) programme was initiated in Nigeria in September 1976 as a national programme directed at primary school enrolment. This programme involved the 6provision of free primary school education, along with significant classroom construction (Osili and Long, 2008). The gross primary enrolment rate doubled between 1974 and 1981 for male students, and nearly tripled for females. Osili and Long (2008) suggest using a double-differences identification strategy in which women from cohorts born between 1970 and 1975 in high-intensity states5 act as the treatment group. As per Osili and Long, rather than opt for a control group immediately older than the treatment group, we examine women born between 1956-1961 in order for our analysis to not be confounded by over- and under-age enrolments in primary school. We also examine the effect on cohorts born between 1965 and 1970, which may be partially affected by the reform given over-age enrolment in primary school. Two measures of treatment intensity are defined: a dummy for those living in high-intensity states, and a measure of federal capital funds dispersed per capita in each state in 1976 (the year of the reform)6. The effect of UPE on schooling and maternal mortality is then estimated using a difference-in-difference framework: yijk = α+β UPE Cohortjk+γUPE Inputk+δ(UPE Inputk×UPE Cohortjk)+X ′ ijkθ+εijk. (2) The schooling and maternal mortality outcomes (y) for individual i in cohort j and state k are regressed on controls for year of birth, religion, ethnicity, and additional controls for groups differentially affected by the 1967-70 civil war (see Akresh et al. (2012) for further discussion.) The UPE Cohort dummy captures whether an individual was of primary school age at the time of the reform (year of birth between 1970 and 1975 inclusive), and UPE Input is a dummy for whether an invididual resides in a treatment high-intensity state, or the investment per capita of a given individual’s state as discussed above. For educational outcomes we expect that δ— the reduced form estimate of the effect of UPE—should be positive, and, if education reduces maternal mortality, that δ < 0 for maternal mortality outcomes. Further details and summary statistics for outcome and treatment variables are provided in section 4. 3.2.2 Zimbabwe 1980 Agüero and Bharadwaj (2010) use extensions of schooling availability to black citizens in 1980 after independence in Zimbabwe to isolate the effect of education on health investments. This extension in years of education was observed in high-school age students, with the authors demon- 5Prior to the UPE program the non-Western states of Nigeria had low primary enrolment rates and low levels of investment per capita. With the introduction of the UPE, these states experienced a considerable expansion in educational investments (Osili and Long, 2008), and so are defined as ‘high-intensity’ states. 6Two additional treatment measures are examined by interacting funds per capita with treatment state, and funds per capita with number of years affected by the treatment. Further details are available as notes to Table 6a 7strating a discontinuity between the ages of 14 and 15. The reform in question occurred in April 1980, and has been used in the literature in earlier papers (see for example Edwards (1995)). In estimating the effect of the reform we follow Agüero and Bharadwaj (2010) in estimating around the discontinuity in educational attainment between the cohorts aged 14 and 15 in 1980: yij = β1DumAgeij+β2DumAgeij×(Age80ij−14)+β3(1−DumAgeij)×(Age80ij−14)+X ′ ijθ+εij . (3) Here DumAge takes the value of 1 for those aged 14 or less in 1980, and 0 for all women aged over 14 in 1980, and linear trends are fitted on each side of this cutoff.7 The Xij vector includes controls for age at 1980, survey fixed-effects, region fixed effects, and a rural dummy variable. In this context β1 is the reduced-form estimate of the effect of the expansion on education and maternal mortality outcome variables. 3.2.3 Kenya 1985 A primary school reform in Kenya in 1985 resulted in a change in the structure of education from a “7-4-2- 3” system (7 years primary education, 4 years of lower secondary, 2 years of upper secondary, and 3 years of tertiary) to an “8-4-4” system. Significantly, this means an increase by one year in required primary schooling to receive the Kenyan Certificate of Primary Education (KCPE). This reform was sharp, taking place nationally in January of 1985 by allowing no students to progress from primary to secondary education in this school year. Chicoine (2011) shows that this has a significant impact on years of completed schooling, both for those women who complete only primary school, as for those women who progress to secondary education. Chicoine (2011) demonstrates that for those individuals born prior to or in 1963, there is null (or very low) probability that they will be treated by the 1985 reform in Kenya. For those born between 1963 and 1972, the probability of treatment is an increasing function of birth year, and for those born post-1972 they are treated with a probability of 1 (where ‘treatment’ implies matriculating to secondary school after 1985, and hence completing an additional year of primary schooling). A treatment variable is then defined in line with the probability that an individual in each cohort is affected by the reform. This variable is approximately identical to that defined by 7This is in line with that suggested by Agüero and Bharadwaj (2010). We also present results estimated with a quadratic and cubic trend. 8Chicoine (2011). We use this treatment in a reduced form equation of the following form: yijq = α+ βTreatjk + age ′ ijqγ + qob trend ′ jqδ + X ′ ijqθ + εijq, (4) where as in the prior subsections we estimate the effect of the plausibly exogenous treatment on individual i, from cohort j, and, in this case, as per Chicoine (2011), control for quarter of birth q. The vectors age (in years) and quarter of birth trend include cubic and quadratic terms respectively, while Xijq includes a quarter of birth fixed effects, ethnicity, and rural/urban controls. 4 Data 4.1 Cross-Country Regressions Measures of educational attainment come from the cross–country dataset compiled by Barro and Lee (2013, 2010) which provides data on total years of schooling, years of schooling by education level (primary, secondary and tertiary), plus a measure of the proportion of the population with each level of education (primary, secondary and tertiary) or with no formal education. The Barro-Lee data allows us to observe the evolution of national attainment figures over time in five-year age groups, both for the entire population, and for the subgroup of females. The age- specific nature of this dataset allows for us to limit our analysis to individuals of fertile age, which for simplicity we define as between the ages of 15 and 39.8 We also construct educational measures based upon individual data collected in the Demographic and Health Surveys (DHS). This data allows for us to observe the education outcomes reported by a representative sample of individuals in a subsample of countries, and allow us to test the robustness of the results using the Barro-Lee data. The DHS education data can be constructed at the regional (sub-country) level and we take advantage of this for our country-specific estimates. The education data are merged by country and year with information on maternal mortality ratios (hereafter MMR) provided by the World Health Organization (WHO, 2012). The MMR indicator is calculated by the World Health Organization (WHO) using figures collected in house- hold surveys and administrative records for maternal mortality and live births. Maternal deaths are classified (ICD 10) as those mothers who die from any cause “related to or aggravated by pregnancy or its management (excluding accidental or incidental causes) during pregnancy and 8Similar estimates are run for the age group 15-49. The qualitative implications of the estimates do not change. 9childbirth or within 42 days of termination of pregnancy. . . ”.9 Given concerns of underreporting by member states and the difficulty in compiling accurate measures of maternal mortality, point estimates are reported along with confidence intervals in the MMR dataset used. Merging Barro-Lee and MMR data results in a panel of 146 countries with observations over a period of 20 years (1990-2010). Both sets of data provide observations on a 5-yearly cycle; 1990, 1995, 2000, 2005 and 2010. To this main dataset we add measures for the log of GDP per capita, immunization rates—which we consider an imperfect proxy for national health expenditures—and fertility which are accessed from The World Bank Data Bank. We construct two further variables for the analysis by combining results from household surveys with available, but sporadic, cross-country measures. In order to control for births attended by skilled health staff and for the adolescent fertility rate, we use household data from the Demographic and Health Surveys (DHS) for that subset of countries and time periods for which a DHS survey is available. Where DHS data is not available, we access more complete cross-country data provided by The World Bank (2013) Data Bank. A precise description of what these variables represent and how they are constructed is provided in appendix A. The full merged dataset is available for download from the authors’ webpage. The estimation sample with non-missing data for all relevant variables contains 108 countries observed quinquennially through 1990-2010. The panel is unbalanced containing gaps in years in which attended births or adolescent fertility observations are not available. A full list of countries and years of availability is presented in table 6 in appendix B, and summary statistics are presented in table 2a below. Trends in maternal mortality and education in the cross-country sample are displayed in figures 1-2. There is considerable variation both across time and continents, but overall a consis- tently positive trend in years of schooling, and a negative trend in regional maternal mortality figures. Exceptionally, in sub-Saharan Africa, maternal mortality stagnated between 1990-1995, beginning to fall in line with other regions during the late 1990s and early 2000s. Despite reductions in maternal mortality in sub-Saharan Africa, this region still had over- whelmingly the highest rates of maternal death associated with childbirth in 2010 (Figure 4). Whilst certain countries in South, Central and East Asia, along with Middle East and North Africa, and Latin America have achieved ratios of less than 100 deaths per 100,000 births, all but a small hand-full of sub-Saharan countries still have ratios which exceed 300 per 100,000 births. 9A full description is provided by The World Health Organization at http://apps.who.int/gho/ indicatorregistry/App_Main/view_indicator.aspx?iid=26 104.2 Natural Experiments In order to focus on country-specific case studies, we use micro-level survey data from the De- mographic and Health Surveys (DHS). These surveys focus on maternal and child health, and have a module focusing on maternal mortality. This maternal mortality module of the DHS is based upon the sisterhood method, where surveyed women are asked about sisters who have died, and whether death was during or related to child-birth (Rutstein and Rojas, 2006). These surveys allow us to calculate cohort specific maternal mortality rates, which can be combined with similarly calculated cohort specific educational attainment figures. Given that we do not have educational data on women who have died during childbirth and are not available to be surveyed, we use the microdata from the DHS respondents themselves. Effectively we work with two individual-level databases: a maternal mortality database of all sisters of respondents and their survival status, and an education database with years of completed education of all survey respondents. Given that we observe data on each respondent’s sisters and their survival status, our mater- nal mortality measure in this case is at the level of the woman, not at the level of the birth. While in the cross-country specification (1) we examine deaths per 100,000 births, in each country case we examine deaths per woman (or the probability that a woman dies in child birth) over her fertile life. This latter measure is referred to as maternal mortality rate, and reflects both the risk of maternal death per birth and the frequency of births in a population. The maternal mortality rate and the maternal mortality ratio are not directly comparable without some measure of the fertility in a given country. In the results section of this paper we use fertilty data described in section 4.1 to provide a rough comparison of the results from cross-country and country-specific estimates. The DHS surveys interview women aged 15-49 at the time of the survey. To correct for the fact that many women are interviewed before they have completed their fertility (younger cohorts in the sample), as a robustness check we use the correction method discussed by Rutstein and Rojas (2006) and Stanton et al. (1997). This involves weighting each age cohort by an adjustment factor, which is calculated based upon its exposure to maternal mortality.10 We note however that in the empirical strategies outlined in section 3.2, all identification is either based upon a double-difference estimator, or includes appropriate cohort-specific trends, so we expect that variation in exposure to maternal mortality should be accounted for even in the absence of this correction. 10These adjustment factors have been calculated as 0.107, 0.206, 0.343, 0.503, 0.664, 0.882, and 0.900 for five age groups from 15-19 to 45-49 (respectively). 11Table 2b presents summary statistics from the DHS data for each of these country-specific ex- periments. In each case we present descriptive statistics for years of education, maternal mortality rate, and treatment intensity, along with the time period examined in the natural experiment of interest. In Zimbabwe and Kenya, the two countries in which the natural experiments occurred at the secondary or upper-primary level, average educational attainment is approximately 7-8 years for all women in the sample, while in Nigeria, where the natural experiment of interest in- volves primary school construction, average education in the cohorts of interest is approximately 5 years. In each case, over the entire range of cohorts examined, betwee 1 and 2 in every 100 women dies during child birth. 5 Results 5.1 Global 5.1.1 The Sensitivity of Maternal Mortality to Education (and Health Interven- tions) Table 3 provides results for our principal cross-country specification (1), estimated on cross- country panel data for 108 developing countries observed at five yearly intervals in 1990-2010, and using the percentage of people in a country and year who have attained specified levels of education. We present similar results disaggregated by country income group and regional classifications as tables 4a and 4b respectively. In order to test the robustness of our results to alternative functional form specifications of the education variable, we run regressions using years of education and years of education squared, and these results are included as appendix tables 8a and 8b. We start with a simple panel data specification in which the identifying variation is within countries over time, with country fixed effects absorbing all country-specific time-invariant components of, for instance, political institutions, climate, culture and deep-set attitudes towards women that may produce a correlation between education and maternal mortality. We then add flexible controls for aggregate trended unobservables and for country and time-varying variables that we expect are correlated with education and maternal mortality. The pooled within-country correlation of maternal mortality rates with women’s education is significantly negative for the percent of the population attaining each education level. The estimates suggest that the MMR-returns to investment in women’s education are positive and weakly diminishing in level of education (this is confirmed by fitting a quadratic in total years 12of education: the quadratic term is positive and significant but the turning point lies outside the sample range). The point estimates are consistently largest for primary education but they are not statistically significantly different from the estimates for higher levels of education. The estimates are robust to the introduction of year dummies, and further time-varying controls which reflect changes in the health and development status of a country. When examining coefficients by region- and income-specific groups (tables 4a and 4b), we find that the relationship between educational improvements and MMR reduction is larger in lower income environments (consistent with their higher baseline rates of maternal mortality and their lower baseline levels of education). Column (2) is the favoured specification. It shows that, conditional upon aggregate trends in maternal mortality, if an additional one percent of the population enrolled in primary education this would lower the number of maternal deaths by 11 per 100,000 live births (The omitted group is the percent of the population with no formal education). This coefficient is estimated with precision (a p-value less than 0.01). It denotes a large impact, equivalent to 0.04 of a standard deviation and 5% of the mean maternal mortality rate in the sample. It is illustrative to consider the impact of moving the entire population of individuals with no education—on average 17.5% of women—into enrolment in primary education. Our estimates suggest this would lower MMR by 196 per 100,000 births, which is 0.68 of a s.d of total MMR. The addition of controls in the remaining columns of the table lowers the coefficient on primary education but in no case is it significantly different from the coefficient in column 2. In the richest model, conditional upon controls for economic and health variables (column 7), the point estimate on primary education is reduced to almost two thirds of its initial size (column 2). For policy purposes it can be useful to look at the estimates conditional and unconditional upon alternative sets of controls since cohort changes in primary education will sometimes evolve similarly to changes in the controls that we introduce. We examine this in the following section. Conditional on the controls introduced, the coefficients on secondary and tertiary education also remain significant. An additional percent of the population moving into secondary education lowers maternal mortality by 10 in 100,000 births. The effect of moving the sample population of uneducated women of fertile age into secondary education is estimated to create a mean reduction in maternal mortality of 172 (0.77 s.d.) in addition to the reduction of 196 associated with their passage through primary education. So the total impact of moving all uneducated women into secondary education is a reduction in MMR of 368 deaths per 100,000 births. In columns 3-7 we introduce a succession of controls for the state of the economy, public 13health care, and fertility. We control for the logarithm of per capita GDP so as to assess whether education is simply proxying income. We find that it is not; the coefficients on education are unchanged and the coefficient on GDP is insignificantly different from zero (column 3). This is important, given that the most recent WHO factsheet on maternal mortality highlights differences in maternal mortality ratios by income and not education (WHO, 2012). Another widely recognised stylized fact highlighted in the WHO factsheet is that maternal mortality is a function of health services. We investigate this, introducing into the equation, DPT immunization rates for children as an indicator of the quality of public health provision.11 The coefficient on primary education falls by about one quarter but this drop is not statistically significant. Immunization has a direct impact on maternal mortality. A one standard deviation increase in immunization rates (which is 15.9% points) lowers maternal mortality by 39 per 100,000 birth, which is 0.14 s.d. or 20% of the mean (column 5). We also include in the equation an index of health provision that is more directly related to maternal mortality. Skilled care before, during and after childbirth is recognised in the public health community to save the lives of both women and new-born babies. Maternal deaths tend to occur because of bleeding and infection and timely management and treatment can make the difference between life and death (WHO, 2012). India and Nepal for instance have, in the last decade, introduced financial incentives to encourage women to give birth in facilities with birth attendants rather than at home (Bauhoff et al., 2012; Powell-Jackson and Hanson, 2012). We find that the percentage of births attended by a skilled professional is associated with lower maternal mortality conditional upon immunization (column 8). A one s.d. increase in the percentage of births attended (which is 27.6% points) results in a drop in MMR of 41 which is 0.14 s.d. or 19% of the mean. Conditioning upon birth attendance does not alter the immunization coefficient. The measure of maternal mortality that we use here is per live birth and so it is mechanically a function of the number of live births in the country and, plausibly, a function of fertility (the number of births per woman). Women in developing countries have on average many more pregnancies than women in developed countries, and their lifetime risk of death due to pregnancy is higher. A woman’s lifetime risk of maternal death—the probability that a 15 year old woman will eventually die from a maternal cause—is 1 in 3,800 in developed countries, versus 1 in 150 in developing countries. Introducing the fertility rate in column 7, we see that an additional birth per mother is associated with an increase in maternal mortality of 28 per 100,000 live births or 11UNICEF for instance regards immunization rates as indicating the quality of overall public health provision in developing countries. Most immunization is directed at children under the age of 12 (and sometimes up to 24) months and our measure of immunization rates pertains to children but it includes tetanus and tetanus injections for pregnant mothers are part of routine prenatal care and in this way immunization rates may also directly influence maternal mortality. 14that a one s.d. increase in fertility (which is 1.7 births) leads to 48 additional maternal deaths, which is 0.16 of a s.d. or 22% of mean fertility. In the main specification however it should be noted that the effect of total fertility is weakly estimated, and not significant at typical levels of confidence. Teenage fertility is highly prevalent in developing countries, and this is closely associated with women dropping out of school early. Adolescent women face relatively high risks of com- plications and death as a result of pregnancy than older women (Conde-Agudelo et al., 2005; Patton et al., 2009). Whether this is because such women are negatively selected or because they are biologically more at risk is unclear and we know of no causal evidence on this. Introducing into the model the percentage of women who give birth at age 15-19, we find a significantly positive association. A one s.d. increase in the percent of teen births (which is 46 percentage points) results in 81 more maternal deaths per 100,000 births, which is 0.27 s.d. and 37% of the mean. Overall, in standardised terms, the association of maternal mortality with fertility and the prevalence of teen births is stronger than its association with either health interventions or education. It is well established that education tends to lower fertility and raise age at birth. In standardized terms, the impacts of primary and secondary education, DPT immunization rates and attended delivery on the maternal mortality ratio are almost identical, each conditional upon the others. To the extent that standardized estimates are useful, this suggests that the payoff to basic education in terms of a health sector outcome is equal to or larger than the payoff to a common index of provision in the health sector. This contributes to a literature documenting synergies between investments in education and health. It is of particular import for policy given that WHO policy documents on maternal mortality have under-emphasised education. Alternative specification of education The specification in appendix Table 8a (column 2) suggests that an additional year of education for all women lowers the number of maternal deaths by 56 per 100,000 live births. A one standard deviation increase in years of education, of 3.21 years, would lower MMR by 180 per 100,000 births, which is 0.59 of a s.d. of MMR. The introduction of controls—particularly the addition of immunization—renders the linear years of education variable insignificant, although the coefficient is still negative. In appendix Table 8b we show estimates using a quadratic in years of education, the functional form suggested by a non-parametric plot of the relationship of interest (see Figure 3). The education coefficient is now robust to the introduction of immunization and other controls. With the exception of fertility which is now significant and of larger magnitude, other controls behave similarly in terms of their effects on MMR as the estimates in Table 3 discussed earlier. The linear and quadratic terms in education are significant and they indicate continuously positive but diminishing marginal returns (in terms of MMR reductions) to years of education. The estimates in column 2 show 15that moving from 0 to 1 year of education is associated with a reduction of 224 deaths per 100,000 live births, while moving from 5 to 6 years of education is associated with a smaller, but still very significant, reduction of 98 maternal deaths per 100,000 births. In the model with all controls (column 8) these estimates are attenuated slightly but still suggest significant effects: reductions of 166 deaths and 56 deaths respectively when moving from 0 to 1 or from 5 to 6 years of education. Similarly, we present results examining the total percentage of women enrolled in primary education (Table 9). Here we see that a 1% increase in total primary school enrollment in a given country is associated with an average reduction of 10 maternal deaths per 100,000 live births (unconditional), or a reduction of 7 deaths conditional on the full set of included controls. Given that our summary statistics suggest that some 17.5% of women were unenrolled over all sample years (or 13% in 2010), considerable reductions in maternal death could potentially be achieved by providing education to this group. 5.1.2 Potential Pathways and Correlated Effects The discussion in the preceding section establishes that declines in maternal mortality in de- veloping countries have been associated with gains in education, improvements in public health services and declines in fertility. Since our primary interest in this paper is in the benefits flowing from education, in this section we consider the extent to which improvements in public health services and declines in fertility are themselves driven by improvements in education. We take each of the controls that were used in Table 3 (including GDP which had no significant impact on maternal mortality) and regress them on education conditional upon country fixed effects (Table 5): fertit = α+ educ ′ itβ + W ′ iγ + uit. (5) We see that increases in education are significant predictors of within-country variation in fertility, immunization, the percentage of attended births and log per capita GDP. The dependent variables in Table 5 have been cast as z-scores (see notes to Table 5) so the coefficients on the education variables indicate the standard deviation change in the outcome flowing from a one year increase in education at the specified level. For example, moving an additional percent of non-school attendees into primary education results in a 0.018 s.d. decline in fertility; an increase of approximately 0.014 s.d. in both immunization rates and attended births; and a 0.001 s.d. increase in log GDP p.c. So education not only directly influences maternal mortality, it also is strongly correlated with other determinants of maternal mortality. 16We can plug the relationships estimated in Table 5 into the specifications in Table 3. For instance, using column 7 in Table 3, we found that a one s.d. reduction in fertility led to a 0.16 s.d. decline in MMR. Since Table 5 column 1 tells us that a single percentage increase in primary education years leads to a 0.018 s.d. decline in fertility, it follows that there is a fertility-mediated impact of primary education on MMR of 0.018 × 0.16 or 0.003 s.d. This is in addition to the direct impact of a year of primary education on MMR in column 7, which is 0.025 s.d. When considering that the total percent of women without formal education is much larger than the one percent which the preceding calculations assume, these effects can become quite large. Similar calculations can be made for education-led impacts of immunization and attended delivery on maternal mortality, and we present these calculations in the final row of table 3. Rather than present these calculations as the impact of moving 1% of women into primary education, we present the impact of moving all uneducated women into schooling. We consider this to be a more relevant indicator for policy. Overall, these results show that education is strongly positively correlated with health service provision conditional upon country fixed effects and aggregate trends. While this association may not be entirely causal, causal components such as that educated people demand public health services merit testing. 5.2 Country-Specific Results We take advantage of the reforms described in sections 3.2.1-3.2.3 to estimate the effect that plausibly exogenous within-country variation in education expansion has on maternal mortality rates. Regression results from these country-specific expansions in schooling are presented be- low in Tables 6a-6c. Each table displays reduced form results for the effect of the reform on educational attainment and on maternal mortality. For each case we also present a descriptive (graphical) summary of the evolution of education, maternal mortality, and the date of the ed- ucational reform (Figures 5a-7b). In those graphs where two vertical cut-off points are plotted, the first signifies the tail end of the “treatment” group, while the second represents the begin- ning of the “control” group (with intermediate cohorts considered as transitionary, or partially affected). Whilst these figures are each plotted from data at the national level, the regression analysis presented in Tables 6a-6c, relies on sub-national variation resulting from variation in reform intensity by region. In each case as a robustness check we also examine the effect of some placebo reform. This placebo reform is defined by comparing birth cohorts who had completely finished the relevant level of school at the time of the reform, and hence who were both unaffected by the expansion in education. We expect in this case that the effect of the false reform—on both education and 17maternal mortality—should not be significantly different to zero, while we expect that the true reform should increase educational attainment and reduce maternal mortality. These placebo results are presented in appendix tables 10a-10c With the exception of one of twenty different outcome treatment-outcome pairs, as expected we see that the placebo reform has no significant effect on either schooling or on maternal mortality. This provides some evidence in favour of the functional form of equations (2), (3) and (4). Turning to the true reforms, we find that in all cases these significantly increased educational attainment. In Nigeria the effect of residing in a high-intensity state (column 2 of Table 6a) is approximately 2.2 additional years of education. The effect of extensions in schooling availability due to independence in Zimbabwe is approximately 1.1 additional years,12 and the effect of the changes in the structure of the KCPE was an additional 0.95 years of education on average for female students. Turning to the reduced form effect of these reforms on maternal mortality, we see that in two of the three cases: Nigeria and Zimbabwe the reforms significantly reduce maternal mortality in the treated cohorts. The exception here is Kenya, where we find a weakly positive (but not significantly different from zero) effect. The lack of a result in this country is perhaps unsurprising for two reasons. Firstly the educational expansion occurred at a relatively high level (between seven and eight years), while our earlier results suggest that the effect of education on maternal mortality is greatest when the educational expansion occurs at low levels. Secondly, the DHS data on maternal mortality from Kenya is by far the noisiest, with large fluctuations by birth cohort over the range of the treatment period. In the case of Nigeria we find statistically and economically significant reductions in maternal mortality by the age of 25 in response to the reform. Depending on the treatment examined, the effect is between a 1 and 2% reduction in total mortality (or between 1 and 2 less women in 100 dying in child birth). In the case of Zimbabwe we find a slightly smaller effect of a 0.4% reduction in total mortality. Once again, it is perhaps not surprising that the effect of the Zimbabwe reform is smaller than the effect of the Nigeria reform, given that the Nigeria reform occured at the primary level, while the Zimbabwe reform largely increased educational attainment at the high school level. We can calculate the approximate effect of an additional year of education by dividing the “second stage” effect of the reform on maternal mortality by the “first stage” effect of the reform on education in years. Using the preferred estimates from Tables 6a and 6b, this gives an estimate of −0.0192/2.179 = −0.0088 and −0.00413/1.148 = −0.0036 respectively. 12This is very similar (and not significantly different) to the 1.252 years estimated by Agüero and Bharadwaj (2010) using only two waves of the Zimbabwean DHS. 18How do these estimates compare with the cross-country estimates reported earlier? It turns out that they match up rather unexpectedly well with cross-country estimates for the effect of additional years of education. Recall that the cross-country estimates use WHO maternal mortality ratios which are maternal deaths per live birth while the country specific estimates use DHS maternal mortality rates which are maternal deaths per woman. In order to provide a rough comparison, we use recent data (Central Intelligence Agency, 2012) on fertility per woman for Zimbabwe and Nigeria to convert the maternal deaths per woman estimates calculated in the paragraph above, to deaths per birth. These figures of 3.61 and 5.38 births per woman (respectively), result in per-birth estimates of 0.0088/5.38 = 0.00164, and 0.0036/3.61 = 0.00010. These correspond to an effect of 164 deaths per 100,000 live births (for an expansion at the lower primary level), and 10 deaths per live birth for an expansion in high school. When we use the years of schooling results from table 8b (column 8), we find that an expansion in schooling from 1 to 2 years results in a reduction in MMR of 150.5 deaths per 100,000 births, and an expansion from 7 to 8 years results in a reduction of 14.75 deaths per 100,000 births: quite close to the country-specific results from Nigeria and Zimbabwe. 6 Conclusion In the last two decades, maternal mortality has declined sharply (by 50%) and education has risen sharply. The analysis in this paper suggests that some part of the decline in maternal mortality may be attributed to the rise in education. Using the cross-country data sample, we use the changes in maternal mortality, education and the control variables in the analysis between 1990 and 2010 to crudely simulate the contribution of education. In this twenty year period, the average percent of women with no formal education has fallen from 22 to 13%. Applying this reduction of 9% to the estimates in Table 3 suggests that education will have led to a decline in the maternal mortality ratio of 105 deaths per 100,000 births (unconditional) or about 71 (conditional upon health services and fertility which, we noted, are correlated with education). The actual decline in the maternal mortality ratio in the same twenty year period in our sample is 119. This suggests that education gains in this period can explain roughly half or possibly more of the observed decline in maternal mortality. Compare this with the role of attended births. The percentage of attended births has risen from 72% to 82%. Our estimates suggest that this will have led to a decline in MMR of 15 per 100,000 births. These simulations are only suggestive but the evidence overall does suggest that the rapid growth in education in developing countries may have made a substantial and largely unrecognised contribution to the dramatic declines in maternal mortality witnessed in recent years. 19Figures Figure 1: Educational Attainment by Region: 1990-2010 4 6 8 1 0 1 2 Y e a rs o f E d u c a ti o n 1990 1995 2000 2005 2010 Year Advanced Economies East Asia Europe/Central Asia LAC MENA South Asia SSA Figure 2: Maternal Mortality Ratio by Region: 1990-2010 0 2 0 0 4 0 0 6 0 0 8 0 0 M M R 1990 1995 2000 2005 2010 Year Advanced Economies East Asia Europe/Central Asia LAC MENA South Asia SSA 20Figure 3: Maternal Mortality and Education: Functional Form 0 5 0 0 1 0 0 0 1 5 0 0 2 0 0 0 0 5 10 15 Years of Schooling MMR Fitted Values (linear) Fitted Values (quadratic) Notes to figure: Each point represents a country average of maternal deaths per 100,000 births. Education data is for women aged 15−39. 21Figure 4: Maternal Mortality Ratio by Country (305,1100] (99,305] (46.5,99] (12,46.5] [2,12] MMR 2010 Maternal Mortality Ratio by Country 22Figure 5a: Educational Attainment by Year – Nigeria 3 4 5 6 7 Y e a rs o f E d u c a ti o n 1950 1960 1970 1980 1990 respondent’s year of birth Series is a 3 year moving average of educational attainment Figure 5b: Maternal Mortality by Year – Nigeria 0 .0 0 5 .0 1 .0 1 5 .0 2 .0 2 5 M a te rn a l M o rt a lit y 1950 1960 1970 1980 1990 year of birth of sibling Series is a 3 year moving average of maternal deaths per woman 23Figure 6a: Educational Attainment by Year – Kenya 4 5 6 7 8 9 Y e a rs o f E d u c a ti o n 1950 1960 1970 1980 Respondent’s Year of Birth Figure 6b: Maternal Mortality by Year – Kenya 0 .0 1 .0 2 .0 3 M a te rn a l M o rt a lit y 1950 1960 1970 1980 year of birth of sibling 24Figure 7a: Educational Attainment by Year – Zimbabwe 2 4 6 8 1 0 Y e a rs o f E d u c a ti o n 1940 1960 1980 2000 Respondent’s Year of Birth Figure 7b: Maternal Mortality by Year – Zimbabwe 0 .0 0 5 .0 1 .0 1 5 M a te rn a l M o rt a lit y 1950 1960 1970 1980 1990 year of birth of sibling Series is a 3 year moving average of maternal deaths per woman 25Tables Country Year Expansion Treatment Years Control Years Indonesia 1973-78 Primary school construction 1968-72 1957-62 Nigeria 1976 Universal Primary Education 1970-75 1956-61 Zimbabwe 1980 High school education expansion 1966 1965 Kenya 1985 Additional year of primary school post-1972 pre-1963 Botswana 1986 High school year rearrangement pre-1970, post 1982 1974 Sierra Leone 2001 Free Primary Education 1990-93 1980-86 Table 1: Educational Experiments: treatment and control groups Variable Obs Mean Std. Dev. Min Max Maternal Mortality 710 220.58 300.88 2.00 1900.00 ln(Maternal Mortality) 710 4.30 1.65 0.69 7.55 GDP per capita 702 9189.71 13870.86 64.36 104512.20 ln(GDP per capita) 702 7.97 1.65 4.16 11.56 Teen Births 670 55.68 46.12 2.80 220.56 Immunization 690 84.75 15.90 18.00 99.00 Percent Attended Births 450 77.29 27.59 0.00 100.00 Fertility 718 3.16 1.68 0.89 8.66 Population 670 4.05+07 1.44+08 95150 1.34+09 Education - Female Total Years of Education 730 7.95 3.22 0.41 13.82 Years of Primary Education 730 4.72 1.68 0.30 8.84 Years of Secondary Education 730 2.84 1.70 0.05 7.14 Years of Tertiary Education 730 0.38 0.38 0.00 2.03 Percent No Education 730 17.51 23.23 0.02 94.25 Percent Primary 730 25.10 18.59 0.00 86.70 Percent Secondary 730 44.98 23.93 1.20 99.13 Percent Tertiary 730 12.40 12.33 0.08 65.01 Education - All Percent No Education 730 14.42 19.17 0.01 83.57 Total Years of Education 730 8.13 2.83 1.13 13.54 Years of Primary Education 730 4.88 1.48 0.87 8.85 Years of Secondary Education 730 2.89 1.57 0.10 7.14 Years of Tertiary Education 730 0.36 0.33 0.00 1.86 Notes: Maternal mortality is expressed in terms of deaths per 100,000 live births. Immuniza- tion is expressed as the percent of children of ages 12-23 months who are immunized against diphtheria, pertussis and tetanus (DPT). Fertility represents births per woman, and teen births are expressed as the number of births per 1000 women between the ages of 15–19. Table 2a: Summary Statistics - Cross Country 26Variable Obs Mean Std. Dev. Min Max Panel A – Nigeria Years of Education 13221 4.822 5.349 0 22 Investment per Capita 12748 0.881 0.545 0.014 2.195 Non-West State 13235 0.828 0.377 0 1 Year of Birth (education) 13235 1968.329 5.119 1956 1975 Maternal Mortality 25354 0.019 0.137 0 1 Under 25 Maternal Mortality 29676 0.006 0.074 0 1 Year of Birth (MM) 29967 1968.472 5.381 1956 1975 Panel B – Zimbabwe Years of Education 10195 7.023 3.788 0 21 High School Enrollment 10195 0.439 0.496 0 1 Treated 10201 0.622 0.485 0 1 Year of Birth (education) 10201 1966.128 4.786 1956 1974 Maternal Mortality 23699 0.013 0.115 0 1 Under 25 Maternal Mortality 28631 0.003 0.055 0 1 Year of Birth (MM) 28842 1966.023 4.736 1957 1974 Panel C – Kenya Years of Education 13712 7.168 4.149 0 23 Treated 13712 0.575 0.443 0 1 Year of Birth (education) 13712 1968.389 8.147 1950 1980 Maternal Mortality 22738 0.014 0.116 0 1 Under 25 Maternal Mortality 25616 0.006 0.076 0 1 Year of Birth (MM) 25686 1967.686 7.770 1950 1980 Notes: Year of birth (education) refers to the birth cohorts of respondents to the DHS surveys for whom we observe educational attainment. Year of birth (MM) refers to the birth cohorts in the maternal mortality data, who are sisters of DHS respondents. In panel A investment per capita refers to federal funds dispersed for school construction in 1976 (in naira). We do not observe this for the state Abuja which has existed since 1991 only. In all panels maternal mortality and under 25 maternal mortality refers deaths due to pregnancy divided by the total number of women. Table 2b: Summary Statistics – Natural Experiments 27(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES MMR MMR MMR MMR MMR MMR MMR MMR Primary Education (% Population) -10.48*** -11.21*** -9.311*** -9.280*** -8.540*** -8.127*** -7.653*** -7.030*** (1.397) (1.676) (1.639) (1.638) (1.730) (1.812) (2.256) (2.087) Secondary Education (% Population) -9.454*** -9.865*** -6.633*** -6.621*** -5.810*** -5.092*** -4.469* -3.881* (1.288) (1.350) (1.670) (1.673) (1.706) (1.878) (2.307) (2.085) Tertiary Education (% Population) -10.37*** -11.77*** -5.918** -6.025** -5.565** -5.111** -4.670* -4.232* (1.421) (1.556) (2.534) (2.521) (2.363) (2.472) (2.648) (2.453) year==1995 -7.806 -9.034 -1.246 -0.973 1.797 -3.266 (12.27) (12.77) (12.17) (12.04) (13.70) (14.08) year==2000 -24.58 -26.09 -14.11 -11.73 -6.262 -9.631 (18.32) (19.04) (17.10) (17.21) (20.35) (20.43) year==2005 -47.33** -53.45* -29.80 -30.54 -23.22 -20.39 (22.98) (27.54) (23.61) (23.15) (26.57) (26.31) year==2010 -73.55** -83.85** -58.52* -57.19* -48.50 -36.02 (29.64) (39.22) (33.54) (32.63) (36.01) (35.59) log GDP per capita 10.67 7.476 12.41 10.12 7.619 (20.47) (20.20) (20.07) (19.49) (19.45) Immunization (DPT) -2.448** -2.238** -2.138** -2.099** (0.944) (0.891) (0.903) (0.906) Attended Births -1.455* -1.258 -1.504* (0.856) (0.796) (0.806) Fertility 16.53 3.622 (24.95) (25.49) Teen births 1.706* (0.865) Constant 1,034*** 1,095*** 862.3*** 782.5*** 941.1*** 947.0*** 838.2*** 765.8** (105.7) (120.1) (133.5) (213.8) (208.0) (202.4) (292.8) (297.0) Observations 710 428 428 428 428 428 428 428 R2 0.329 0.459 0.493 0.494 0.531 0.544 0.546 0.556 Number of Countries 142 108 108 108 108 108 108 108 *** p<0.01, ** p<0.05, * p<0.1 Notes: All regressions include fixed-effects by country. For the full list of countries by year see table 7. Results are for the percent of the female population between the ages of 15 and 39 with each level of education in each country. A full description of control variables is available in section 4, and as the note to table 2a. Heteroscedasticity robust standard errors are shown. Table 3: Cross-Country Results of MMR and Female Educational Attainment 28(1) (2) (3) (4) (5) (6) (7) Advanced East Asia Europe and Latin Amer- Middle East South Sub-Saharan VARIABLES Economies and the Central ica and the and North Asia Africa Pacific Asia Caribbean Africa Primary Education (% Population) -1.961*** -3.938 4.958 -4.592 0.271 -1.142 -7.789 (0.292) (6.844) (3.223) (3.235) (1.794) (8.602) (5.649) Secondary Education (% Population) -2.011*** -2.942 6.244 -3.796 -0.282 -0.184 7.034 (0.330) (8.726) (4.051) (3.140) (2.056) (16.06) (8.223) Tertiary Education (% Population) -1.962*** 15.79 7.201 -3.551 -2.878 41.47 -14.02 (0.369) (21.47) (4.413) (4.345) (2.023) (49.24) (15.44) Constant 204.7*** 432.7 -531.9 531.9* 152.0 612.9 864.5** (31.26) (746.3) (367.4) (284.9) (105.8) (438.7) (339.1) Observations 78 42 67 83 34 23 101 R2 0.713 0.667 0.512 0.402 0.600 0.912 0.680 Number of Countries 18 11 16 20 12 6 25 *** p<0.01, ** p<0.05, * p<0.1 Notes: All regressions include fixed-effects by country. Results are for average years of education of females between the ages of 15 and 39 in each country. Results are reported for specification (2) from table 3 which includes country and year fixed effects. Standard errors are clustered by country. Table 4a: Cross-Country Results of MMR and Female Educational Attainment By Region 29(1) (2) (3) (4) VARIABLES Low Lower Upper High Middle Middle Primary Education (% Population) -4.899 -8.120*** -3.953 0.147 (4.068) (2.109) (2.483) (0.149) Secondary Education (% Population) 2.995 -7.953*** -3.312 0.0235 (5.511) (2.071) (2.450) (0.171) Tertiary Education (% Population) -31.05*** -10.65*** -2.543 0.220 (8.182) (3.647) (2.184) (0.323) Constant 920.9*** 962.1*** 413.2* 5.374 (180.2) (164.1) (212.9) (17.76) Observations 79 114 122 113 R2 0.798 0.581 0.364 0.158 Number of Countries 19 28 31 30 *** p<0.01, ** p<0.05, * p<0.1 Notes: All regressions include fixed-effects by country. Results are for average years of education of females between the ages of 15 and 39 in each country. Results are reported for specification (2) from table 3 which includes country and year fixed effects. Countries are classified according to World Bank income groups, and standard errors are clustered by country. Table 4b: Cross-Country Results of MMR and Female Educational Attainment By Income Group 30(1) (2) (3) (4) (5) VARIABLES Fertility Immunization Percent Attend ln(GDP p.c.) Teen Births Primary Education (% Population) -0.0183*** 0.0147*** 0.0144*** 0.00913*** -0.0116*** (0.00157) (0.00282) (0.00241) (0.00144) (0.00263) Secondary Education (% Population) -0.0323*** 0.0267*** 0.0321*** 0.0197*** -0.0262*** (0.00103) (0.00190) (0.00148) (0.00117) (0.00181) Tertiary Education (% Population) -0.0370*** 0.0238*** 0.0326*** 0.0460*** -0.0289*** (0.00151) (0.00229) (0.00274) (0.00241) (0.00209) Constant 2.362*** -1.845*** -2.227*** -1.687*** 1.844*** (0.0788) (0.180) (0.148) (0.0656) (0.169) Observations 718 690 450 702 670 R2 0.713 0.416 0.682 0.563 0.477 Effect Size 0.049 s.d. 0.030 s.d. 0.036 s.d. 0.004 s.d. 0.054 s.d. *** p<0.01, ** p<0.05, * p<0.1 Notes: Each regression includes fixed effects by country, and heteroscedasticity robust standard errors. Each dependent variable has been transformed to a z-score by subtracting its global mean and dividing by its standard deviation. Education measures are for the female population between 15 and 39. For discussion of the effect size, see section 5.1.2. Table 5: Correlations between Education and Health/Development Outcomes 31(1) (2) (3) (4) Panel A: Outcome – Education Intensity 70-75 1.108** 2.179*** 0.0658 0.198*** (0.535) (0.806) (0.0571) (0.0455) Intensity 65-69 0.739** 1.059** 0.104 0.217*** (0.368) (0.455) (0.0652) (0.0533) Intensity 0.425 (0.491) Obs 12,735 12,735 12,735 12,735 R2 0.427 0.425 0.428 0.429 Panel B: Outcome – MMR Intensity 70-75 0.000722 -0.0156 -0.000596 2.37e-05 (0.00980) (0.0135) (0.00101) (0.000887) Intensity 65-69 0.00107 -0.0106 0.00286 0.00249* (0.00731) (0.00880) (0.00213) (0.00150) Intensity -0.000767 (0.00942) Obs 28,694 28,694 28,694 28,694 R2 0.008 0.008 0.013 0.013 Panel C: Outcome – MMR (Under 25) Intensity 70-75 0.00190 -0.0192** -0.00127* -0.00123*** (0.00607) (0.00951) (0.000652) (0.000320) Intensity 65-69 -2.52e-05 -0.0118** -0.000707 -0.00145*** (0.00405) (0.00518) (0.000708) (0.000479) Intensity -0.00705 (0.00619) Obs 28,694 28,694 28,694 28,694 R2 0.005 0.005 0.007 0.007 *** p<0.01, ** p<0.05, * p<0.1 Notes: Columns (1)-(4) represent different measures of treatment intensity. Column (1) uses capital expenditure on school construction in 1976 in each individual’s state as their treatment intensity, column (2) uses a dummy for residence in non-West (high-intensity) states, column (3) uses the number of years exposed to the reform interacted with the high-intensity state dummy as the intensity measure, and column (4) uses capital expenditure interacted with the high-intensity state dummy. The effect of the reform is identified for 1970-1975 birth cohorts (who are fully affected), and 1965-1969 cohorts, who are affected partially via over-age enrollments. Panel A shows the effect of the education reforms on educational attainment, panel B shows the effect on life-time maternal mortality, and panel C the effect on maternal mortality under the age of 25. All regressions are double-differences, however in columns (2)-(4) the intensity dummy is captured by state of residence fixed effects. Additional controls include religion, ethnicity and year of birth fixed effects, plus time trends by state, and controls for the length of exposure to the civil war in Biafra (Akresh et al., 2012). Standard errors are clustered by state and birth cohort. Table 6a: Effect of 1976 Educational Expansion: Nigeria 32Years of Education Maternal Mortality VARIABLES (1) (2) (3) (4) (5) (6) DumAge 1.148*** 0.428** 0.629* -0.00413** -0.000283 0.00265 (0.167) (0.170) (0.318) (0.00143) (0.00451) (0.00701) DumAge×(Age1980-14) -0.194*** -0.552*** -0.842*** 0.000155 0.000186 0.000582 (0.0250) (0.0704) (0.154) (0.000311) (0.00107) (0.00227) (1-DumAge)×(Age1980-14) -0.148*** -0.313** 0.0633 -0.000249 0.00219 0.00516 (0.0248) (0.114) (0.294) (0.000432) (0.00235) (0.00676) Obs 10,195 10,195 10,195 28,631 28,631 28,631 R2 0.268 0.271 0.272 0.002 0.002 0.002 *** p<0.01, ** p<0.05, * p<0.1 Notes: Columns (1) and (4) include a linear trend for age, columns (2) and (5) a quadratic, and columns (3) and (6) a cubic trend. DumAge (treatment) refers to the birth cohort which was 14 years old at the time of the reform (1980). Additional controls included are survey, region and birth cohort fixed effects, along with a rural dummy variable. Standard errors are clustered at region of residence. Table 6b: Effect of 1980 Educational Expansion: Zimbabwe 33(1) (2) VARIABLES Years of Education Maternal Mortality Treatment 0.953*** 0.00689 (0.265) (0.00553) Observations 13,703 25,602 R2 0.203 0.031 *** p<0.01, ** p<0.05, * p<0.1 Notes: Each regression includes a cubic term for age at time of reform, a cuadratic trend for quarter of birth, fixed effects by quarter of birth and ethnicity, and a dummy for rural or urban residence. The nature of the treatment variable is defined in section 3.2.3. Table 6c: Effect of 1985 Educational Expansion: Kenya 34References C. AbouZahr, E. Royston, and WHO. Maternal mortality: a global factbook. Technical report, WHO/MCH/MSM/91.3, Geneva: World Health Organization, 1991. J. Agüero and P. Bharadwaj. Do the more educated know more about health? evidence from schooling and HIV knowledge in Zimbabwe. Mimeo, 2010. S. Ahmed, L. Qingfeng, L. Li, and A. O. Tsui. Maternal deaths averted by contraceptive use: an analysis of 172 countries. Lancet, 380(9837):111 – 125, 2012. R. Akresh, S. Bhalotra, M. Leone, and U. O. Osili. War and stature: Growing up during the nigerian civil war. American Economic Review, 102(3):273–77, May 2012. R. J. Barro and J. W. Lee. A new data set of educational attainment in the world, 1950–2010. Working Paper 15902, National Bureau of Economic Research, April 2010. R. J. Barro and J. W. Lee. A new data set of educational attainment in the world, 1950–2010. Journal of Development Economics, 104:184 – 98, September 2013. S. Bauhoff, M. Mohanan, G. L. Forgia, K. Babiarz, and K. Singh. Impact of the chiranjeevi yojana program on institutional deliveries and birth outcomes in gujarat, india: Difference-in- difference analysis. Mimeograph, RAND Corporation, 2012. Central Intelligence Agency. World fact book, Nigeria/ Zimbabwe, 2012. URL http://www.cia. gov/library/publications/the-world-factbook. L. Chicoine. Education and fertility: Evidence from a policy change in Kenya. Mimeo, 2011. A. Conde-Agudelo, J. Belizán, and C. Lammers. Maternal-perinatal morbidity and mortality associated with adolescent pregnancy in Latin America: Cross-sectional study. Am J Obstet Gynecol., 192(2):342 – 9, February 2005. A. Costello, D. Osrin, and D. Manandhar. Reducing maternal and neonatal mortality in the poorest communities. BMJ: British Medical Journal, 329(7475):pp. 1166–1168, 2004. J. Currie and E. Moretti. Mother’s education and the intergenerational transmission of human capital: Evidence from college openings. The Quarterly Journal of Economics, 118(4):1495– 1532, November 2003. D. Cutler, A. Deaton, and A. Lleras-Muney. The determinants of mortality. Journal of Economic Perspectives, 20(3):97–120, Summer 2006. D. M. Cutler and A. Lleras-Muney. Understanding differences in health behaviors by education. Journal of Health Economics, 29(1):1–28, January 2010. 35G. Edwards. Rapid educational expansion and primary school efficiency in Zimbabwe, 1983-9. Development Southern Africa, 12(1):87–95, 1995. I. T. Elo and S. H. Preston. Educational differentials in mortality: United States. Social Science and Medicine, 42(1):47 – 57, 1996. A. D. Foster and M. R. Rosenzweig. Technological change and the distribution of schooling: evidence from green-revolution india. Journal of Development Economics, 74(1):87–111, June 2004. S. Glied and A. Lleras-Muney. Health inequality, education and medical innovation. Demography, 45(3):741–761, August 2008. D. Goldman and D. Lakdawalla. Understanding health disparities across education groups. NBER Working Papers 8328, National Bureau of Economic Research, Inc, June 2001. D. Goldman and P. Smith. Can patient self-management explain the health gradient? Goldman and Smith (2002) revisited: A response to Maitra. Social Science & Medicine, 70(1):813 – 815, 2010. E. Goodburn and O. Campbell. Reducing maternal mortality in the developing world: Sector- wide approaches may be the key. BMJ: British Medical Journal, 322(7291):pp. 917–920, 2001. G. T. Greco, T. Powell-Jackson, J. Borghi, and A. Mills. Countdown to 2015: assessment of donor assistance to maternal, newborn, and child health between 2003 and 2006. Lancet, 371 (9620):1268 – 1275, 2008. M. C. Hogan, K. J. Foreman, M. Naghavi, S. Y. Ahn, W. Mengru, S. M. Makela, A. D. Lopez, R. Lozano, and C. J. L. Murray. Maternal mortality for 181 countries, 1980-2008: a systematic analysis of progress towards millennium development goal 5. Lancet, 375(9726):1609 – 1623, 2010. R. Jensen and A. Lleras-Muney. Does staying in school (and not working) prevent teen smoking and drinking? Journal of Health Economics, 31(4):644–657, 2012. K. S. Khan, D. Wojdyla, L. Say, A. M. Gülmezoglu, and P. F. V. Look. Who analysis of causes of maternal death: a systematic review. The Lancet, 367(9516):1066–1074, April 2006. A. E. Kunst and J. Mackenbach. The size of mortality differences associated with educational level in nine industrialized countries. American Journal of Public Health, 84(6):932 – 937, 1994. F. Lichtenberg and A. Lleras-Muney. The effect of education on medical technology adoption: Are the more educated more likely to use new drugs? Annales d’Economie et Statistique Special Issue in memory of Zvi Griliches, (79/80), 2005. 36A. Lleras-Muney. The relationship between education and adult mortality in the U.S. Review of Economic Studies, 72(1):189 – 221, 2005. D. S. Manandhar, D. Osrin, B. P. Shrestha, N. Mesko, J. Morrison, K. M. Tumbahangphe, S. Tamang, S. Thapa, D. Shrestha, B. Thapa, J. R. Shrestha, A. Wade, J. Borghi, H. Standing, M. Manandhar, and A. M. de L Costello. Effect of a participatory intervention with women’s groups on birth outcomes in nepal: cluster-randomised controlled trial. The Lancet, 364(9438): 970–979, September 2004. C. McAlister and T. F. Baskett. Female education and maternal mortality: A worldwide survey. J Obstet Gynaecol Can, 28(11):983–990, November 2006. J. McCarthy and D. Maine. A framework for analyzing the determinants of maternal mortality. Studies in Family Planning, 23(1):pp. 23–33, 1992. E. Meara, S. Richards, and D. Cutler. The gap gets bigger: Changes in mortality and life expectancy by education, 1981-2000. Health Affairs, 27(2):350 – 360, 2008. U. O. Osili and B. T. Long. Does female schooling reduce fertility? Evidence from Nigeria. Journal of Development Economics, 57:57–75, 2008. G. C. Patton, C. Coffey, S. M. Sawyer, R. M. Viner, D. M. Haller, K. Bose, T. Vos, J. Ferguson, and C. D. Mathers. Global patterns of mortality in young people: a systematic analysis of population health data. The Lancet, 374(9693):881 – 92, September 2009. T. Powell-Jackson and K. Hanson. Financial incentives for maternal health: Impact of a national programme in Nepal. Journal of Health Economics, 31(1):271 – 84, 2012. H. Richards and R. Barry. U.S. life tables for 1990 by sex, race, and education. Journal of Forensic Economics, 11(1):9 – 26, 1998. C. Ronsmans and W. J. Graham. Maternal mortality: who, when, where, and why. The Lancet, 368(9542):1189–1200, September 2006. M. R. Rosenzweig. Why are there returns to schooling? American Economic Review, 85(2): 153–58, May 1995. M. R. Rosenzweig and T. P. Schultz. Schooling, information and nonmarket productivity: Con- traceptive use and its effectiveness. International Economic Review, 30(2):457–77, May 1989. S. O. Rutstein and G. Rojas. Guide to dhs statistics. Technical report, Demographic and Health Surveys, Calverton, Maryland, 2006. C. Stanton, N. Abderrahim, and K. Hil. Dhs maternal mortality indicators: An assessment of data and implications for data use. Technical report, September 1997. 37The World Bank. The world bank data bank. http://databank.worldbank.org/ddp/home.do, 2013. Accessed: 10/08/2012. J. Trussell and A. R. Pebley. The potential impact of changes in fertility on infant, child and maternal mortality. Studies in Family Planning, 15(6):pp. 267–280, 1984. WHO. Maternal mortality. Fact Sheet 348, World Health Organization, 2012. A. Yazbeck. Challenges in measuring maternal mortality. Lancet, 370:1291 – 92, 2012. A Data Appendix The variable ‘Percent Attended Births’ represents the number of births attended by a skilled physician. This variable has been constructed from World Bank data relating to births attended by skilled health staff (http://data.worldbank.org/indicator/SH.STA.BRTC.ZS) and DHS data constructed from the cross-country micro data. The DHS data comes from a question regarding place of birth, where those mothers who report births in a public or private facility are assumed to have been attended by a skilled health professional. Given the sporadic nature of the data produced, the following procedure was followed: ‘Per- cent Attended Births’ was set equal to the DHS microdata average for each year/country group pair for which this observation was available. In the case that DHS data was not available and World Bank data was, ‘Percent Attended Births’ was set equal to the rate from the World Bank data. Given that MMR and Education data is available in 5 year periods (1990, 1995, 2000, 2005, 2010), for each of these years where the fifth year was not available but there was data in the preceeding 5 year period, the average of the preceeding 5 year period was taken and used as the data point in the fifth year. 38B Appendix Tables Country 1990 1995 2000 2005 2010 Country 1990 1995 2000 2005 2010 Afghanistan x x Lesotho x x x x Albania x x x x x Liberia x x Algeria x x x x Libya x x Argentina x x x x Lithuania x x x x Armenia x x x x Luxembourg x x Australia x x x x x Malawi x x x x x Austria x x x x x Malaysia x x x x x Bahrain x x x Maldives x x x Bangladesh x x x x x Mali x x x x x Barbados x x x x Malta x x Belgium x x x x x Mauritania x x x Belize x x x x Mauritius x x x x Benin x x x x Mexico x x x x Bolivia x x x x x Mongolia x x x Botswana x x x Morocco x x x x Brazil x x x x x Mozambique x x x x Bulgaria x x x x x Namibia x x x Burundi x x x x Nepal x x x x Cambodia x x x x Netherlands x x x x x Cameroon x x x x x Nicaragua x x x x Canada x x x x x Niger x x x x Chile x x x x Norway x x x x x China x x x x x Pakistan x x x x x Colombia x x x x x Panama x x x x Croatia x x x x Paraguay x x x x Cuba x x x Peru x x x x x Cyprus x Philippines x x x x x Denmark x x x x x Poland x x x x Ecuador x x x x Portugal x x x Estonia x x x x Qatar x x x Fiji x x x Moldova x x x x Finland x x Romania x x x x x France x x x x x Rwanda x x x x x Gabon x x Senegal x x x x Germany x x x x x Serbia x x x Ghana x x x x x Singapore x Guatemala x x x x x Slovenia x x x x Guyana x x x x Sudan x x Haiti x x x Swaziland x x x x Honduras x x x x x Sweden x x x x x Hungary x x x x x Switzerland x x x x x India x x x x Tajikistan x x x x Indonesia x x x x x Thailand x x x Iraq x x Togo x x x x x Ireland x x x x x Tonga x x x x Israel x Tunisia x x x x Italy x Turkey x x x x x Jamaica x x x x Uganda x x x x x Japan x x x x x Ukraine x x x x Jordan x x x x Tanzania x x x x x Kazakhstan x x x x Uruguay x x x Kenya x x x x x Viet Nam x x x x Kuwait x x x x Zambia x x x x x Latvia x x x x Zimbabwe x x x x Table 7: Full Country Data 39(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES MMR MMR MMR MMR MMR MMR MMR MMR Years of Education -49.50*** -56.27*** -26.22* -24.85* -19.40 -12.01 0.218 2.207 (8.061) (10.00) (13.48) (13.56) (12.74) (12.93) (11.32) (9.842) year==1995 -10.53 -14.70 -4.278 -5.525 3.425 -5.867 (12.99) (14.64) (13.54) (13.09) (13.04) (14.78) year==2000 -29.27 -34.85* -18.48 -17.46 1.219 -6.044 (17.91) (20.35) (18.41) (18.06) (18.81) (19.62) year==2005 -50.47** -67.63** -35.17 -39.40 -14.46 -11.45 (21.66) (31.77) (27.52) (26.43) (27.11) (25.69) year==2010 -77.88*** -105.7** -70.66* -73.18* -43.45 -23.81 (27.99) (46.09) (39.79) (38.21) (39.27) (35.90) log GDP per capita 24.59 18.77 25.46 15.79 11.04 (22.95) (21.59) (20.77) (21.04) (20.46) Immunization (DPT) -3.303*** -2.979*** -2.461** -2.354** (0.952) (0.931) (0.960) (0.979) Attended Births -1.908** -1.068 -1.499 (0.899) (0.939) (0.935) Fertility 59.40*** 30.28 (18.63) (22.48) Teen births 2.905** (1.340) Constant 611.0*** 658.7*** 455.9*** 263.5 529.3*** 538.6*** 204.3 172.4 (63.58) (79.66) (97.98) (201.4) (185.7) (176.4) (185.6) (206.9) Observations 710 428 428 428 428 428 428 428 R2 0.164 0.246 0.288 0.293 0.363 0.385 0.416 0.448 Number of Countries 142 108 108 108 108 108 108 108 *** p<0.01, ** p<0.05, * p<0.1 Notes: All regressions include fixed-effects by country. For the full list of countries by year see table 7. Results are for average years of education of females between the ages of 15 and 39 in each country. A full description of control variables is available in section 4, and as the note to table 2a. Heteroscedasticity robust standard errors are shown. Table 8a: Cross-Country Results of MMR and Female Educational Attainment (years) 40(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES MMR MMR MMR MMR MMR MMR MMR MMR Years of Education -233.0*** -236.5*** -207.1*** -206.3*** -191.5*** -185.4*** -191.9*** -184.4*** (23.83) (26.67) (24.03) (23.55) (26.85) (29.12) (37.32) (36.65) Years of Education Squared 12.34*** 12.51*** 12.53*** 12.49*** 11.68*** 11.44*** 11.73*** 11.31*** (1.347) (1.567) (1.522) (1.490) (1.581) (1.672) (2.040) (2.026) year==1995 -8.620 -9.644 -4.161 -4.580 -6.012 -8.210 (10.81) (12.35) (12.01) (11.68) (13.14) (13.64) year==2000 -25.12* -26.49 -17.91 -17.58 -20.62 -21.82 (14.71) (16.99) (15.91) (16.01) (18.89) (19.10) year==2005 -50.16*** -54.36** -37.12* -38.49* -42.52* -40.69 (17.23) (24.80) (22.21) (21.47) (25.56) (25.21) year==2010 -75.56*** -82.38** -64.34** -65.31** -69.95** -63.63* (21.61) (34.77) (30.85) (30.36) (34.74) (34.07) log GDP per capita 6.023 3.976 6.519 7.607 6.602 (17.30) (16.38) (15.60) (15.89) (16.02) Immunization (DPT) -1.843* -1.765** -1.818** -1.812** (0.934) (0.885) (0.867) (0.873) Attended Births -0.636 -0.740 -0.870 (0.841) (0.827) (0.833) Fertility -9.667 -15.13 (22.73) (23.58) Teen births 0.793 (0.895) Constant 1,163*** 1,183*** 981.7*** 933.1*** 1,038*** 1,030*** 1,097*** 1,057*** (95.07) (105.2) (99.20) (156.1) (147.2) (145.3) (212.5) (226.8) Observations 710 428 428 428 428 428 428 428 R2 0.425 0.538 0.580 0.580 0.601 0.604 0.604 0.606 Number of Countries 142 108 108 108 108 108 108 108 *** p<0.01, ** p<0.05, * p<0.1 Notes: All regressions include fixed-effects by country. For the full list of countries by year see table 7. Results are for average years of education of females between the ages of 15 and 39 in each country. A full description of control variables is available in section 4, and as the note to table 2a. Heteroscedasticity robust standard errors are shown. Table 8b: Cross-Country Results of MMR and Female Educational Attainment (quadratic years) 41(1) (2) (3) (4) (5) (6) (7) (8) VARIABLES MMR MMR MMR MMR MMR MMR MMR MMR Primary Enrollment -9.805*** -10.40*** -9.043*** -8.967*** -8.141*** -7.744*** -7.623*** -6.951*** (1.283) (1.419) (1.523) (1.517) (1.662) (1.764) (2.286) (2.142) year==1995 5.297 3.082 10.14 10.72 11.37 5.545 (10.09) (11.27) (11.17) (11.17) (12.33) (13.37) year==2000 -1.735 -4.656 6.282 8.910 10.20 5.786 (12.21) (14.41) (13.71) (13.92) (16.17) (16.73) year==2005 -13.81 -23.34 -1.080 -0.410 1.358 3.043 (11.72) (21.49) (19.86) (19.22) (21.42) (21.19) year==2010 -29.60** -45.22 -21.70 -19.16 -17.07 -5.752 (14.38) (32.51) (29.38) (28.45) (30.47) (29.53) log GDP per capita 14.37 10.72 14.39 13.82 10.96 (20.29) (19.65) (19.49) (19.43) (19.31) Immunization (DPT) -2.419** -2.252** -2.229** -2.186** (1.004) (0.972) (0.987) (0.992) Attended Births -1.103 -1.054 -1.309 (0.906) (0.843) (0.855) (mean) fertility 3.739 -9.717 (25.27) (25.81) Teen births 1.769* (0.925) Constant 1,026*** 1,082*** 976.8*** 864.0*** 1,018*** 1,025*** 1,000*** 922.3*** (105.4) (118.9) (126.4) (203.4) (200.8) (193.8) (293.4) (306.7) Observations 710 428 428 428 428 428 428 428 R2 0.325 0.450 0.467 0.468 0.505 0.512 0.512 0.524 Number of Countries 142 108 108 108 108 108 108 108 *** p<0.01, ** p<0.05, * p<0.1 Notes: All regressions include fixed-effects by country. For the full list of countries by year see table 7. Results are for total percent of females between the ages of 15 and 39 in each country who are enrolled in at least primary school. A full description of control variables is available in section 4, and as the note to table 2a. Heteroscedasticity robust standard errors are shown. Table 9: Cross-Country Results of MMR and Female Educational Attainment (percent primary enrollment) 42(1) (2) (3) (4) Panel A: Outcome – Education Intensity 56-60 1.935** 1.529 -0.315 0.237 (0.864) (1.183) (0.340) (0.165) Intensity -0.717 (0.928) Obs 2,617 2,617 2,617 2,617 R2 0.300 0.297 0.299 0.300 Panel B: Outcome – MMR Intensity 56-60 -0.0147 -0.00636 -0.00298 -0.00336 (0.0104) (0.0115) (0.00384) (0.00214) Intensity 0.00140 (0.0110) Obs 6,508 6,508 6,508 6,508 R2 0.015 0.014 0.015 0.015 Panel C: Outcome – MMR (Under 25) Intensity 56-60 3.52e-05 0.00186 -0.00169 -0.00280 (0.00642) (0.00598) (0.00229) (0.00211) Intensity -0.0103 (0.0101) Obs 6,508 6,508 6,508 6,508 R2 0.010 0.010 0.010 0.011 *** p<0.01, ** p<0.05, * p<0.1 Notes: For a full description of outcomes and treatments see Table 6a. A placebo treatment here is defined by comparing two groups who had already left primary school by the time of the reform. The birth cohorts from 1956-1961 were defined as the placebo ‘treatment’ and the cohorts from 1950-1955 were defined as controls. Table 10a: 1976 Educational Expansion Placebo: Nigeria 43Years of Education Maternal Mortality VARIABLES (1) (2) (3) (4) (5) (6) DumAge -0.464 -0.201 -0.241 0.00439 -0.00142 -0.000423 (0.259) (0.261) (0.382) (0.00307) (0.00605) (0.00684) DumAge*(Age1980-14) -0.357*** 0.0358 -0.131 0.000489 -0.000297 -0.00266 (0.0309) (0.0985) (0.205) (0.000268) (0.00133) (0.00338) (1-DumAge)*(Age1980-14) -0.0513 -0.220* -0.171 0.000598 -0.00253 -0.000121 (0.0522) (0.105) (0.356) (0.000542) (0.00253) (0.00678) Obs 6,770 6,770 6,770 20,836 20,836 20,836 R2 0.234 0.237 0.237 0.001 0.001 0.001 *** p<0.01, ** p<0.05, * p<0.1 Notes: For a full description of outcomes and treatments see Table 6b. A placebo treatment here is defined by comparing two groups who had already left primary school by the time of the reform in 1980. The placebo ‘treatment’ was defined between the cohorts born in 1960 and 1961 (20 years old at the time of the reform), and the same window is used (16 years) as the real treatment in Table 6b. Table 10b: 1980 Educational Expansion Placebo: Zimbabwe 44(1) (2) VARIABLES Years of Education Maternal Mortality Treatment -0.0994 -0.0152 (0.531) (0.00978) Observations 13,703 25,602 R2 0.202 0.031 *** p<0.01, ** p<0.05, * p<0.1 Notes: For a full description of outcomes and treatments see Table 6c. A placebo treatment here is defined by comparing two groups who had already left primary school by the time of the reform in 1985. The placebo ‘treatment’ was defined as occurring in 1977, and hence affecting (at least partially) birth cohorts from 1955 to 1963, rather than the true affected cohorts of 1964 to 1972. Table 10c: 1985 Educational Expansion Placebo: Kenya 45