WATER GLOBAL PRACTICE QUALITY UNKNOWN BACKGROUND PAPER The Nitrogen Legacy SEPTEMBER 2019 Esha Zaveri, Jason Russ, Sébastien Desbureaux, The Long-Term Effects of Water Pollution Richard Damania, on Human Capital Aude-Sophie Rodella, and Giovanna Ribeiro About the Water Global Practice Launched in 2014, the World Bank Group’s Water Global Practice brings together financing, knowledge, and implementation in one platform. By combining the Bank’s global knowledge with country investments, this model generates more firepower for transformational solutions to help countries grow sustainably. Please visit us at www.worldbank.org/water or follow us on Twitter at @WorldBankWater. About GWSP This publication received the support of the Global Water Security & Sanitation Partnership (GWSP). GWSP is a multidonor trust fund administered by the World Bank’s Water Global Practice and supported by Australia’s Department of Foreign Affairs and Trade, the Bill & Melinda Gates Foundation, the Netherlands’ Ministry of Foreign Affairs, Norway’s Ministry of Foreign Affairs, the Rockefeller Foundation, the Swedish International Development Cooperation Agency, Switzerland’s State Secretariat for Economic Affairs, the Swiss Agency for Development and Cooperation, U.K. Department for International Development, and the U.S. Agency for International Development. Please visit us at www.worldbank.org/gwsp or follow us on Twitter #gwsp. The Nitrogen Legacy The Long-Term Effects of Water Pollution on Human Capital Esha Zaveri, Jason Russ, Sébastien Desbureaux, Richard Damania, Aude-Sophie Rodella, and Giovanna Ribeiro SEPTEMBER 2019 © 2019 International Bank for Reconstruction and Development / The World Bank 1818 H Street NW, Washington, DC 20433 Telephone: 202-473-1000; Internet: www.worldbank.org This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Rights and Permissions The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Please cite the work as follows: Zaveri, Esha, Sébastien Desbureaux, Richard Damania, Aude-Sophie Rodella, and Giovanna Ribeiro. 2019. “The Nitrogen Legacy: The Long-Term Effects of Water Pollution on Human Capital.” World Bank, Washington, DC. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights​ @­worldbank.org. Cover photos (left to right): Franck Barske from Pixabay, Wietze Brandsma from Pixabay, Yogendra Singh from Pexels Cover design: Jean Franz, Franz and Company, Inc. Contents Abstract1 1. Introduction 1 2. Data 3 3. Empirical Methods 8 4. Results 9 5. Robustness Checks 12 6. Evidence From Other Regions 14 7. Mechanisms 16 8. Conclusion 16 Notes 18 References 18 model Appendix 1  Spatial Stream Network ­ 21 Additional references 22 Figures 1. Upstream-Downstream Hydrologic Breakdown 6 2. Countries Studied, Africa 7 3. Different Window Periods Of Exposure 13 Tables 1. Descriptive Statistics 10 2. The Long-Term Impacts of Upstream Pollution on Health 10 3. The Long-Term Impacts of Local Pollution on Health Using Instrumental Variables 11 4. Falsification Test 13 5. Main Results with Trends for Districts 14 6. Alternative Clustering 14 7. Impacts in Vietnam 15 8. Impacts in Africa 15 The Nitrogen Legacy iii Abstract The fallout of nitrogen pollution is considered one of the largest global externalities facing the world, health. In this paper we present new evidence that nitrogen pol- impacting air, water soil and human ­ c apital. Numerous studies have lution in water is an important determinant of variations in human ­ shown a link between adverse conditions experienced during early-life, such as those caused by con- outcomes. However, causally derived links between ­ flict and disease, and adult ­ early-life exposure to nitrogen pollution in water and later-life health outcomes have not been extensively e ­ xplored. In this paper we combine data from the Demographic and Health Survey (DHS) dataset across several ­countries, India, Vietnam and 33 African countries to analyze the causal links between pollution expo- sure experienced during the very earliest stages of life and later-life ­ health. Our results show that pollution exposure experienced in the critical years of development – from the period of birth up until year three – is associated with decreased height as an adult, a well-known indicator of overall health ­ hecks. Because adult height is related to educa- and productivity, and is robust to several statistical c tion, labor productivity, and income, this also implies a loss of earning ­ potential. The analysis begins within an assessment in India where data are more available and is then extended to geographic set- ­ frica. Results are consistent and show that ear- tings including Vietnam, and across 33 countries in A ly-life exposure to nitrogen pollution in water can lower height-for-age scores during childhood in Africa. These findings add to the evidence on the enduring conse- Vietnam and during infancy in ­ ­ ntervention. quences of water pollution and identify a critical area for policy i Keywords: Water Pollution, Health, Nitrogen JEL Codes: O13, O15, Q15, Q25, Q54 1. Introduction A hundred years since the ingenious experiments in nitrogen fixation by Haber and Bosch which resulted in the development of the first nitrogen-based synthetic fertilizers, the fallout of nitrogen ­ pollution is considered one of the most important environmental issues of the twenty-first century (Kanter, 2018). Recent studies suggest that nitrogen may be the world’s largest global externality, due ­ l. 2016). The world has also surpassed to its effects on human health and the environment (Keeler et a the planetary boundary for nitrogen-- a level of human interference beyond which environmental damage increases dramatically and possibly permanently (Steffen et a ­ ­ l. 2015). In water, excess reactive nitrogen can promote the growth of algae, which can trigger toxic blooms ­ ealth. It is one of the few water pol- that can kill fish, and nitrate in drinking water can harm human h lutants that is trending upwards nearly everywhere, including in developed countries like the United States, despite strong regulation (Keiser and Shapiro 2018). The legacy effects of nitrogen pollution on Quality. We are grateful to Kelly Baker This research was part of the World Bank report titled Quality Unknown: The Invisible Crisis of Water ­ and seminar and conference participants at the International Food and Policy Research Institute (Delhi), Center for Policy Research (Delhi), Indian Statistical Institute (Delhi), the annual meetings of the Association of Environmental and Resource Economists (AERE) (2019), and ­eedback. Ali Sharman provided excellent the Workshop on Environmental Economics and Data Science (TWEEDS) (2019) for helpful f assistance. All errors are our own. research ­ The Nitrogen Legacy 1 the environment are likely to endure decades after nitrogen inputs have ceased with long time lags between the adoption of conservation measures and any measurable improvements in water quality al. 2018)1. In humans, the health impacts can be acute causing infant death due to met- (Van Meter et ­ oxygen.2 hemoglobinemia, or the blue baby syndrome, that reduces the blood’s ability to transport ­ However, causal evidence for the long-term and legacy health impacts of early-life exposure to nitro- ­ imited. gen pollution are still l This paper addresses this gap by examining the impact of nitrate-nitrogen pollution on height and Africa. India, well-being in India, along with supporting evidence from Vietnam and 33 countries in ­ provides a compelling setting in which to study the impacts of water pollution caused by nitrate-­ ­ nitrogen. The Green Revolution, starting in the mid-1960s, was a watershed moment in Indian agriculture. Along with a rapid increase in agricultural productivity, it also led to a dramatic rise in the ­ potassium ­ consumption of synthetic nitrogenous fertilizers such as nitrogen-phosphate-­ (NPK). The five-fold rise in the use of NPK fertilizers per hectare of cultivated land since the mid-1960s resulted in profound changes to the nitrogen cycle with impacts on India’s waters — runoff of excess nitrogen from fields increased concentrations of nitrate in the waters to unsafe levels (Fields, 2004). But agri- ­ ollution. According to the first-ever decade-long nitro- culture is just one of the sources of nitrogen p gen assessment conducted for India by the Indian Nitrogen Group (ING), a voluntary body of over a hundred scientists and other stakeholders, sewage and organic solid wastes are some of the fastest growing sources of nitrogen pollution in the country (INA, 2017). Most previous work that provide causally interpretable estimates have primarily focused on short- run and immediate birth outcomes (Brainerd and Menon, 2014; Jones, 2019)3. These studies show early-life exposure to nitrogen-related pollution can lead to infant mortality (Brainerd and that ­ Menon, 2014) and low birthweight (Jones, 2019). The well-established fetal origins literature suggests that intrauterine health impacts can lead to lasting health damages, and that low birthweight is asso- ciated with shorter height in adulthood (Barker, 1990; Almond and Currie, 2011; Currie and Vogl, 2013; Christian et a ­ l., 2013; Almond, Currie and Duque, 2018). And low birthweight is a well-known marker for many health problems later in life, including coronary heart disease (Barker 1995), ­ l. 1998), and decreased glucose tolerance (and thus a higher propensity for obesity) (Ravelli et a al. 2011). Still, there has been no attempt to quantify increased rates of all-cause mortality (Risnes et ­ the full extent of health damages, especially the irreversible and lagged human capital impacts, as a water. result of nitrogen pollution in ­ Evidence for lagged human capital impacts in environmentally vulnerable and poor locations is ­ mpacts. This is particu- severely limited due to the paucity of longitudinal data that can trace long-term i larly challenging for water pollution since monitoring of water quality is sparse in space and time, and is site-specific. In this paper, we exploit temporal and geographic variation in nitrogen pollution exposure ­ with a newly constructed database of water quality that combines in situ monitoring station data with a geospatial statistical model for stream networks developed by ver Hoef and Peterson (2010). We carefully integrate early-life exposure to nitrogen pollution between the critical years of development – from the period of birth up until year three – with women’s health outcomes, climatic factors, correlated pollut- ­ nalysis. ants, household inputs and other socioeconomic demographics for our a 2 The Nitrogen Legacy Our research design exploits the direction of river flow and the upstream-downstream geographic ­ l. 2018; Garg et a relationship used in past literature (Do et a ­ l. 2016) to estimate a pollution-health relationship. Because the costs imposed by water pollution are largely felt in downstream regions, the ­ regions. To analysis focuses on the impact of upstream pollution on health outcomes in downstream ­ isolate the average pollution spillover at downstream locations, the analysis uses a rich set of ­ controls. These are meant to control for time-invariant, location-specific characteristics such as local soil quality and natural resource endowments, as well as factors that vary by year and month, such as weather, and development. The analysis also controls for national trends in economic output and technological ­ ­ olicies. To ensure that later-life time-varying factors that are specific to states to capture state-level p health outcomes are measured in the same location of conception and birth where exposure occurred, the sample is restricted to individuals that have never migrated from their place of ­birth. In this way, our empirical strategy controls for a wide number of potential confounders in an effort to identify causal effects. To test for external validity, similar analyses are conducted in geographies outside of India - ­ Africa. Vietnam and 33 countries in ­ Our results find that nitrogen exposure experienced by infants can have durable, long-term impacts that stretch well into ­adulthood. In India, women exposed to nitrogen pollution in their earliest years of life are shorter on average in adulthood than women of similar circumstances who were not pollution. Early-life exposure to nitrogen pollution also lowers later-life labor exposed to such ­ ­ welfare. This finding is robust to several productivity and depresses adult wages decreasing overall ­ ­sensitivity and falsification ­tests. Analyses across different geographic settings in Vietnam and Africa that measure the impact of nitrogen pollution during early-life provide further supporting evidence India. Taken together, this paper provides new evidence that early-life expo- for the results found in ­ sure to nitrogen pollution has enduring and irreversible costs on human capital with decreases in height observed across different life-stages: in adulthood (India), in childhood (Vietnam), and in infancy ­(Africa). ­ ollows. In Section 2 we describe the health and water quality The rest of the paper is organized as f data, the construction of the main variables used in the analysis as well as the procedure we use to pollution. Section 3 outlines our empirical strategy and Section 4 match the health data to upstream ­ results. Robustness checks are provided in Section 5. Section 6 investigates the external discusses the ­ Africa. validity of the results reported for India in other geographic settings such as Vietnam and ­ Section 7 discusses the plausible mechanisms linking nitrogen pollution to height impacts, and Section 8 ­concludes. 2. Data In this section, we describe the data sources that were used in the empirical analysis, and the construc- analysis. tion of the main variables in the ­ 2.1. Health Data The data on our outcomes of interest come from the fourth round of the National Family and Health India. The NFHS is the Demographic and Health Survey (DHS) equivalent Surveys (NFHS) conducted in ­ The Nitrogen Legacy 3 India. The survey was conducted between January 2015 – December 2016 and covered all areas of the in ­ country. In a departure from the previous DHS surveys, the sample for this survey was designed to be ­ level. Close to 600,000 households were interviewed which included representative at the district ­ 0.7 million eligible women in the age group 15-49. The main variables in this analysis come from the woman’s questionnaire where a number of anthropometric measures are c ­ ollected. variable. The micro-econometric literature often uses We make use of adult height as our main health ­ adult height as a proxy indicator for overall health and long-term adult well-being since it reflects the ­ dolescence. A rough consensus drawn from accumulation of shocks to health through childhood and a this literature is that an improvement in health associated with a 1-centimeter increase in adult height raises productivity by 3.4 percent (Kraay, 2018). Respondent’s height is reported in centimeters in the DHS ­data. ­ ocation. We utilize this The DHS also records how long the individual has resided in the current l information to restrict the sample to only those women whose birth-place coincides with the cur- location. This allows us to guard against the possibility of mis-measuring exposure to nitrogen rent ­ pollution if the woman was born in a different location than the location in which she currently ­ resides. We trace the birth-year histories of all adult women ranging from 1966 to 1999, a period when the effect of the Green Revolution was already in force yet nitrogen fertilizers were still use. increasing in ­ 2.2. Water Quality Data The Central Water Commission (CWC) within the Ministry of Water Resources monitors data on ambient ­ ountry. We compile and harmonize a rich dataset of water pollution mea- water quality throughout the c surements between the years 1963-2017 along a network of 375 river monitoring stations throughout India. Many gaps in the data exist since the pollution measures are not consistently recorded over the ­ unbalanced. Over the 1963-2017 period, 75% of water quality entire sample time frame and the panel is ­ data are missing (60% between 1986-2017). To circumvent these problems, we build on a new class of spatial statistical network model for stream data to interpolate and fill in missing observations across monitor-year pairs (ver Hoef and Peterson, 2010). The model takes advantage of the fact that water qual- ity in a station downstream depends on environmental conditions and human activities upstream of the ­ etwork). Therefore, spatial station, and on the water quality “received” from upstream (directed n covariates in a well-defined upstream area and the spatial dependence between observations based on stream distance allow to model water quality and predict it in unsampled ­locations. We train such model observations. We then collapse monitor level observations to the district ­ to fill-in missing ­ level. A more complete description of the model is provided in Appendix 1. We focus on cumulative exposure to nitrate-nitrogen when the concentrations exceed safety thresholds of 10mg/l from the year of birth to age 3. Prior work suggests that the first 1000 days of a ­ child’s life are the most critical for early childhood development and for determining whether a child stunted. It has also been shown that height at age three strongly predicts adult height will grow up ­ (Maccini & Yang, 2009). Lower height-for-age scores can lead to severe consequences for cognitive ­ dulthood. development, overall health, and even socio-economic conditions that carry into a 4 The Nitrogen Legacy We assign each woman a fractional measure of the share of years exposed to high levels of nitrate-­ born. Since districts have nitrogen between the year of birth and age 3 in the district where she was ­ time. We then compare later-life split over time, we use parent districts to allow comparability across ­ health outcomes among cohorts with more and less pollution exposure after accounting for a rich set of ­controls. Even though direct measures of drinking water quality are unavailable, in-situ monitoring data serve as a reasonable proxy for proximate levels of nitrates in drinking ­water. This is because nitrates are noto- riously expensive and difficult to clean out of water, and cannot be sufficiently treated using conven- tional ­methods.4 Evidence from a slew of countries around the world, including Morocco, Niger, Nigeria, Senegal, India, Pakistan, Japan, Lebanon, Philippines, the Gaza Strip and Turkey, show that nitrates in ­ l., 2018).5 drinking water often cross conventional safety thresholds (Ward et a 2.3. Additional Controls We control for both average rainfall in millimeters and average temperature in degrees Celsius as ­ l. 2019, these have been shown to impact adult outcomes (Maccini and Yang 2009, Fishman et a Hyland and Russ 2019) and are known to also interact with nitrate loadings in waterways (Zheng et ­al. Department. In some specifications, we 2016). These are obtained from the Indian Meteorological ­ ­ ataset. They are an oft-used measure of domestic pol- also control for fecal coliforms from the CWC d ­ perations. It is mea- lution, and are a major focus of water supply, sanitation, and hygiene (WASH) o sured as the “most probable number” of coliform organisms per 100 mL of water (MPN/100 ml, thousands). reported in ­ 2.4. Matching Health Data To Water Quality Data The primary challenge to evaluate the pollution-health relationship is the endogeneity of pollution ­ activities. In the exposure. Pollution is not randomly assigned and is often the byproduct of productive ­ ­ aste. case of nitrates in the water, it is largely a byproduct of intensive agriculture and untreated urban w Thus, a naïve approach which examines impacts of local pollution on local impacts will likely conflate the positive effects of increased production with the negative externalities of water pollution, and effect. To circumvent this bias, we construct a measure of upstream pollution underestimate the latter’s ­ flow. Similar techniques to identify upstream-downstream relationships using the geography of river ­ al., 2018’ Do et ­ have been applied in recent economics literature (Garg et ­ al., 2018 and Keiser, 2018). This choice is predicated on the fact that the decision to pollute upstream is orthogonal to downstream downstream. health, while geography dictates that pollution flows ­ We make use of a digital elevation model from the Shuttle Radar Topography Mission (SRTM) mis- sion to identify the direction of stream flow and to track upstream and downstream through surface waters in I ­ ndia. We link the districts in our sample to all other districts that are upstream from it as connected by the stream network as shown in Figure 1. Since water quality decays over time, we distance between upstream and downstream district-pairs such that the upstream district bound the ­ apart. For any given downstream district, we then is the closest upstream district, and up to 300 km ­ concentrations of nitrate-nitrogen pollution in the upstream d calculate the average ­ ­ istricts. The Nitrogen Legacy 5 FIGURE 1. Upstream-Downstream Hydrologic Breakdown Note: The map shows direction of streamflow from upstream to downstream districts. 2.5. Data for Vietnam and Sub-Saharan Africa To test for external validity of the results from India, we also measure the impact of nitrogen pollution ­ frica. on health in other geographic settings: Vietnam and 33 countries in A Data for Vietnam Water quality data comes from the Mekong River Commission (MRC) which collects data for four coun- tries (Cambodia, Lao PDR, Thailand, and Vietnam) spanning the years 1985-2010, and covers the main tributaries of the Mekong ­River. Our health data comes from the latest Vietnam Living Standards Survey (VLSS) of 1997–98 where we focus on the health outcomes of children aged 4 to 12 ­years. The VLSS 97–98 ­ ountry. was a nationally representative survey that sampled almost 6,000 households across the c For each member of the surveyed household, the survey contains information on gender, year of birth, 6 The Nitrogen Legacy ­ utcomes. At the household level, information is available on the ethnicity of age, and anthropometric o residence. Our sample is restricted to those that have always the household head and the province of ­ ­ lace. Because nitrate-nitrogen levels in Vietnam are relatively lower, exposure to resided in the same p nitrate pollution is examined at levels that are above the 75th percentile in the distribution, or roughly mg/L. Following a similar methodology described in 2.4, each VLSS commune is matched to its 2 ­ counterpart. upstream pollution ­ Data for Sub-Saharan Africa We use data from 90 Demographic and Health Surveys (DHS) spanning over a period of 23 years to analysis. Figure 2 shows the 31 countries account for all child and household variables presented in the ­ Egypt. The dots represent included in the analysis from Sub-Saharan Africa, as well as Morocco and ­ ­ ive. We focus on anthro- the approximate locations of the communities where households in the survey l pometric measures of children up to 5 years of ­ age. We convert children heights into Z-scores using the FIGURE 2. Countries Studied, Africa Program. Note: The map shows locations of enumeration areas that were surveyed as part of the DHS ­ The Nitrogen Legacy 7 WHO growth standards (WHO 2006). Doing so allows us to assess child height relative to well-nourished sex. For our main outcome variable, we use height-for-age Z-score (HAZ) children of the same age and ­ and low HAZ ­(i.e. HAZ below -2) which reflects ­stunting. Water quality data comes from a machine learn- al. (2019). Each birth record is then matched to nitrate pollution ing algorithm presented in Damania et ­ ­ pstream. These urban centers where identified using data flowing from urban centers that are farther u of urban agglomerations from Africapolis (OECD/SWAC (2018)). 3. Empirical Methods To estimate the long-run health impacts of childhood exposure to nitrogen pollution, the research design exploits quasi-random variation in exposure to nitrogen pollution experienced by different birth districts. Specifically, the analysis compares height outcomes between exposed cohorts in different ­ and non-exposed cohorts, controlling for average differences in these outcomes across birth years and districts. The estimating equation for individual-level outcome Y of person i during time t and across ­ below. born in district d and state s is presented ­ [birth , 3 ] U idt Yidt = α + β N + λ Xdst + γ Did W + ρ mt + ρ s + tµd + ∈idt [birth , 3 ] U idt N , where superscript U denotes upstream, is the fraction of years from the time of birth to age 3 that individual i was exposed to nitrate-nitrogen levels from upstream areas that exceeded safe limits in their birth district ­d. It serves as a measure of cumulative pollution exposure in early life during development. These values are recorded generally accepted critical periods for biological growth and ­ from upstream districts exploiting the natural flow of rivers and the fact that pollution flows down- ­ ealth. In this way, we stream even as the decision to pollute upstream is orthogonal to downstream h exploit quasi-random variation in pollution that originates upstream and yet flows downstream to districts. The analysis then uses these spillovers to ascertain how much of the health impact other ­ incidence. persists in the next district downstream of pollution ­ The analysis compares later-life height among cohorts with duration of nitrate-nitrogen exposure, state-trends. In this way, the analysis controlling for birth year, birth month, district fixed effects and ­ exploits within-district variation in birth timing relative to pollution exposure to identify b. The birth-year and birth-month fixed effects (rmt) are included to account for age effects in health ­ outcomes as well as unobserved national or seasonal shocks such as macroeconomic conditions or ­seasonal weather patterns, which might otherwise confound the relationship between pollution expo- ­ eight. Similarly, district fixed effects are included to control for any time-invariant unob- sure and h health. For example, access to local nutrition servable differences between districts that can affect ­ ­ocation. The programs is one such factor that may be constant across individuals born in the same l analysis also includes state-trends (rst) to flexibly control for heterogeneous changes in demographic ­ tates. factors, technological progress in agriculture and other policies that differ across s ­ nalysis. Xdst are a v A number of other district and household specific variables are included in the a ­ ector of district time-varying variables (include temperature and precipitation and concentrations of other water quality indicators like fecal ­coliform). Dids w are controls for household characteristics such as religion 8 The Nitrogen Legacy ­ ontext. Lastly, we use cluster-robust standard errors to account for and caste that are salient to the Indian c time. Our baseline within-district clustering of errors and arbitrary correlation of observations across ­ specification, therefore, compares two women from the same district who are subjected to different levels of nitrate-nitrogen exposure based on their year of birth, over and above any unobserved shocks to height that vary by the year of birth, and any long-run trends (or annual patterns) in height in the state of b ­ irth. Thus, following established statistical methods in applied economics, the relationship between water pollution and height ( b ) is identified by removing any confounding differences attributable to location time. The reduced-form relationship provides a causal estimate of the health damages caused by and ­ al. (2018), downstream spillovers of pollution, adding to related work on pollution spillovers by Do et ­ Garg et a ­ l., (2018), Keiser and Shapiro (2018), Lipscomb and Mobarak (2017) and Sigman (2002, 2005). Further, since the identification strategy uses multiple exposure events over time and space, it allevi- ates concerns that the results are being driven by confounding factors to health that may be correlated with single ­ events. So far, our estimation strategy allows us to quantify the persistence of water quality impacts in down- spillovers. We are also interested in the within-district stream districts by focusing on downstream ­ externality: to what extent does nitrogen pollution within a given district affect health outcomes in the same district? To address this question, we instrument local pollution concentrations in a given location ­ oncentrations. This effectively uses variations in local water quality that are and time with upstream c concentrations. The validity of this approach rests on the assumption induced by exogenous upstream ­ that river flow is unidirectional and pollution from far away distances affects health, but only through concentrations. The first-stage (equation (2)) and second stage (equation (3)) its effect on local pollution ­ below. of the two-stage least squares strategy are presented ­ [birth 3 ] [birth , 3] N idt = α ′ + β ′ N idt U, + λ ′ Xdt + γ ′ Did w ′ + ρ s t ′ + µd ρ mt ′ + ∈idt (2) ˆ idt Yidt =α + β N [ birth , 3 ] + λ Xdt + γ Did w ρ mt + ρ st µ d + ∈idt (3) Coefficient b in equation (3) gives us the estimated impact of pollution exposure on health in the average ­ ­ district. Together with the spillover health impact in downstream districts estimated in equation (1), we are able to measure the full external health costs imposed by p ­ ­ ollution. It is important to highlight that this paper does not include a structural model that describes the results. Therefore, we interpret our main result as a reduced form rela- mechanism(s) for our baseline ­ health. We provide discussion on the possible mecha- tionship between nitrogen pollution and adult ­ nisms in Section 7. 4. Results Summary statistics are provided in Table 1. About 3% of the sample experienced high levels of nitrate-ni- trogen pollution in water (exceeding 10 mg/l) in the year of birth and on average women were exposed ­ hree. Table 2 presents the main to high levels of nitrate-nitrogen pollution for 2% of their lives up to age t specification. We results from estimating equation (1). Column (1) presents results from the preferred ­ find that exposure to nitrate pollution that exceeds safety standards over the entire period decreases The Nitrogen Legacy 9 TABLE 1. Descriptive Statistics Variable Mean ­Std. ­Dev. Min Max Height (cm) 151.584 6.315048 80 209.2 Mean upstream nitrate-N concentrations 1.776662 2.887341 0 20.18191 1[Exceedance of nitrate-N in year of birth] 0.0257691 0.1584502 0 1 Fraction of early childhood nitrate-N exposure 0.0190851 0.0732628 0 0.5 Annual Precipitation (mm) 882.8155 500.6614 99.29888 4463.666 Average temperature in the Wet Season 28.57736 1.785982 22.61299 32.91105 Average temperature in the Dry Season 20.75181 2.642523 13.20256 27.45295 CWC. Sample based on 19,138 Notes: Table shows descriptive statistics from the DHS surveys as well as the water quality data from ­ respondents who have not migrated from their birth ­ place. TABLE 2. The Long-Term Impacts of Upstream Pollution on Health Dependent variable: Height (cm) (1) (2) (3) (4) (5) (6) Fraction early childhood N −2.246*** −3.044*** −1.963*** exposure (0.497) (0.996) (0.506) Exposure in-utero 0.541 (0.463) Exposure at birth −0.385 (0.458) Exposure at age 1 −0.411 (0.392) Observations 19138 17399 17618 17417 13862 19138 mean Dependent Variable 151.6 151.6 151.6 151.7 151.4 151.6 R-sq 0.0793 0.0812 0.0795 0.0769 0.0656 0.0908 RMSE 6.082 6.076 6.093 6.114 6.089 6.046 Birth-year Fixed Effects Y Y Y Y Y Birth-Month Fixed Effects Y Y Y Y Y District Fixed Effects Y Y Y Y Y Y State Trends Y Y Y Y Y Y Weather controls Y Y Y Y Y Y Fraction early childhood FColi Y exposure Birth-Year by Month Fixed Y Effects Eq. (1) via ordinary least squares ­ Notes: Table shows results from estimating ­ regression. (OLS). Each column displays estimates from a separate ­ Fraction of early childhood exposed to N pollution is the fraction of years from year of birth to age 3 that nitrate pollution exceeds safety guidelines. Standard errors are clustered at the district level, and are presented in ­ ­ parentheses. ***, **, * denote statistical significance at respectively. the 1%, 5% and 10% levels ­ 10 The Nitrogen Legacy ­ entimeters. At the mean fraction of early life exposed to pollution, this decrease in height by 2.24 c centimeters. Columns (2), (3) and (4) use an indicator for high-level exposure in-utero, in height is 0.5 ­ exposure. The results show a lowering the birth year and at age one rather than a cumulative measure of ­ of height with exposure but these effects are not significant compared to the effect from cumulative exposure in column (1). Column (5) includes an indicator for whether concentrations of fecally derived bacteria related to poor sanitation from upstream locations are above desired limits from the time of birth to age three, to confirm that it is not this correlated water quality indicator that is driving the result. Exposure to nitrogen pollution continues to be statistically significant, and the magnitude is ­ even ­higher. This suggests that exposure to nitrogen pollution matters for health in addition to exposure bacteria. In Column (6), stricter control of birth month-by-birth year fixed effects from excreta-related ­ are included to control for unobserved factors that are constant across all individuals born in the same month. Results are qualitatively ­ year and ­ similar. The results show that exposure to nitrate pollution ­ entimeters. that exceeds safety standards over the entire period decreases height by 1.96 c So far, the results have focused on the persistence of water quality impacts in downstream areas by pollution. In Table 3, we provide the measuring the direct spillover externality imposed by upstream ­ estimated impact of the within-district externality by measuring the impact of pollution on health in the same district using the 2SLS procedure outlined in equations (2) and (3). TABLE 3. The Long-Term Impacts of Local Pollution on Health using Instrumental Variables (1) (2) (3) (4) First-stage Second-stage First-stage Second-stage Upstream: Fraction early childhood N exposure 0.748*** 0.745*** (0.166) (0.163) Local: Fraction early childhood N exposure −2.819*** −2.604*** (0.645) (0.634) Observations 17755 17755 17755 17755 mean Dependent Variable 151.6 151.6 R-sq 0.0186 0.0187 RMSE 0.0407 5.956 0.0407 5.921 Kleibergen-Papp F-stat 20.34(F=16.38) 20.94(F=16.38) Birth-year Fixed Effects Y Y Birth-Month Fixed Effects Y Y District Fixed Effects Y Y Y Y State Trends Y Y Y Y Weather controls Y Y Y Y Birth-Year by Birth-Month Fixed Effects     Y Y Eq. (2) and ­ Notes: Table shows results from estimating ­ SLS). Each column displays estimates from a Eq. (3) using Two-Stage Least Squares (2­ regression. Fraction of early childhood exposed to N pollution is the fraction of years from year of birth to age 3 that nitrate pollution separate ­ guidelines. Columns 2 and 4 show 2nd-stage results and columns 1 and 3 show 1st-stage ­ exceeds safety ­ results. The endogenous analog. For Kleibergen-Paap rkWald F Stat, Stock-Yogo variable(Local: Fraction early childhood N exposure) is instrumented using its upstream ­ parentheses. Critical value for 15% maximal instrumental weak identification critical value for 10% maximal instrumental variable size in ­ parentheses. ***, **, * denote statistical variable size equals 8.96. Standard errors are clustered at the district level, and are presented in ­ respectively. significance at the 1%, 5% and 10% levels ­ The Nitrogen Legacy 11 significant. When The first-stage is strong across all columns and the upstream concentrations are ­ local concentrations are instrumented with upstream pollution levels in the second-stage, all specifica- tions yield statistically significant estimates of the pollution impact and the effect of nitrogen pollution magnitude. Diagnostic statistics for instrument relevance in water on height is negative, and large in ­ such as the Kleibergen-Paap F (Kleibergen and Paap, 2006) statistic shows that the instrument is very ­ strong. The F-statistic exceeds the Stock-Yogo (Stock and Yogo, 2005) weak identification critical value size.6 The point estimates from the 2SLS procedure are rela- for 10% maximal instrumental variables ­ tively much larger than the corresponding downstream spillover impact in Table 2 supporting the logic that as water pollution decays with river flow and time, the downstream impacts are likely to be smaller ­mpact. The results show that exceeding the nitrate-­ in magnitude than the within-district health i ­ entimeters. nitrogen safety standards over the entire period decreases height by 2.81 c Because adult height is associated with income, this implies a productivity loss of around 7% using decrease in height estimates under full exposure derived from column 1 and using estimates of the eco- nomic returns to height assumed in the World Bank Human Capital Project (Kraay, 2018). When using estimates of a decrease in height under mean exposure derived from column 1, this translates into a 1.7% potential.7 fall in productivity or earning ­ 5. Robustness Checks ­ esults. We carry out several robustness exercises to further corroborate our baseline r In order to examine the possibility that these results are driven by spurious spatial or temporal patterns, the analysis is subjected to falsification ­ ­ tests. The first test involves re-estimating equation (1) while replacing each individual’s exposure condition with exposures that occur for 6 different four-year periods before or after birth up to age 9. The resulting coefficient estimates are plotted in Figure 3 against exposure. All the “shifted” coefficients are smaller than the “true” the different window periods of ­ coefficient, plotted at 0-3, and are all statistically ­ ­ insignificant. The second test involves replacing the upstream pollution variable with a falsified value using pollu- tion data from the nearest off-river region farther downstream—a location that is disconnected from river flow dynamics and from where the pollution cannot flow (upstream) to areas where the health ­ easured. In the case that the ‘falsified’ upstream pollution variable shows a significant outcomes are m impact on health, then it would be likely that our baseline results are capturing spurious spatial correlations. Table 4, however, reveals ­ ­ otherwise. There is no significant impact of the falsified value on health suggesting that the upstream variable utilized in the analysis is indeed isolating quasi-random pollution. variation in ­ In Table 5, in addition to the district fixed effect we also include district time trends to address the results. The results are of concern that broad secular trends at the district level might be influencing our ­ ­ evel. the same sign and magnitude as our baseline estimates, and remain significant at the 5 percent l In Table 6, we cluster the standard errors by state in DHS, as well as survey cluster in DHS, instead of ­district. Standard errors are more or less similar using either of these alternatives, and the results remain unchanged. significant and ­ 12 The Nitrogen Legacy FIGURE 3. Different Window Periods of Exposure 4 Coe cient on fraction childhood exposure 2 0 –2 –4 L6–L3 L5–L2 L4–L1 L3–0 L2–1 L1–2 0–3 1–4 2–5 3–6 4–7 5–8 6–9 Woman’s height (cm) Note: Estimated coefficients from variants of the main regression equation, in which the period of pollution exposure is shifted by 6 four-year period. Each marker’s vertical position therefore measures the estimated impact of exposure at periods (horizontal axis) from the main 0-3 ­ exposure. For example, the purple marker represents the impact of exposure discussed in the ­ the appropriate period of ­ report. Other markers exposures. Error bars represent 95% confidence ­ represent the impact of “placebo” ­ intervals. TABLE 4. Falsification Test Placebo Districts (1) (2) Fraction childhood N exposure 0.106 −0.010 (0.430) (0.413) Observations 23338 23338 R-sq 0.0773 0.0683 RMSE 5.796 5.821 Birth-year Fixed Effects Y Y Birth-Month Fixed Effects Y Y District Fixed Effects Y Y State Trends N Y Weather controls Y Y Notes: Columns 1 and 2 show results from a placebo test, in which the upstream district for each observation is replaced by a different, upstream. Each column displays estimates from a separate ­ neighboring district that is not ­ regression. Fraction of early childhood exposed to guidelines. Standard errors are clustered at N pollution is the fraction of years from year of birth to age 3 that nitrate pollution exceeds safety ­ parentheses. ***, **, * denote statistical significance at the 1%, 5% and 10% levels ­ the district level, and are presented in ­ respectively. The Nitrogen Legacy 13 TABLE 5. Main Results with Trends for Districts Dependent variable: Height (cm) (1) (2) Fraction childhood N exposure −2.273** −2.392*** (0.887) (0.891) Observations 19450 19138 R-sq 0.0846 0.0835 RMSE 6.047 6.064 Birth-year Fixed Effects Y Y District Fixed Effects Y Y District Trends Y Y Weather controls N Y Eq. (1) via ordinary least squares ­ Notes: Table shows results from estimating ­ regression. (OLS). Each column displays estimates from a separate ­ Fraction of early childhood exposed to N pollution is the fraction of years from year of birth to age 3 that nitrate pollution exceeds safety guidelines. Standard errors are clustered at the district level, and are presented in ­ ­ parentheses. ***, **, * denote statistical significance at the respectively. 1%, 5% and 10% levels ­ TABLE 6. Alternative Clustering Dependent variable: Height (cm) (1) (2) Fraction childhood N exposure −2.246 −1.963 s.e. clustered by district ­ (0.497)*** (0.506)*** s.e. clustered by state ­ (0.552)** (0.488)** s.e. clustered by survey cluster ­ (0.928)*** (0.944)*** Observations 19138 19138 R-sq 0.0793 0.0908 Birth-year Fixed Effects Y Birth-Month Fixed Effects Y District Fixed Effects Y Y State Trends Y Y Weather controls Y Y Birth-Year by Month Fixed Effects Y Eq. (1) via ordinary least squares ­ Notes: Table shows results from estimating ­ regression. (OLS). Each column displays estimates from a separate ­ Fraction of early childhood exposed to N pollution is the fraction of years from year of birth to age 3 that nitrate pollution exceeds safety guidelines. Standard errors are clustered at the district level, state level and survey cluster level and are presented in ­ ­ parentheses. respectively. ***, **, * denote statistical significance at the 1%, 5% and 10% levels ­ 6. Evidence from Other Regions In Vietnam, home to one of the fastest-growing and urbanizing societies in the world, agricultural ­ evelopment. But in parts of the growth and intensification have played significant roles in spurring d deepening. In intensively farmed country, the environmental footprint of the agricultural sector is ­ pollution. This is particularly so in the areas, agriculture has become a significant contributor to water ­ intensively farmed Mekong delta region (Cassou, Jaffee, and Ru 2018; Chea, Grenouillet, and Lek 2016). 14 The Nitrogen Legacy To measure the consequences of nitrogen pollution, the analysis focuses on children aged four to twelve years surveyed in the latest Living Standards Survey of 1997–98. Table 7 shows that repeated exposure to nitrate pollution for the first three years of life substantially lowers height-for-age scores in ­ eviations. These childhood, with full exposure decreasing height-for-age scores by 0.7 standard d effects occur despite nitrate-nitrogen concentrations being below the recommended safety thresholds ­ ontaminants. of 10 mg/L and emerge even after accounting for exposures from other c ­ rowing. Other sources In Africa, although present-day fertilizer usage is lower than in Asia, it is g of nitrate exposure include expanding urban centers that lack wastewater treatment facilities f arming. The analysis is based on the entire universe of child records up to and increased livestock ­ records. The results in Table 8 show that in age 5 years across 33 countries in Africa from DHS ­ utero exposure to nitrate pollution emanating from upstream urban agglomerations lowers the TABLE 7. Impacts in Vietnam Height-for-age scores (1) (2) Fractional Exposure to Nitrate-Nitrite −0.776** −0.779** (0.338) (0.337) Birth-year Fixed Effects Y Y Birth-Month Fixed Effects Y Y Commune Fixed Effects Y Y Province Trends Y Y Other controls N Y N 691 691 R-sq 0.132 0.156 level. Notes: Statistical significance is given by * p<0.10 ** p <0.05 ***p < 0.01. Standard errors in parentheses are clustered at the commune ­ Other controls include precipitation, temperature, ethnicity (tribe), sex, conductivity, phosphorus, water-treatment at home, household asset value, years of education of head, farm/non-farm household. TABLE 8. Impacts in Africa In-utero exposure Stunting HAZ Stunting HAZ Downstream of N pollution 0.0172*** −0.0729*** (0.00636) (0.0228) Downstream of N pollution x Rural 0.0209*** −0.0848***   (0.00597) (0.0222) Fixed effects Year-Month of Birth, Grid Cell Other controls Y Y Y Y N 204,886 204,886 204,886 204,886 R-Sq 0.106 0.143 0.106 0.144 level. Other Notes: Statistical significance is given by * p<0.10 ** p <0.05 ***p < 0.01. Standard errors in parentheses are clustered at gridcell ­ controls include household variables – if it is in a rural location, indicator for improved sanitary facilities, improved water source and no sanitation facility (open defecation), child age in months, age of mother at birth giving, if child is a girl, a household wealth index, body mass index (BMI) of mother, an index of mother empowerment (health decisions), mother’s years of education and mother’s partner’s years of education – and community variables – percentage of improved water source, improved sanitation and open defecation, and total population of trend. urban area; temperature and precipitation; and year specific country ­ The Nitrogen Legacy 15 height-for-age scores and increases the likelihood of stunting for children younger than five years, ­ xposure. The negative effects are most pronounced downstream from even at low levels of nitrate e ­ igher. Stunting already remains a widespread urban centers where nitrate levels are relatively h problem in Sub-Saharan Africa, where more than 35 percent of children younger than five years ­ ndicators). This suggests an urgent need for potable are considered stunted (World Development I agglomerations. water treatment in urban ­ 7. Mechanisms These results are perhaps the first demonstration of such widespread links between exposure to ­ utcomes. Nevertheless, they are con- elevated nitrate levels during early-life and long-run health o sistent with several well-established streams of biomedical literature that are indicative of such a ­ link. First, increased dietary-nitrate intake has been associated with hypothyroidism and thyroid ­ l. 2012; Ward et a cancer (Aschebrook-Kilfoy et a ­ l. 2010, 2018). The thyroid is an important gland for ­ egulation. Hypothyroidism in children is there- regulating hormone production and metabolism r fore linked to stunting of growth and a delay in the process of maturation (Wilkins 1953). Thus, the path from increased nitrate consumption from water, to diseases of the thyroid, to stunted growth direct. Another potential causal link between nitrates in and development is seemingly clear and ­ water. water and reduced health and growth is through the buildup of algae and bacteria in ­ blooms. These bacteria can emit Nitrogen in waterways often causes cyanobacteria fueled algal ­ illnesses. cyanotoxins that are toxic to humans and, if consumed, can lead to diarrhea-related ­ Repeated bouts of diarrhea increase the probability of nutritional deficiencies in children and thus development. Exposure to such toxins can also adversely affect birth outcomes by stunted child ­ lowering infant birthweight (Jones 2019), an important predictor of stunting later in childhood al. 2013). (Christian et ­ Finally, and related to the prior point, exposure to higher levels of pathogens can disrupt the gut ­ microbiome. The first months after birth are particularly critical for establishing the composition of the ­ l. 2019). There is evidence in gut microbiome that persists for the rest of a person’s life (Robertson et a the medical literature that this microbiome is difficult to permanently change later in life, although this debate. If true, then the resulting change in gut microbiome from exposure to nitrate-­ matter is under ­ induced toxins like those from cyanobacteria could permanently handicap the digestive system of indi- viduals and reduce their capacity to absorb nutrients throughout their l ­ ives. However, more research is required on how and when exposure of fetuses and young children to high nitrate levels influence microbiome function, growth, and development, especially in settings in which pathogenic infections problematic. and food insecurity are ­ 8. Conclusion Recent studies have focused attention on the loss of human lives and immediate birth outcomes as a ­ ollution. In a departure from previous studies, this paper underscores the long-lasting result of water p ­ ndure. We find a health damages, and decreased economic capability that survivors of water pollution e 16 The Nitrogen Legacy statistically significant negative effect of early-life exposure to nitrogen pollution on women’s height in India with supporting evidence of a decrease in child height in Vietnam, and infant HAZ scores and Africa. These results are robust to several checks for confounding increased incidence of stunting in ­ factors. By demonstrating the long-term effects of nitrogen pollution, our results draw attention to the ­ critical role that local environmental spillovers play for population health outcomes, and highlight the pollution. need for closer policy attention to nitrogen ­ The policy relevance of our results is underscored by the fact that health effects also emerge at levels well below prescribed limits, raising questions about what constitutes safe standards for nitrates in water. Emerging evidence from epidemiological studies have also found relationships between nitrate ­ ingestion and cancer, thyroid disease, and adverse pregnancy outcomes, such as neural tube defects, at al., forthcoming; Ward et ­ concentrations below regulatory limits (Temkin et ­ al. 2018). Even as far back U.S. National Academy of Sciences warned that “there is little margin of safety” as 1977, a report by the ­ in the 10 mg/L safety limit for nitrates (National Research Council 1977). More research and assessments claims. It is possible that across even more geographies and populations are needed to make definitive ­ future research will uncover even more health effects as more data becomes available to link exposures today. However, the body of evidence so far suggests that began decades ago to diseases that develop ­ that there still remains a great deal of uncertainty surrounding drinking water standards for nitrates set ­ gencies. The magnitude of people impacted by nitrate contaminated water is, there- by environmental a thought. fore, likely to be much larger than presently ­ This work also speaks to the consequentiality of fertilizer subsidies in developing countries that are ­ se. In India, a system of domestic price controls by way of tipped in the favor of nitrogen fertilizer u large subsidies has significantly distorted market prices for nitrogen fertilizer compared to other nutrients resulting in an inefficient balance of fertilizer application (Gulati and Banerjee 2015). By 2015, subsidy costs amounted to $11.6 billion per year in India, roughly five times more than what was recorded 15 years earlier (Gulati and Banerjee 2015). This is exemplified by the wide gap between global and Indian domestic nitrogen prices ---world prices were almost four times higher than regu- lated Indian prices in 2014 (Huang, Gulati, and Gregory 2017). In recent years the government has made efforts to improve nitrogen use efficiency in agriculture and has has mandated urea manufac- turers to produce neem-coated ­urea. Since neem acts as a nitrification inhibitor, it allows a more grad- ual release of nitrogen into the soil thereby improving nitrogen use ­efficiency. More research is needed to quantify the environmental and economic consequences of such measures, and its impacts on water ­pollution. Finally, unlike much of the literature on water quality and health that focuses on developed coun- tries, this work adds to the growing evidence on water pollution impacts in the developing world that is subject to different exposure profiles, institutions and levels of economic ­ development. While our analysis controlled for correlated pollutants where possible, it was primarily focused on a single ­ pollutant. It is possible that the combined health impacts of co-occurring pollutants are different or al. 2019). More work is needed to investigate these issues in the develop- even more harmful (Stoiber et ­ ing ­world. The Nitrogen Legacy 17 Notes 1. For instance, even if runoff of nitrogen was fully stemmed, it will still take 30 years to realize the 60% decrease in load needed to reduce eutrophication in the Gulf of Mexico (Van Meter, 2018). 2. This health hazard was responsible for triggering the creation of drinking water standards for nitrates at 10 parts per ­m illion. Note that 10 mg/L as nitrate-nitrogen (NO3-N) is approximately equivalent to the World Health Organization (WHO) guideline of 50 mg/L as NO3. 3. A number of biomedical and epidemiological studies in the United States and other countries have documented a relationship between agrichemical exposure and birth defects such as Down’s syndrome and Spina Bifida, especially for children conceived during the crop-sowing months, and among children of agrichemical applicators who are consistently exposed to toxins 4. Indeed, even in the US the percentage of public water systems that have violated safety limits for nitrates in drinking water have increased in the 15 year period between 1994 and 2009, due to the difficulty of coping with the rising nitrate pollution and the concomitant rise in the al. 2018). costs of water treatment (Ward et ­ 5. In Senegal, studies have recorded nitrate-nitrogen levels going beyond 40 mg/l, more than 4 times the safety limit for NO3­-N. Extremely high levels of nitrate have also been reported in The Gaza Strip, where nitrate reached concentrations of 500 mg/L NO3 in some areas (10 times the safety limit for NO3), and more than 50% of public-supply wells had nitrate concentrations above 45 mg/L NO3. Other site-specific studies in India have found nitrates in drinking water supplies to be particularly high in rural areas, where average levels are reported to be between 46 mg/L NO3 and 66.6 mg/L NO3 with maximum levels exceeding 100 mg/L NO3 in several r ­ egions. al. (2007) and Bazzi and Clemens (2013) provide explanations of these t 6. Baum et ­ ­ ests. 7. As a robustness check, we also make use of the Indian Human Development Survey (IHDS) to measure the impact of early-life exposure to nitrogen pollution on later-life wages using a similar methodology described in Section 2. The IHDS is a nationally representative survey. The survey provided a more complete recording of men’s earnings that we use as our main variable of i ­ ­nterest. In unreported results, we find that full exposure in early-life decreases wages in adult life by 9%, on average, providing direct proof of the impact on labor ­productivity. References D., and ­ Almond, ­ Currie. 2011. “Killing Me Softly: The Fetal Origins ­ J. ­ Hypothesis.” Journal of Economic Perspectives 25 (3): 153–72. D., ­ Almond, ­ ­. ­ J. Currie, and V ­I.” Journal of Economic Literature 56 (4): Duque. 2018. “Childhood Circumstances and Adult Outcomes: Act I 1360–446. Aschebrook-Kilfoy, ­ S. ­ B., ­ J. ­ L. Heltshe, ­ M. ­ R. Nuckols, ­ M. Sabra, ­ A. ­ R. Shuldiner, ­ D. Mitchell, ­ B. ­ T. ­ M. Airola, ­ Y. Zhang, and R. Holford, ­ ­ H. ­ M. ­ Ward. 2012. “Modeled Nitrate Levels in Well Water Supplies and Prevalence of Abnormal Thyroid Conditions among the Old Pennsylvania.” Environmental Health 11 (1): 6. Order Amish in ­ Barker DJ (1990) The fetal and infant origins of adult ­ disease. BMJ 301:1111. D. ­ Barker, ­ disease. Bmj, 311(6998), 171-174. J. (1995). Fetal origins of coronary heart ­ Brainerd, Elizabeth and Menon, Nidhiya, 2014. Seasonal effects of Water quality: the hidden costs of the Green revolution to infant and child India. Journal of Development ­ health in ­ Economics. 107, 49–64. E., ­ Cassou, ­ M. Jaffee, and ­ S. ­ Ru. 2018. The Challenge of Agricultural Pollution: Evidence from China, Vietnam, and the ­ J. ­ Philippines. Bank. Washington, DC: World ­ R., ­ Chea, ­ ­ .­ G. Grenouillet, and S ­ ap.” Lek. 2016. “Evidence of Water Quality Degradation in Lower Mekong Basin Revealed by Self-Organizing M PloS ONE 11 (1): e0145527. P., ­ Christian, ­ ­ .­ S. E, Lee, M ­ .­ D. Angel, L ­.­ S. Adair, S E. Arifeen, P ­ . Ashorn, F ­.­ ­ .­ C. Barros, C ­ .W. Fawzi, W H. Fall, W ­ .­ ­ . Hao, and G Hu. 2013. “Risk of Countries.” International Childhood Undernutrition Related to Small-for-Gestational Age and Preterm Birth in Low- and Middle-Income ­ Journal of Epidemiology 42 (5): 1340–55. J. and Vogl, T Currie, ­ ­ . 2013. Early-life health and adult circumstance in developing countries, Annual Review of Economics, Annual Reviews, vol. 5(1), ­ ­ pp. 1–36, 05 18 The Nitrogen Legacy S. 2004. Global nitrogen: cycling out of ­ Fields, ­ control. Environ Health Perspect 112(10):A556-63. R., Desbureaux, ­ Damania, ­ A. ­ S., Rodella, ­ J., & Zaveri, ­ S., Russ, ­ Crisis. World Bank: E. (2019). Quality Unknown: The Invisible Water ­ Washington, ­DC. Do, Quy-Toan, Shareen Joshi, and Samuel ­Stolper. 2018. Can environmental policy reduce infant mortality? Evidence from the Ganga Pollution ­ Cases. Journal of Development Economics 133 : 306-325. R., Carrillo, ­ Fishman, ­ J. (2019). Long-term impacts of exposure to high temperatures on human capital and economic p P., & Russ, ­ ­ roductivity. Journal of Environmental Economics and Management, 93, 221-238. Garg, Teevrat, Hamilton, Stuart, Hochard, Jacob, Plous, Evan, Talbot, John, 2016. Not So Gently Down the Stream: River Pollution and Health Indonesia. Grantham Research Institute on Climate Change and the ­ in ­ Environment. Working Paper #234. A., and ­ Gulati, ­ Banerjee. 2015. “Rationalizing Fertilizer Subsidy in India: Key Issues and Policy ­ P. ­ Options.” Working Paper 307, Indian Council Delhi. ­ for Research on International Economic Relations, New ­ /Rationalizing_Fertilizer_Subsidy​ https://www.academia.edu/14970315­ _in_India-Key_Issues_and_Policy_Options. J., ­ Huang, ­ I. Gregory, ­ A. Gulati, and ­ report. ­ editors. 2017. Fertilizer Subsidies: Which Way Forward? IFDC/FAI ­ https://ifdcorg.files.wordpress. com/2017/02/fertilizer-subsidieswhich-way-forward-2-21-2017­.pdf. M., & Russ, ­ Hyland, ­ ­ frica. World Development, 115, 30-45. J. (2019). Water as destiny–The long-term impacts of drought in sub-Saharan A (INA). 2017. Eds Abrol, ­Y. ­P. et ­al. The Indian Nitrogen Assessment ­ B. ­ Jones, ­ Experiment.” Journal of A. 2019. “Infant Health Impacts of Freshwater Algal Blooms: Evidence from an Invasive Species Natural ­ Environmental Economics and Management 96: 36–59. D. ­ Kanter, ­ ­ hange.” Climatic Change 147 (1–2): 11–21. R. 2018. “Nitrogen Pollution: A Key Building Block for Addressing Climate C B. ­ Keeler, ­ J. ­ L., ­ S. Polasky, ­ D. Gourevitch, ­ F. Isbell, ­ C. ­ W. Tessum, ­ D. Hill, and J J. ­ D. ­ ­. ­ ­ itrogen.” Science Marshall. 2016. “The Social Costs of N Advances 2 (10): e1600219. ­ ata. 2019. J Assoc Environ Resour Econ 6 (4): 669-707. Keiser D, The missing benefits of clean water and the role of mismeasured pollution d D.A., and J Keiser, ­ Shapiro. 2018. “Consequences of the Clean Water Act and the Demand for Water Q ­.S. ­ ­ uality.”  Quarterly Journal of Economics 134 (1): 349–96. ­ aap. 2006. Generalized reduced rank tests using the singular value d Kleibergen, Frank and Richard P ­ ecomposition. Journal of Econometrics 133 (1): 97–126. ­ .. 2018. Methodology for a World Bank Human Capital Index ( Kraay, Aart C ­ o. WPS 8593. Washington, ­ English). Policy Research working paper; n D.C. : World Bank ­ ­ Group. ­ ., and Jay M Peterson, Erin E Hoef. 2010. A mixed-model moving-average approach to geostatistical modeling in stream n ­ . Ver ­ ­ etworks. no. 3: 644-651. Ecology 91, ­ A., 2017. Decentralization and Water pollution spillovers: evidence from the redrawing of county bound- Lipscomb, Molly, Moshfiq Mobarak, ­ Brazil. ­ aries in ­ Econ.Stud. 84 (1), 464–502. Rev. ­ S. ­ Maccini, ­ D. (2009). Under the weather: Health, schooling, and economic consequences of early-life r L., & Yang, ­ ­ainfall. American Economic Review, 99(3), 1006–1026. www.africapolis.org (accessed 05 February 2019) OECD/SWAC (2018), Africapolis (database), ­ A. ­ Ravelli, ­ J. ­ C., van der Meulen, ­ R. ­ H., Michels, ­ P. ­ J., Osmond, ­ C., Barker, ­ J., Hales, ­ D. ­ N., & Bleker, ­ C. ­ P. (1998). Glucose tolerance in O. ­ famine. The Lancet, 351(9097), 173-177. adults after prenatal exposure to ­ K. ­ Risnes, ­ L. ­ R., Vatten, ­ J., Baker, ­ J. ­ L., Jameson, ­ K., Sovio, ­ E., ­ U., Kajantie, ­ ­ . (2011). Birthweight and mortality in adulthood: ... & Sundh, V meta-analysis. International journal of epidemiology, 40(3), 647-661. a systematic review and ­ Am. ­ Sigman, Hilary, 2002. International spillovers and Water quality in rivers: do countries free ride? ­ Rev. ­ Econ. ­ Proceed. 92 (4), Pap. ­ 1152–1159. The Nitrogen Legacy 19 ­ olicies. ­ Sigman, Hilary, 2005. Transboundary spillovers and decentralization of environmental p Environ. ­ J. ­ Manag. 50, 82–101. Econ. ­ W., ­ Steffen, ­ K. Richardson, J ­ . Rockstrom, S ­ .E. Cornell, ­ I. Fetzer, E R. Biggs, et ­ ­ .M. Bennett, ­ al. 2015. “Planetary Boundaries: Guiding Human Planet.” Science 347: 736. Development on a Changing ­ H. and Motohiro ­ Stock, James ­ Regression. In Identification and Inference for Yogo. 2005. Testing for Weak Instruments in Linear IV ­ ­ d. Donald W Econometric  Models: Essays in Honor of Thomas Rothenberg, e K. Andrews and James H ­ .­ Stock. Cambridge University ­ .­ ­Press. T., ­ Stoiber, ­ D. Andrews, ­ A. Temkin, ­ O. ­ C. Campbell, and ­ Naidenko. 2019. Applying a Cumulative Risk Framework to Drinking Water V. ­ Commentary.” Environmental Health 18 (1): 37 Assessment: A ­ A., ­ Temkin, ­ S. Evans, ­ T. Manidis, ­ C. Campbell, and O ­ .­ Naidenko. ­ V. ­ Forthcoming. “ Exposure-Based Assessment and Economic Valuation of ­ ater.” Environmental ­Research. Adverse Birth Outcomes and Cancer Risk Due to Nitrate in United States Drinking W K. ­ Van Meter, ­ P., & Basu, ­ J., Van Cappellen, ­ B. (2018). Legacy nitrogen may prevent achievement of water quality goals in the Gulf of N. ­ ­Mexico. Science, 360(6387), 427-430. L. 1953. “Disturbances in ­ Wilkins, ­ Growth.” Bulletin of the New York Academy of Medicine 29 (4): 280. Ward, ­ R. Jones, ­ M., ­ J. Brender, ­ T. de Kok, ­ P. Weyer, ­ B. Nolan, ­ M. Villanueva, and ­ C. ­ Breda. 2018. “Drinking Water Nitrate and Human S. van ­ Review.” International Journal of Environmental Research and Public Health 15 (7): 1557. Health: An Updated ­ M. ­ Ward, ­ B. ­ H., ­ P. ­ A. Kilfoy, ­ J. Weyer, ­ K. ­ E. Anderson, ­ R. Folsom, and ­ A. ­ R. ­ J. ­ Cerhan. 2010. “Nitrate Intake and the Risk of Thyroid Cancer and Disease.” Epidemiology (Cambridge, ­ Thyroid ­ Mass.) 21 (3): 389. L., Cardenas, M Zheng, ­ B., & Wang, L ­ .­ ­ . (2016). Temperature effects on nitrogen cycling and nitrate removal-production efficiency in bed zones. Journal of Geophysical Research: Biogeosciences, 121(4), 1086-1103. form-induced hyporheic ­ 20 The Nitrogen Legacy Appendix 1 Spatial Stream Network ­ model Water quality downstream directly depends on upstream water quality as well as weather related ­ etween. Models developed by ver Hoef, Peterson variables and anthropogenic activities happening in b and Theobald (2006) and ver Hoef and Peterson (2010) allow to statistically represent these stream framework. They allow dependencies within a network and predict water quality in a spatially valid ­ to flexibly control for spatial auto-correlation between observations belonging to the same river net- work based on stream distances, to take into account accumulation of pollutants as well as their dilution. They present important improvement to classic geospatial models based on Euclidean dis- ­ settings. tances which were to be proved to be biased in such ­ Here, we used the model developed by ver Hoef and Peterson (2010) to fill missing observations in the CWC nitrogen data between 1986 and 2017 where over 60% of the nitrogen observations are miss- ing in this dataset, limiting our understanding of the evolution of water quality over the ­period. More specifically, we used the openSTARS package in R (Kattwinkel and Szcos 2018) to derive a topograph- ­ ndia. First, a Digital Elevation Model from the SRTM mission ically correct stream network for all I India. For computational limits, the original DEM 30 meters was used to derive all streams across ­ ­esolution. Second, the upstream area of each CWC station model was resampled at a 100m r determined. Third, the stream distance between each station belonging to a given network was was ­ c alculated. Fourth, the annual level of rainfall, average temperature, average elevation and average ­ pollution. slope were computed to better account for dilution of ­ ­ l. 2014) to model the determinants of water quality in Then, we used the SSN package (ver Hoef et a stations. The original model developed by the authors is: CWC ­ ni = Xib + Su + Sd + Se + Wg + ¨i Where ni is the nitrogen level in station i, Xi = (Xi1 …Xiq) are environmental covariates defined over the station. Su,Sd,Se, are a set of spatially auto-correlated random variables that mod- upstream area of each ­ network. The main dependency we want to capture is the upstream to els spatial dependence inside a ­ (Su). The authors also provide the possibility to incorporate downstream relation between stations ­ (Se). Following downstream to upstream dependencies (Sd), as well as standard Euclidean relationships ­ common practices in spatial statistics, we assumed exponential spatial dependencies between ­observations. Finally, Wg represents a possible set of fixed effects, such as watershed fixed ­ effects. The model was estimated for each year between 1986 and 2008 – the year of birth of the last woman in the analysis. Years before 1986 were excluded for an insufficient number of observa- DHS data used in the ­ tions (<100). levels. To do so, we create loops to Our objective year is to find the model that predict best nitrogen ­ estimate for each year 93 models that represent all the possible combinations of covariates and spatial ­ ­ trategy. dependencies. Models were validated through a Leave One Out Cross Validation (LOOCV) s ­ riteria. The maximization The final model was chosen based on a Mean Square Prediction Error (MPSE) c The Nitrogen Legacy 21 of the predictive power of the model was achieved by introducing one trick in the original approach: we included as a predictor the average value of nitrate in a station between 1986-2008. For each year, the model was trained on available observations and predictions of nitrate was done for missing ­observations. ­ utcomes. The final dataset was then used to study the long term impact of nitrogen level on health o Additional References E. 2018, openSTARS: open source implementation of the STARS ArcGIS ­ Kattwinkel M, Szocs ­ Seehttps://github.com/MiKatt​ toolbox. ­ /­openSTARS M., Erin Peterson, and David ­ Ver Hoef, Jay ­ distance.” Environmental and Theobald. “Spatial statistical models that use flow and stream ­ Ecological statistics 13.4 (2006): 449-464. al. “SSN: An R package for spatial statistical modeling on stream ­ Ver Hoef, Jay, et ­ networks.” Journal of Statistical Software 56.3 (2014): 1-45. 22 The Nitrogen Legacy SKU W19078