Policy Research Working Paper 10617 Is Natural Capital a Complement to Human Capital? Evidence from 46 Countries Richard Damania Luis Diego Herrera Garcia Hyungju Kim Leonardo Viotti Esha Zaveri Stefanie Onder Chrissie Pantoja Environment, Natural Resources and Blue Economy Global Practice & Sustainable Development Practice Group November 2023 Policy Research Working Paper 10617 Abstract The environment has long been the foundation of human The paper focuses on the hydrological services provided by flourishing, but its continued degradation is threaten- forests and exploits quasi-random variation in deforesta- ing to reverse recent development gains, especially in tion upstream to assess the impacts on waterborne disease human health. This paper analyzes the possible comple- outcomes for rural households downstream. The results not mentarity between natural and human capital by linking only indicate increases in diarrheal disease incidence among high-resolution deforestation data with health outcomes children under 5 years old, but also offer new evidence of for 0.7 million children across 46 countries. Forest loss is early-life exposure to deforestation on childhood stunting, a often a consequence of economic activities that may confer well-known indicator of later-life productivity. A case study market and other benefits. At the same time, it can adversely for Peru shows similar results for diarrheal disease, but a affect the provision of forest ecosystem services and reduce weaker effect of forest loss on stunting. The paper concludes the associated socioeconomic and environmental benefits that maintaining natural capital has the potential to gener- for rural communities. The net effect is thus ambiguous. ate meaningful improvements in long-run human capital. This paper is a product of the Environment, Natural Resources and Blue Economy Global Practice and the Office of the Chief Economist, Sustainable Development Practice Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at rdamania@worldbank.org, lherreragarcia@worldbank.org, hkim32@worldbank.org, leonardoviotti@gmail.com, ezaveri@worldbank.org, sonder@worldbank.org, and chrissiepantoja@gmail.com. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Is Natural Capital a Complement to Human Capital? Evidence from 46 Countries Damania, Richard* ⓡ Luis Diego Herrera Garcia* ⓡ Hyungju Kim* ⓡ Leonardo Viotti** ⓡ Esha Zaveri*† ⓡ Stefanie Onder* ⓡ Chrissie Pantoja*** JEL classification: I115, Q51, Q53, Q56, Q57 Keywords: Health and Economic Development, Valuation of Environmental Effects, Water Pollution, Sustainability, Ecosystem Services ⓡ The author order was randomized using the author randomization tool of the American Economic Association and has been archived at this link. Affiliations: *The World Bank, Washington, DC ** University of Pittsburgh, Pittsburgh, PA *** Duke University, Durham, NC † Corresponding author 1. Introduction The relationship between the state of the environment and human health has become a key topic in global policy discussions due to growing concerns about the rapid depletion of natural resources, and its implications for human health and eventually sustainable development. The environment has long been the foundation of human wellbeing and its degradation is threatening to reverse recent improvements in human health (Whitmee et al., 2015), with possible long-term effects on labor productivity and income (Pattanayak and Pfaff 2009; Pattanayak et al. 2017). The COVID-19 pandemic has demonstrated the critical need to improve our understanding of the multiple interactions between the environment and humans (FAO, UNEP WHO, and WOAH, 2022; World Bank, 2022), which is especially critical for developing countries. Not only are low- and middle-income economies heavily dependent on natural resources (World Bank 2021), but, on average, they also experience more rapid environmental degradation (FAO, 2020; World Bank, 2021), and have worse health outcomes than higher-income countries (Deaton 2013). Nevertheless, the underlying mechanisms of how natural resource degradation affects human health have yet to be fully explored. In this paper, we focus on one specific type of natural capital – forest ecosystems – which provide a range of socioeconomic and environmental benefits to local and global communities (Chaplin-Kremer et al., 2023; Grammatikopoulou and Vačkářová, 2021; Jenkins and Schaap, 2018; Marta-Pedroso et al. 2014), but have experienced extensive loss and degradation over the last few decades (Hansen et al., 2013, FAO 2020). Forest loss is often a consequence of economic activities that confer commercial and financial benefits. At the same time, it adversely affects the provision of forest ecosystem services and the associated benefits to humans. The net effect is thus ambiguous, requiring an empirical investigation of the underlying causal mechanisms. In this paper, we focus on a forest ecosystem service that has not yet been studied extensively in the literature exploring the relationship between forests and health – namely, hydrological services – and exploit quasi- random variation in deforestation upstream to isolate the impacts on waterborne disease outcomes for rural households downstream. A better understanding of these causal links can shed new light on the possible complementarity between natural and human capital, which can inform the design of environmental, health, and development strategies (Pattanayak et al., 2017), as investments in both nature and health are necessary to sustain economic growth (Johnson et al., 2021; Ogundary & Awokuse, 2018; Weil, 2007; Narayan et al., 2010; Weil, 2014). In addition, if natural and human capital are indeed complementary factors of production, a degradation of natural resources will not only have environmental and health impacts but will also have significant implications for economic development. Recent work suggests that introducing limited substitutability into production functions can limit economic growth (Smulders and van Soest, 2023) as well as a country’s fragility and resilience to external shocks (Karayalcin and Onder, 2023). Several recent studies have explored the causal relationships between forest loss and human health, especially with regard to malaria. Malaria is a pervasive disease that affects populations in tropical and subtropical regions and is transmitted by mosquitos, causing fever and flu-like illness and in some case nausea and diarrhea. While deforestation can increase malaria risk factors in some settings, increased malaria risk does not necessarily translate to higher malaria rates in humans, suggesting that context matters. Bauhoff and Busch (2020) find that the deforestation-malaria relationship does not necessarily generalize across Sub-Saharan Africa. Nevertheless, others find positive relationships at the country level for Brazil (Olson et al., 2010; Terrazas et al., 2015), Paraguay (Wayant et al., 2010), Malaysia (Fornace et al., 2016), Nigeria (Berazneva and Byker, 2017) and Indonesia (Garg 2019; Chakraborty, 2021). 2 However, forest loss can also impact human health by interacting with other forms of natural capital, such as air and water. Air and water quality are critical determinants of human health. According to the World Health Organization (WHO), 94 percent of the world's population suffer direct exposure to unsafe concentrations of particulate matter (PM), a standard indicator of air pollution. Air pollution can harm health in both the short- and long-term (Juginović et al. 2021; Giani et al. 2020; Manisalidis et al., 2020; World Bank, 2016), while air purification can be extremely costly (Cohen et al., 2017; Domingo et al., 2021; Huangfu & Atkinson, 2020; Lelieveld et al., 2015; Ritchie & Roser, 2013). Polluted water can cause acute and chronic diseases, including diarrheal disease (DD) and stunting (Zaveri et al., 2020). Today, DD is among the leading causes of death among children under five (Liu et al., 2012; Prüss-Ustün et al., 2014). Moreover, an estimated 13.5 percent of global stunting prevalence is attributable to DD (Danaei et al., 2016), which is widely recognized as a major impediment to development (Galasso & Wagstaff, 2018). Deforestation and its links to air pollution are often due to forest fires. For instance, prenatal exposure to the 1997 Indonesian forest fires and the resulting exposure to dangerous levels of air pollution have been found to adversely impact child height-for-age scores as well as later life wages for the approximately 1 million Indonesian workers born during this period (Tan-Soo & Pattanayak, 2019). Air pollution generated from Indonesian forest fires can also significantly impact neighboring countries through severe episodes of “seasonal haze”, resulting in increased air-related diseases, including acute respiratory infections (ARI) in Singapore (Sheldon & Sankaran, 2017). Forest fires can also have global impacts. For example, reductions in deforestation rates and associated deforestation fires in the Amazon rainforests, known as “the lungs of the planet”, can significantly reduce PM, preventing thousands of excess deaths (Cobelo et al., 2023; Reddington et al., 2015). However, fewer studies have investigated the causal links between the loss of forest cover and the hydrological services they provide and human health. Empirical studies in forest science and hydrology show that deforestation can impact access to clean water through changes in water yield and quality. Although forests— especially fast-growing young plantations—may use more water than older, mixed forests and therefore reduce freshwater availability, there is growing evidence that, with proper management, forests can help enhance the resilience of water supplies (Miller et al., 2020). Forest loss upstream can also decrease soil infiltration of water and the removal of pathogen, and increase soil erosion (Filoso et al., 2017; El Kateb et al., 2013; Zheng, 2006), which can result in lower stream water loads of Escherichia coli, higher pH of water and lower stream water turbidity (Cunha et al., 2016). Such water conditions can also render point-of-use chlorination less effective especially for households in low-income countries who cannot afford expensive water treatment (Amrose et al 2015). Herrera et al. (2017) use a multi-country database to explore the connections between deforestation, watershed conditions, and childhood DD and find that higher upstream tree cover is associated with a lower probability of DD downstream. This effect is significant for rural households but not urban ones, suggesting differing dependence on watershed conditions. In rural areas, the effect of a 30 percent increase in upstream tree cover is similar in magnitude to the effect of improved sanitation but smaller than for improvement of water sources, wealth, or education. Country-level studies further provide evidence that deforestation reduces access to clean drinking water (Mapulanga and Naito, 2019) and undermines the effectiveness of point-of-use chlorination (Rasolofoson et al., 2021), thus further increasing the risk of DD. Given the importance of upstream forest cover loss to DD, it is possible that upstream forest cover loss could also influence stunting rates through the water channel. Past work has shown that repeated bouts of DD increase the probability of nutritional deficiencies in children, thus stunting child development (Zaveri et al., 2020). Exposure to water pollution can also adversely affect birth outcomes by lowering infant birthweight (Jones, 3 2019), an essential predictor of stunting later in childhood (Christian et al., 2013). However, to the best of our knowledge no evidence yet exists that links the impacts of deforestation to stunting rates. In this paper, we contribute to this emerging literature by taking a closer look at the water-borne channels and the causal effects of forest loss on human health. Such an endeavor is challenging due to the multiple factors determining health outcomes and the degree of complementarity that must be accounted for by including socioeconomic, demographic, infrastructure, environmental, behavioral, and institutional factors (Pattanayak & Pfaff, 2009). We exploit the wealth of data available in the Demographic and Household Surveys (DHS), which includes not only contemporaneous health and socioeconomic outcomes, but also full birth histories for all live births that occur in the years before the survey. We restrict our sample to 0.7 million children from rural areas 1 in 46 countries from 2000-2020 for which we have spatial coordinates. This allows us to link the health outcomes with high-resolution deforestation data and study the effects of forest loss on the incidence of DD (by linking lags of forest loss to the survey year) as well as stunting (by linking lags of forest loss to the birth year). Identifying health impacts of forest loss is challenging given a trade-off between health costs and economic benefits that may be associated with agricultural expansion or urbanization following forest loss. To address the potential bias from non-random forest loss and confounding trends, we examine a possible water pathway from deforestation to health outcomes by exploiting quasi-random variation in deforestation upstream to assess the impacts on waterborne disease outcomes for rural households downstream. This allows us to isolate a water- specific channel through which forest loss impacts health. We also carry out a series of falsification and robustness checks to rule out spurious temporal or spatial correlation and that the observed health outcomes are caused by water- and not air- or vector-borne diseases. We then replicate the global analysis for Peru, which is one of the 10 most forest-rich countries in the world and has experienced rapid land use change over the last two decades, with 3.4 million hectares of forest lost between 2001 and 2020. One key advantage of this case study is that we are able to complement the DHS data with additional health surveys implemented by Peru’s National Institute of Statistics (a.k.a., INEI) using the DHS methodology. This provides us with a longer repeated cross section and the ability to control more comprehensively for time and location specific trends. Our analysis indicates that deforestation increases DD incidence among children under five in downstream communities, but also offers new evidence of early-life exposure to deforestation on childhood stunting, a well- known indicator of later-life productivity. Our results suggest that a 1 percent decline in upstream forest cover increases the probability of DD by 5.8 percent. These results are consistent with previous evidence (Herrera et al., 2017) and highlight the vital role that upstream forests play in regulating water flow and quality downstream. We also present new evidence on the long-lasting and significant health impacts of deforestation, with a 1 percent decline in upstream forest cover around the time of birth increasing the probability of stunting by 3.9 percent. This effect is particularly large in early childhood, measured by the first thousand days of life. These global results also hold for Peru for DD but less so for stunting, which could be due to significant government- and donor-supported health and sanitation campaigns during our sample period aimed to reduce stunting. Our analysis has both analytical and practical implications. From an analytical point of view, this study contributes to a growing literature on the complementarities between human and natural capital. It provides further evidence of the importance of the water channel for short- and long-term health outcomes. On the practical side, our results suggest that maintaining (and potentially restoring) natural capital has the potential 1 We restrict our sample to non-migrant children, i.e., children whose households have either always lived in their current location, or that have migrated to their current location before the child was conceived. 4 to generate meaningful improvements in long-run human capital and that sustainable management of forests should be part of the menu of interventions targeting stunting in rural communities. The rest of the paper is organized as follows: section 2 explains the data sources and the process of combining the health outcomes with the geospatial information to create a rich micro database, section 3 describes our empirical strategy. Section 4 analyzes the results, and conducts falsification and robustness checks to demonstrate that our results are driven by the water channel. Section 5 concludes. 2. Data Analyzing the impact of upstream forest cover loss on downstream health outcomes requires combining household surveys with high-resolution spatial data. We use data from DHS and link the spatial coordinates of the household clusters with geospatial data on forest cover (Hansen et al., 2013), upstream and downstream watersheds (Lehner et al., 2008) and climate (Hersbach et al., 2018). 2.1. Demographic and Household Surveys (DHS) Data The DHS Program, sponsored by the United States Agency for International Development (USAID), provides technical assistance for the implementation of nationally representative, stratified, two-stage cluster sample household surveys that collect data on population, health, and nutrition for over 90 developing countries around the world. We focus our analysis on 91 surveys representing 46 developing countries 2 for the 2000-2020 period. To be included in our sample, each DHS survey must georeference clusters (shown in Figure 1), so that we can match the health outcomes and household characteristics to the environmental and climate data. Following Herrera et al. (2017), we restrict the sample to rural communities, as they are directly dependent on watershed conditions, and focus only on households that have either always lived in their current location or have migrated to their current location before the child was conceived (the rural non-migrant sample). This ensures that health outcomes are measured in the same location of conception and birth where exposure occurred. Given that the health impacts of low water quality are most significant for young children, we restrict the sample further to ages 0 to 5 years, yielding a final sample of 0.7 million children. 2 The 46 countries that are used in our analysis were selected based on the following criteria: countries that provide GPS coordinates for each cluster (this rules out most DHS surveys, as only 60 countries provide geo-referenced data), countries that surveyed individuals between 2000-2020, and countries that provide migration data. For the list of countries included in our DHS sample, please refer to Table A1. 5 Figure 1: Location of DHS clusters, 2000-2020. Notes: Rural clusters are identified as black dots while urban cluster as yellow dots. Table 1 shows the regional distribution of the 0.7 million children in our dataset. Since DHS data are only collected in developing countries, most of our observations are from Sub-Saharan Africa and South Asia with each constituting around 41 percent of the total number of children. India constitutes most of the data due to its population size, but most of the countries in our sample (58 percent) are in the Sub-Saharan Africa region. The sample selection of countries is, thus, not representative across regions, with our final selection of surveys and countries determined by the availability of spatial coordinates and migration data. Table 1: DHS data for rural non-migrant children by region, 2000-2020 Regions Countries Surveys Clusters Children East Asia and Pacific 3 7 2,980 36,806 Europe and Central Asia 4 4 801 6,575 Latin America & Caribbean 6 10 4,714 41,591 Middle East & North Africa 3 8 3,305 34,407 South Asia 3 9 37,773 300,953 Sub-Saharan Africa 27 53 16,785 303,760 Total 46 91 66,358 724,092 Notes: Our sample only includes children residing in rural clusters, whose households have either always lived in the same location or those that have migrated to their current location before the child was conceived (non-migrant sample). Countries are grouped in regional aggregates based on the World Bank Group regional classification. For our sample of rural non-migrant children, we construct health indicators on DD and stunting. The DHS surveys record whether each child has experienced DD in the last 24 hours or within the last 2 weeks. We use this information to construct a dummy variable for whether the child had DD. The DHS surveys also provide data on height-for-age z-scores that can be used to construct stunting measures. We construct a dummy variable for severe stunting if a child’s height-for-age z-score is 3 standard deviations below the mean of the WHO Child Growth Standards. We also compile data on demographic controls such as child’s sex, age, and birth order, as well as the mother’s age at birth. Finally, we use the WHO/UNICEF Joint Monitoring Programme (JMP) definition 6 to identify which households have access to safe drinking water through “improved water sources”, such as piped water, boreholes, and packaged water. 3 Households with unsafe drinking water typically access “unimproved water sources”, including unprotected well and surface water. Contaminated water sources can lead to many chronic diseases, particularly in rural areas of developing countries, where a significant percentage of the poor population – more than a quarter in our sample – continues to use surface water as their main source of drinking water (WHO and UNICEF, 2021). Table 2 reports access to improved water sources in our DHS sample, distinguishing between rich and poor non-migrant households residing in rural and urban areas. While 93 percent of households in urban areas have access to improved water sources, that number decreases to around 73 percent in rural areas. Wealth differences further aggravate access to improved water, with only 56 percent of poor rural households having access to improved water compared to 92 percent of their rich counterparts. These statistics imply that those who are most vulnerable to water pollution are poor people residing in rural areas. Table 2: Percent share of household’s source of drinking water by urban/rural and wealth Water source Urban Rural Urban Rich Urban Poor Rural Rich Rural Poor Improved drinking water 92.60 73.26 96.66 71.69 92.26 55.65 Unimproved drinking water 7.40 26.74 3.34 28.31 7.74 44.35 Total 100 100 100 100 100 100 Notes: The household wealth index from DHS was used to divide households into 5 quintiles based on their level of wealth. 4 Rich in the table refers to the richest quintile (quintile = 5) and poor to the poorest quintile (quintile = 1). The average incidence of DD and stunting among under five-year-old in our sample is 12.6 and 15.5 percent, respectively (Table A1), but there is considerable variation across countries (Table A2). Figure 2a shows the average incidence of DD is particularly high in Latin American, Caribbean and Sub-Saharan African countries, where every fifth child surveyed experiencing DD on average among the top 10 countries. 5 Stunting rates also very greatly across countries, ranging from 29 and 26 percent in Timor-Leste and Madagascar, respectively, to a mere 3 percent in Colombia, The Gambia, and the Dominican Republic (Figure 2b). Stunting rates are generally high in Sub-Saharan Africa and in all Asian countries in our sample. 3 The following water sources are included in the JMP definition of improved drinking water sources: piped water, borehole or tube wells, protected dug wells, protected springs, rainwater, and packaged or delivered water. Any other sources are considered as an unimproved water source. 4 The Wealth index from DHS is defined after accounting for water, sanitation, and hygiene (WASH) variables. We eliminate the WASH variables to create our own wealth index using the Principal Component Analysis. 5 The highest incidence of DD (in percent) was experienced by Bolivia (25.3), Haiti (23.0), Burundi 2.0), Uganda (22.0), Senegal (21.5), Burkina Faso (20.0), The Gambia (20.0), Cambodia (19.8), and Malawi (19.6). 7 Figure 2a: Diarrheal disease rates of rural non-migrant children by country (in percent), 2000-2020 Figure 2b: Stunting rates of rural non-migrant children by country (in percent), 2000-2020 Notes: A child is defined as non-migrant, if its household has either always lived in the current location, or has migrated to the current location before the child was conceived. DD rates are computed as the share of rural non-migrant children that experienced DD in the last 24 hours or within the last 2 weeks of being surveyed. Severe Stunting rates are computed as the share of rural non-migrant children with height-for-age z-scores that are 3 standard deviations below the mean of the WHO Child Growth Standards. Health impacts also vary greatly within countries across age groups and between rural and urban clusters. The incidence of DD is highest among the youngest children, with up to 20 percent of one-year-old children having experienced DD at the time of survey (Figure 3a). This incidence nearly halves by age two and declines even further for children aged three and four years. The incidence of severe stunting mimics this pattern, but declines are much less pronounced with high stunting rates persisting for older children (Figure 3b). Differences across rural and urban areas are particularly pronounced for stunting rates which are on average twice as high in rural areas. These differences are likely driven by socioeconomic status and the availability of services, as there is no significant rural-urban divide for DD rates. 8 Figure 3a: Diarrheal disease rates for non-migrant children by age and urban/rural area Figure 3b: Stunting rates for non-migrant children by age and urban/rural area Notes: See notes to Figure 2 for details on variable construction. To ensure that we can map the child and household information to the environmental and climate data, we need to use the coordinates of the cluster within which each household was surveyed. However, the DHS does not report exact coordinates. Instead, the clusters included in the survey randomly displace or jitter the GPS coordinate within 5 kilometers (km) for rural clusters. 6 Given that the location of the clusters is uncertain, we create 5km buffers around the coordinates provided by DHS (Figure 4). We then use these buffers to distinguish the impact of environmental variables in the vicinity of the cluster (i.e., inside the 5km buffer) or upstream and downstream of the cluster (using watersheds upstream or downstream of the 5km buffer). The next section further discusses how the environmental and climate variables were constructed and matched to the DHS clusters. 6 For urban clusters the coordinates are randomly jittered within a 2km radius. 9 Figure 4: Cluster coordinates and 5km buffer 2.2. Hydrological Data To identify areas upstream and downstream of DHS clusters we used the HydroBASINS product of HydroSHEDS (Lehner et al., 2008), which represents a series of vectorized polygon layers that depict sub-basin boundaries at a global scale. HydroBASINS divides a basin into two sub-basins at every location where two river branches meet which each having their respective upstream area of at least 100 square kilometers (km²). The data is further divided into 12 nested levels using a modification of the Pfafstetter level system (Verdin and Verdin, 1999), where level 1 corresponds to continental scale, and level 12 has an average area of 130km². HydroSHEDS’ coding scheme also allows for the analysis of watershed topologies, by providing identifiers for the polygon immediately downstream of each sub-basin. We overlay household cluster-buffer polygons with level 12 HydroSHEDS sub-basins and determine in which sub-basin(s) each cluster-buffer is contained. The index system is then used to identify sub-basins upstream. We define our upstream area as the union of all sub-basins immediately upstream of the sub-basin(s) that contain the cluster-buffer, removing the area of the cluster-buffer in case it intersects the upstream sub-basins. The average size of the upstream watersheds used in our analysis is 456km2. However, this size varies considerably depending on the number and size of the upstream watersheds, ranging from merely 2km2 to nearly 2,300km2. We follow this procedure in the inverted direction to identify downstream polygons for placebo tests (Figure 5). 10 Figure 5: Cluster buffer with upstream and downstream sub-basins 2.3. Environmental and Climate Data For forest cover loss, we use the Global Forest Change dataset by Hansen et al. (2013). They quantify the global forest change from a supervised pixel-based image classification of the Landsat satellite images at a spatial resolution of 30 meters, 7 using 2000 forest cover estimates as well as annual indicators from 2001 to 2020 for forest loss and gain. To estimate forest cover changes, we compute standing forests in 2000 and track forest cover loss until 2020. We count the number of forest pixels lost each year within each cluster (converting total counts to percentages), thus creating a panel of annual deforestation at the cluster-level from 2001-2020 (Figure 6). For our identification strategy, it is critical to account for annual deforestation in the watershed upstream from the cluster. We also need annual deforestation downstream for our falsification test. We thus follow the same approach to compute annual deforestation in the watersheds that are located upstream and downstream from each cluster. 7 Trees are defined as all vegetation taller than 5 meters in height. 11 Figure 6: Cluster buffer, sub-basins, and deforestation since 2000 As precipitation and temperature have been known to be significant predictors of DD (Halpenny et al., 2012; Escamilla et al., 2011), we also construct climate controls for each cluster within its buffer area. In particular, we construct average annual temperature in Celsius and total precipitation in meters for each cluster, using the ERA5-Land product (Muñoz-Sabater et al. 2021). This data is provided as monthly aggregates at 11,132m resolution. To match the temporal resolution of the deforestation data, we further aggregate this data to create yearly means for temperature and total monthly precipitation. 2.4. Data for Peru Case Study Peru was chosen as a case study for our analysis, given that it is one of the top ten forest-rich countries in the world, which has experienced both rapid changes in land use and health outcomes over the last two decades. Compared to other countries in Latin America, only Brazil has larger forest cover, with Peruvian forests representing 54 percent of the country’s territory (FAO and SERFOR, 2017). However, the country has experienced rapid tree cover loss of about 3.15 million hectares between 2004 and 2020, equivalent to a 4 percent decrease in tree cover since 2000, resulting in 2.03 Gt of uncompensated CO2e emissions.8 Five Amazon regions - Loreto, Ucayali, San Martin, Madre de Dios, and Amazonas - were responsible for 78 percent of all tree cover loss, 9 which was driven by the growing demand for fuel wood, exports of hardwoods, and expansion into the Peruvian Amazon for cultivation, livestock, and the extraction of minerals (Swenson et al., 2011; Piu and Menton, 2014). Despite tree cover loss, child health indicators have improved in the last two decades. Particularly, the incidence of diarrhea and stunted growth decreased significantly more in children in rural areas than in urban areas due to various government- and donor-supported interventions and social programs. Specifically, the Peruvian government began a handwashing initiative in 2003 with technical assistance from the World Bank through public education and media messaging to increase handwashing among mothers and young children, 10 which 8 Global Forest Watch. “Tree cover loss in [Peru/Madre de Dios]” from www.globalforestwatch.org. 9 Global Forest Watch. “Location of tree cover loss in Peru”. From www.globalforestwatch.org. 10 The Global Public-Private Partnership for Handwashing (2003-2010) led by the World Bank was implemented by Peru, Senegal, Tanzania, and Vietnam. 12 reduced the incidence of childhood diarrhea (Glassman and Temin, 2016). Additionally, the launch of the cash transfer program Juntos (Spanish for “together”) and national nutrition strategy named CRECER (Spanish for “to grow”) in 2005 and 2007, respectively further reduced the prevalence of childhood malnutrition (Marini et al. 2017, Huicho et al. 2020). Other social programs such as Wawa Wasi (Quechua for “Children’s home”) and Qali Warma (Quechua for “vigorous child”), 11 serve impoverished children with the aim of improving early childhood development (MIDIS, 2012). However, despite these efforts, childhood DD and stunting continue to be public health problems according to the Global Burden Disease study (IHME, 2020). Peru’s National Institute of Statistics (a.k.a, INEI) runs several rounds of health surveys based on the DHS methodology, which we combine with the survey data collected through the USAID’s program (Figure 7). The final database contains health and demographic data for 17 surveys for the 2004-2020 period (for more detail see Annex Table A3); a significantly larger number of surveys compared to other countries in the DHS program, making Peru a good case study to replicate the global analysis. The data covers 24 regions and one constitutional province, 12 which includes 196 provinces, and 1,533 municipalities. The sample encompasses 55,021 rural non- migrant children, whose locations are represented by time-varying clusters (see Annex Table A4). That is, unlike the Global DHS data, the Peru data contains clusters with GPS coordinates that vary over time, with some clusters having more than one GPS coordinate from 2013 onwards. This change was introduced by INEI to conduct more surveys in different compact groups per cluster. To ensure that we can link the health outcomes to the environment and climate data, we created a unique identifier that can track the changes by GPS coordinates. Figure 7: Location of DHS-type surveys in Peru, 2004-2020. Notes. GPS location points are calculated according to Peru DHS unique identifiers. To ensure comparability with the global analysis, we produce the same set of health and socioeconomic controls. This includes measures of the incidence of childhood diarrhea and stunting, as well as the child’s age, sex, and 11 Quechua is the most widely spoken indigenous language in Peru, and it is one of the country’s official languages. 12 The constitutional province is considered a region. 13 birth order, the mother’s age at the child’s birth, and whether the household has access to improved drinking water sources. Similar to the global analysis, we find that poor rural households have more limited access to safe drinking water with 14.4 percent of poor rural households not having access to safe drinking water, while less than 1 percent of poor households in urban areas suffer from this problem (see Table A5). Regarding the child health indicators, the average incidence of DD and severe stunting is 12 and 7 percent, respectively, which have declined significantly over the last 20 years (Figure 8). The trend of the incidence of DD and severe stunting in Peru is similar to the regional average in Latin America and the Caribbean with higher impacts in rural areas and for one-year-olds especially for DD (Figure A1). Remarkably, due to the public health interventions over the last 20 years, the total number of cases of DD and severe stunting of non-migrant children have declined by roughly 10 percent and 15 percent, respectively. Especially, the decrease of stunting, particularly in rural areas of Peru is seen as a “success story” in the fight against malnutrition (Santos et al. 2021). By contrast, since a part of the rural population has limited access to improved drinking water and sanitation, DD rates still turn out to be higher than stunting rates. Figure 8: Diarrheal Disease and Stunting rates for non-migrant children by age and urban/rural area, Peru case study Notes: See notes to Figure 2 for details on variable construction. The outlier in 2008 is driven by the exclusion of 437 clusters without GPS coordinates that were added to the sample that year. If they are included, the DD rate for 2008 decreases in urban areas to 15.57 percent and rural areas to 15.75 percent in line with the broader trend in the data. 3. Empirical Strategy The primary challenge faced in the literature on both the forest-health nexus and regional environmental spillovers is that the location of forest loss is potentially nonrandom and confounding trends may be associated with health outcomes. To circumvent this bias and to examine a possible water pathway from deforestation to health outcomes, we use the hydrological relationship across watersheds to examine the impact of upstream forest loss on health outcomes downstream. This, in turn, allows us to isolate whether there is a water-specific channel through which forest loss impacts health. Forested watersheds are particularly critical to maintain water quality and stabilize the hydrological cycle. Deforestation can decrease the capacity of soil infiltration of water, increase soil erosion, and decrease the quality of water at the source as well as in downstream watersheds (Filoso et al., 2017; El Kateb et al., 2013; Zheng, 2006; Mapulanga and Naito, 2019). Water filtration services by forests are thus likely to be particularly important for those living primarily in low- and middle-income countries, who use unimproved drinking water sources (Table 2). When upstream areas of watersheds are cleared, downstream water can have higher pH, 14 become more contaminated, or more turbid (Cunha et al., 2016), which render less effective the point-of-use chlorination that is commonly used by households in low-income countries (Amrose et al 2015). To estimate the health impacts of forest cover loss via the water channel, we therefore begin by investigating impacts on diarrhea. Our focus on DD is deliberate since different diseases have different disease ecologies and thus different mechanisms of transmission (Pattanayak and Pfaff, 2009; Berazneva and Byker, 2017; Garg, 2019). For our main specification we estimate equation (1) on the pooled sample using Ordinary Least Squares (OLS), where is a dummy for whether child i experienced DD at the time of survey in year t. For each child we also record its location in cluster c in province p of country g: = + 1 ∆,−1 + 2 ∆−1,−2 + 1 + 2 −1 + + + + (1) ∆,−1 and ∆−1,−2 measure annual changes in the share of forest loss in the upstream catchment area relative to survey year t and t-1, respectively. These values are measured for upstream catchments, exploiting the natural flow of rivers and the fact that water flows downstream. Forest loss upstream is, thus, orthogonal to downstream health. In this way, we exploit quasi-random variation in forest cover loss that originates upstream to estimate the impacts on water quality and health outcomes downstream. Additionally, since DD may vary over time, we also control for country-by-survey year fixed effects ( ) to account for the year when information on the child’s DD condition was collected as well as unobserved national macroeconomic conditions which might otherwise confound the relationship between forest exposure and DD. Province fixed effects ( ) are included to control for any time-invariant unobservable differences between provinces that could affect health. 13 For example, access to local nutrition programs is one such factor that may be constant across individuals residing in the same location. Several other cluster- and household-specific variables are included in the analysis. and −1 are time- varying variables at the cluster level, measuring annual mean temperature and total precipitation within the 5km buffer area for survey year t and t-1, respectively. These climate variables are potentially confounding factors, as they are known predictors of DD (Halpenny et al., 2012; Escamilla et al., 2011). are controls for household characteristics such as age, sex, and birth order of the child, as well as the mother’s age at the time of birth. Since access to clean drinking water reduces the incidence of water-borne diseases like diarrhea, in some specifications we also control for whether a household has access to an improved water source at the time of the survey using the WHO/JMP definition. 14 Lastly, we cluster robust standard errors at the province level to account for arbitrary correlation of observations across time and within provinces. To estimate stunting impacts of childhood exposure to forest cover loss via the water channel, we use a similar research design but modify the timing of exposure. Specifically, we exploit quasi-random variation in upstream forest cover loss experienced by different birth cohorts and compare height outcomes between exposed and non-exposed cohorts, controlling for average differences in these outcomes across birth years and across clusters. For our main specification we estimate equation (2) on the pooled sample where is a dummy for whether child i was severely stunted at the time of survey in year t and was born in year b: = + ∆,−3 + 1 + + 2 −1 + + + (2) 13 When estimating equation (1) for the Peru sample, we control for district fixed effects, as access to public health facilities and local nutrition programs is constant across individuals residing in the same district. 14 The WHO/JMP definition of improved drinking water sources includes piped water, boreholes or tube wells, protected dug wells, protected springs, rainwater, and packaged or delivered water. 15 ∆,−3 denotes the cumulative change in upstream forest cover loss over the three years prior to birth year b. We assign the calendar year in which the child spends in utero to be “one year before birth” (in utero year), 15 so that tree loss in the next calendar year is tree loss in the “year of birth.” We also control for province fixed effects ( ) country-by-birth year ( ). The country-by-birth-year fixed effects are included to account for age effects in health outcomes as well as unobserved national or seasonal shocks, such as macroeconomic conditions or weather patterns, which might otherwise confound the relationship between forest exposure and height. The analysis, thus, exploits within-province variation in birth timing relative to forest loss exposure to identify . Equation (2), therefore, compares two types of children from the same province who are subjected to different levels of forest loss 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 province of birth. The reduced-form relationship provides a causal estimate of the downstream health damages caused by upstream environmental degradation, adding to related work on pollution spillovers by Garg et al. (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 alleviates concerns that the results are being driven by confounding factors to health that may be correlated with single events. 4. Results 4.1. Global Results Table 3 presents estimates of the impact of upstream forest loss on downstream DD using a pooled OLS regression for the global sample. Columns 1, 2 and 3 report parameter estimate of equation (1), sequentially including controls related to climate, individual and household demographic characteristics, and access to improved water sources, respectively. Across all three specifications, we find a positive and statistically significant effect at the five percent level of upstream forest loss on DD incidence in the year preceding the survey. Our preferred specification in column 3 indicates that when the share of forest loss increases from 0 to 1 (a 100-percentage point change), DD incidence increases by 0.69 percentage points. The final row of the table interprets the magnitude of the estimated impact for a one percent forest loss in terms of mean levels of DD incidence. Using the sample average of the probability of DD of 0.12, a one percent decline in upstream forest cover increases the probability of DD by 5.77 percent. These results are consistent with previous evidence (Herrera et al., 2017), and highlight the important role that upstream forests play in regulating water flow and quality. Table 3: Impact of upstream forest loss on DD incidence, 2004-2020 (global rural non-migrant sample) (1) (2) (3) (4) (5) Dependent variable: Incidence of Diarrheal Disease Upstream forest loss Year t-1 to t 0.748** 0.693** 0.693** (0.338) (0.342) (0.341) Year t-2 to t-1 -0.450 -0.389 -0.399 (0.351) (0.357) (0.357) 15 In utero years are defined as 9 months before the birth date. For example, if a child is born in January of 2003, the birth year will be 2003 and the in-utero year will be 2002. If, on the other hand, a child is born in December of 2003 both the birth and in utero year will be 2003. 16 Downstream forest loss Year t-1 to t 0.342 (0.298) Year t-2 to t-1 0.0904 (0.306) Upstream forest loss Year t to t+1 -0.223 (0.441) Year t+1 to t+2 0.595 (0.426) N 496,484 496,484 496,484 496,484 381,251 Fixed Effects Y Y Y Y Y Climate Controls Y Y Y Y Y Demographic Controls N Y Y Y Y Improved Water Source N N Y Y Y R-sq 0.0343 0.0519 0.0519 0.0519 0.0554 Mean diarrheal incidence 12.02% 12.02% 12.02% Impact of a 1 percent loss in upstream forest cover 1 6.23% 5.77% 5.77% year ago as a percent of mean diarrhea incidence Notes: This table presents pooled OLS estimates of equation (1) where the dependent variable is the DD incidence for non- migrant children under the age of 5 between 2004-2020 using the global sample. Forest loss is measured in terms of the share of forest loss over the upstream catchment area relative to survey year t of a child. Demographic controls include the age, sex, and birth order of the child, as well as the mother's age at the time of birth and child’s age in months. Climate controls are annual mean temperature, and total precipitation for survey year t and t-1. Improved water source controls for whether the household of child i has access to safe drinking water. Fixed Effects include country-by-survey-year and province fixed effects. Robust standard errors clustered at the province level are reported in parentheses. * p<0.10, ** p<0.05, *** p<0.01. To validate the research design, we also perform several falsification tests to examine the possibility that these results are driven by spurious spatial or temporal patterns. Following Zaveri et al. (2020), we replace our measure of upstream forest loss with a falsified variable capturing forest loss in the downstream watershed instead. Since water cannot flow upstream, it should not impact health outcomes of households in the cluster. If we were to find a significant health impact, then it would be likely that our baseline results are capturing spurious spatial correlation and other mechanisms of disease transmission. Results in Column 4 show that there is no statistically significant effect of the falsified forest loss variable on the incidence of DD. In addition to the spatial falsification test, we examine the possibility of spurious temporal correlation by controlling for leads of upstream forest loss. Column 5 of Table 3 reports that there is no statistical significance of future upstream forest loss on the incidence of DD. This is consistent with the preceding periods’ upstream forest cover loss increasing the incidence of DD through a water mechanism. These results are also robust to estimating the main specification using logit or probit models (see Table A6). We also test for the impacts of upstream forest loss on malaria and acute respiratory infections, which are vector-borne diseases transmitted by mosquitos. The channels through which upstream deforestation shapes diarrhea are unlikely to impact these diseases and, therefore, they can act as a placebo check to detect spurious association between upstream forest loss and health outcomes. Note that although a body of research has investigated the relationship between forest loss and malaria with mixed findings (e.g., Bauhoff and Busch, 2020; 17 Chakraborty, 2021; Garg 2019; Berazneva and Byker, 2017), these studies analyze the impact of local forest loss as opposed to the upstream forest loss, which is the focus of this paper. As reported in Table 4, we find no evidence for the vector-borne disease channel, which suggests that our results are driven by the water channel. Table 4: Impact of upstream forest loss on vector-borne disease incidence, 2004-2020 (global rural non- migrant sample) (1) (2) (3) (4) Dependent variable: Incidence of Malaria Dependent variable: Incidence of Acute Respiratory Infection Upstream Forest Loss Year t-1 to t 0.240 0.191 (0.571) (0.198) Year t-2 to t-1 -0.727 -0.239 (0.655) (0.241) Downstream Forest Loss Year t-1 to t 0.198 0.149 (0.532) (0.112) Year t-2 to t-1 -0.535 -0.173 (0.529) (0.154) 0.198 0.149 N 515,452 515,452 529,623 529,623 Fixed Effects Y Y Y Y Climate Controls Y Y Y Y Demographic Controls Y Y Y Y R-sq 0.071 0.071 0.026 0. 026 Notes: This table presents pooled OLS estimates of equation (1) where the dependent variables are the airborne disease incidences of malaria and Acute respiratory infection (ARI) for non-migrant children under the age of 5 between 2004-2020. Malaria is a dummy variable for whether the child had fever in the two weeks prior to the survey. ARI is a dummy variable that returns 1 if a child was coughing in last two weeks with chest related short/rapid breathing and 0 otherwise. See notes to Table 3 for details on specification and variable construction for the other controls. Robust standard errors clustered at the province level are reported in parentheses. * p<0.10, ** p<0.05, *** p<0.01. In addition to impacting short-term health, can upstream forest cover loss also have long-term health impacts? The well‐established fetal origins literature suggests that intrauterine health impacts can lead to lasting health damages (Barker, 1990; Currie and Vogl, 2013; Christian et al., 2013; Almond et al., 2018). The best overall measure of early life child well-being is stunting, which measures height/length by age against WHO standards. Lower height‐for‐age scores can lead to severe consequences for cognitive development, overall health, and even socio‐economic conditions that carry into adulthood (Hoddinott et al., 2013). Childhood stunting has been linked to a wide range of later-life health outcomes including lower economic productivity (Kraay, 2018). Several potential pathways might be possible between forest cover loss and stunting through water pollution. Past work has shown that repeated bouts of DD increase the probability of nutritional deficiencies in children and, thus, stunted child development (Zaveri et al., 2020). Exposure to water pollution can adversely affect birth outcomes by lowering infant birthweight (Jones, 2019), an important predictor of stunting later in childhood (Christian et al., 2013). Exposure to water pollution can also disrupt the gut microbiome. The first months after birth are particularly critical for establishing the composition of the gut microbiome that persists for the rest of a person’s life (Robertson et al., 2019). There is evidence in the medical literature that this microbiome is difficult to permanently change later in life, although this matter is under debate (REF). If true, it could reduce an individual’s capacity to absorb nutrients throughout their lives, impacting development. 18 Given the importance of upstream forest cover loss to DD, it is possible that upstream forest cover loss could also influence stunting rates. However, thus far the discussion around targeting drivers of stunting has largely ignored the role of forests. To investigate whether upstream forest cover loss can also have long-term health impacts via stunting, we modify our empirical strategy to account for quasi-random variation in upstream forest cover loss around the time of each child’s birth in the DHS sample. For our main specification we estimate equation (2) on the pooled sample. Since children are born at different times, each child has their own exposure based on their birth year. This contrasts with the contemporaneous analysis presented in Table 3 where exposure to forest loss was the same for all children of the same cluster and was solely based on the year they were surveyed. This specification also includes country-by-birth-year fixed effects to control flexibly for changes over time in stunting and forest cover loss. The results in column 2 of Table 5 indicate that when the share of forest loss in the year preceding birth, i.e., when the child is in-utero, up to the year of birth increases from 0 to 1 (a 100 percentage point change), the likelihood of severe stunting increases by 0.71 percentage points, a result that is statistically significant at the 5 percent level. In other words, a 1 percent decline in upstream forest cover significantly increases the likelihood of severe stunting by 0.0071 percentage points. Using the sample average of the probability of severe stunting of 0.172, a 1 percent decline in upstream forest cover around the time of birth increases the probability of severe stunting by 4.14 percent. The final row of the table interprets the magnitude of the estimated impact of a 1 percent forest loss in terms of mean levels of severe stunting in the sample. Other lags of upstream forest cover loss are positive, but insignificant, suggesting that the transmission channel is indeed through low water quality when the child is in-utero. Table 5: Impact of upstream forest loss on childhood stunting, 2004-2020 (global rural non-migrant sample) (1) (2) (3) (4) (5) Dependent variable: Incidence for Severe Stunting Upstream Forest Loss Birth Year b-1 to b 0.626** 0.713** 0.608* (0.312) (0.322) (0.335) Birth Year b-2 to b-1 -0.158 -0.279 (0.281) (0.264) Birth Year b-3 to b-2 0.379 (0.279) Birth Year b-3 to b 0.236* (0.133) Downstream Forest Loss Birth Year b-3 to b 0.209 (0.181) N 372,040 372,040 372,040 372,040 372,040 Fixed Effects Y Y Y Y Y Climate Controls Y Y Y Y Y Demographic Controls Y Y Y Y Y R-sq 0.042 0.042 0.042 0.042 0.042 Mean stunting incidence 17.22% Impact of a 1 percent loss in upstream 3.64% 4.14% 3.53% forest cover 1 year ago as a percent of mean stunting incidence Notes: Table presents estimates of equation (2) where the dependent variable is the stunting incidence for non-migrant children under the age of 5 between years 2004-2020. Forest loss is measured in terms of the share of forest loss over the upstream catchment area relative to birth year 'b' of a child. See notes to Table 3 for details on the variable construction 19 for the other controls. Fixed Effects include country-by-birth-year and province fixed effects. Robust standard errors clustered at the province level and are reported in parentheses. * p<0.10, ** p<0.05, *** p<0.01. However, to alleviate concerns that the results are being driven by confounding factors to health that may be correlated with single events, we also use a cumulative measure of forest loss that accounts for all the lags together. The results in columns 3 indicate a positive coefficient on the upstream cumulative forest loss measure that is statistically significant at the one percent level, although the magnitude is lower. Finally, to validate the research design, we also run a falsification test using the share in forest loss downstream as opposed to upstream from the cluster as was done when estimating the impact on DD in column 4 of Table 3. The results in column 5 show that there is no statistically significant effect of the spatially falsified cumulative forest loss variable on the incidence of severe stunting. This bolsters the argument that upstream forest cover loss increases the incidence of stunting through a water mechanism. We also find evidence that the impacts vary across age (Figure 9). We estimate a variant of the model in Column (1) in Table 5 by interacting the share of forest loss in the year preceding birth with indicator variables for the different age groups in the sample. We do not find impacts in the first year following birth. However, the effects are largest for toddlers aged 13-24 months after which the impact declines. This also coincides with the time window (the first one thousand days) considered to be critical for determining whether a child will grow up stunted (Robertson et al., 2013). Figure 9: Impact of upstream forest loss on childhood stunting by age, 2004-2020 (global rural non-migrant sample) Notes: Estimated coefficients on the share of forest loss in the year preceding birth. Error bars represent 95% confidence intervals. As previously discussed, the analysis is restricted to rural clusters in the DHS. This deliberate restriction stems from the understanding that rural areas are more susceptible to direct influences from upstream watershed conditions (Herrera et al., 2017). Residents in these regions typically reside in close proximity to forested areas and are thus more affected by changes in the ecosystem services they provide. Moreover, as poor rural residents in our sample generally have lower access to improved water sources, the quality of their drinking water will likely be more impacted by poorer water filtration and greater soil erosion in a degraded watershed upstream. We would thus expect to only find an impact in rural and not urban clusters. The findings presented in Table 6 validate this assumption, as they indicate minimal or no discernible impact of upstream forest loss on DD or 20 severe stunting in urban clusters. This observation suggests varying degrees of dependence on watershed conditions between rural and urban clusters, highlighting the crucial role of upstream forests specifically for rural areas. Table 6: Impact of upstream forest loss on children health, 2004-2020 (global urban non-migrant sample) (1) (2) (3) (4) (5) Dependent variable: Diarrheal Disease Dependent variable: Severe Stunting Upstream Forest Loss Survey Year t-1 to t 0.396 0.332 0.324 (0.281) (0.279) (0.278) Survey Year t-2 to t-1 -0.214 -0.161 -0.163 (0.275) (0.266) (0.265) Birth Year b-1 to b 0.228 (0.301) Birth Year b-3 to b -0.0642 (0.144) N 184,151 184,151 184,151 129,217 129,217 Fixed Effects Y Y Y Y Y Climate Controls Y Y Y Y Y Demographic Controls N Y Y Y Y Improved Water N N Y N N R-sq 0. 031 0. 048 0.048 0. 044 0.044 Notes: Table presents estimates where the dependent variable is the health (diarrheal disease and severe stunting) incidence for non-migrant children residing in urban areas that are under the age of 5 between years 2004-2020. See notes to Table 3 and 5 for details on the variable construction and specifications of equations (1) and (2), respectively. Robust standard errors clustered at the province level and are reported in parentheses. * p<0.10, ** p<0.05, *** p<0.01. 4.2. Peru Case Study Having established global results, next we test our model for Peru, a country of particular interest for our study given the considerable threats to forests and the ecosystem services they provide. The model for this subset shows qualitatively similar results for DD in Table 7 as our global sample analysis. The results in columns 1, 2, and 3 in Table 7 indicate that when the share of forest loss in the year prior to the survey increases from 0 to 1 (a 100-percentage point change), DD incidence increases by 2.06 percentage points on average. In other words, a 1 percent decline in upstream forest cover significantly increases the likelihood of DD by 0.0206 percentage points. The magnitude of this effect is more than double the effect in the global sample, with a 1 percent decline in upstream forest cover increasing the probability of DD by approximately 16.29 percent. The same falsification tests are also carried out for the Peru sample, including using the falsified forest loss variable (column 4) and urban sample (column 5). The coefficient estimates are insignificant, providing further evidence that we are able to isolate the water channel. Table 7: Impact of upstream forest loss on DD incidence, 2004-2020 (Peru sample) (1) (2) (3) (4) (5) Dependent variable: Incidence for Diarrheal Disease Upstream forest loss Year t-1 to t 2.056** 2.047** 2.059*** -0.155 (0.826) (0.798) (0.796) (0.558) 21 Year t-2 to t-1 -0.835 -0.756 -0.782 -0.066 (0.799) (0.790) (0.790) (0.631) Downstream forest loss Year t-1 to t 1.057 (0.670) Year t-2 to t-1 -0.161 (0.651) N 38,652 38,652 38,652 38,652 83,322 Location of cluster Rural Rural Rural Rural Urban Fixed Effects Y Y Y Y Y Climate Controls Y Y Y Y Y Demographic Controls N Y Y Y Y Improved Water N N Y Y Y R-sq 0.057 0.075 0.075 0.075 0.037 Mean diarrheal incidence 12.62% Impact of a 1 percent loss in upstream forest 16.29% 16.24% 16.34% cover 1 year ago as a percent of mean diarrhea incidence Notes: This table presents pooled OLS estimates of equation (1) where the dependent variable is the DD incidence for non- migrant children under the age of 5 between 2004-2020 in the Peru sample. See notes to Table 3 for details on the variable construction. Fixed Effects include survey-year and district fixed effects. Robust standard errors clustered at the district level are reported in parentheses. * p<0.10, ** p<0.05, *** p<0.01. However, the results for the impact of upstream forest loss on the incidence of severe stunting reported in Table 8 show a much weaker impact. We only find a positive coefficient for upstream forest loss in the in-utero year, which is statistically significant at the 10 percent level, when the lags are excluded. This weaker effect relative to the global analysis is likely not surprising given that severe stunting rates in Peru saw large declines of nearly 15 percent between 2004 and 2020 due to numerous government and donor-supported sanitation and health programs. We also do not find statistically significant effects for the falsification test using downstream forest loss (column 5) or the urban sample (column 6). Table 8: Impact of upstream forest loss on childhood stunting, 2004-2020 (Peru rural non-migrant sample) (1) (2) (3) (4) (5) (6) Dependent variable: Incidence for Severe Stunting Upstream Forest Loss Birth Year b-1 to b 0.867* 0.845 0.739 22 (0.481) (0.514) (0.500) Birth Year b-2 to b-1 0.041 -0.053 (0.473) (0.509) Birth Year b-3 to b-2 0.0293 (0.500) Birth Year b-3 to b 0.326 -0.035 (0.212) (0.065) Downstream Forest Loss Birth Year b-3 to b 0.194 (0.226) N 33,683 33,683 33,683 33,683 33,683 75,930 Location of cluster Rural Rural Rural Rural Rural Urban Fixed Effects Y Y Y Y Y Y Climate Controls Y Y Y Y Y Y Demographic Controls Y Y Y Y Y Y R-sq 0.096 0.096 0.096 0.096 0.096 0.030 Mean stunting incidence 6.95% Impact of a 1 percent loss in 12.47% upstream forest cover 1 year ago as a percent of mean stunting incidence Notes: Table presents estimates of equation (2) where the dependent variable is the stunting incidence for non-migrant children under the age of 5 between years 2004-2020. See notes to Table 5 for details on the variable construction. Fixed Effects include child’s-birth-year and month, and district fixed effects. Robust standard errors clustered at the district level and are reported in parentheses. * p<0.10, ** p<0.05, *** p<0.01. 5. Conclusion In this paper, we analyzed the complementarities between natural and human capital using a global analysis of 0.7 million rural non-migrant children across 46 countries. By linking high-resolution forest loss data with individual level health outcomes, our analysis shows that deforestation upstream has significant implications for DD incidence and stunting among children under 5 years old in downstream communities. Moreover, we provide novel evidence that these effects can have long lasting and significant impacts on human capital through stunting. This effect is particularly large if children are exposed to poor water quality in the first thousand days of life. Results for Peru confirm the sign and statistical significance of the result for DD, which is twice as large as the global results. However, the impact on stunting is less pronounced likely due to targeted government and donor health and sanitation campaigns. Our analysis has both analytical and practical implications. From an analytical point of view, our analysis complements a growing literature on the complementarities between human and natural capital and provides further evidence of the importance of the water channel for short- and long-term health outcomes. On the practical side, our results suggest that maintaining natural capital has the potential to generate meaningful improvements in long-run human capital and that sustainable management of forests should be part of the menu of interventions targeting stunting in rural communities. The study of the complementarities or substitutability between different forms of capital is a critical yet data demanding task to understand the sustainability of economic development (Cohen et al., 2019). Any future research that analyzes the substitutability and/or complementarity between different types of capital or across subcategories of capital in production or consumption process will help improve our understanding significantly. 23 References Ahern, M., and K. Sari. 2013. “The Health Impacts of Floods.” Taylor and Francis Group, January, 1–219. https://doi.org/10.4324/9781849771351. Almond, D., J. Currie, and V. Duque. 2018. “Childhood Circumstances and Adult Outcomes: Act II.” Journal of Economic Literature 56 (4): 1360–446. Amrose S, Burt Z and Ray I 2015 Safe drinking water for low-income regions Annu. Rev. Environ. Resour. 40 203–31 Andréassian, V. 2004. “Waters and Forests: From Historical Controversy to Scientific Debate [Review Article].” JHyd 291 (1): 1–27. https://doi.org/10.1016/J.JHYDROL.2003.12.015. Barker DJ (1990) The fetal and infant origins of adult disease. BMJ 301:1111. Berazneva, J., and T. S. Byker. 2017. “Does Forest Loss Increase Human Disease? Evidence from Nigeria.” American Economic Review 107 (5): 516–21. https://doi.org/10.1257/AER.P20171132. Berthe, F., S. Bali, and G. Batmanian. 2022. “Putting Pandemics Behind Us Investing in One Health to Reduce Risks of Emerging Infectious Diseases.” The World Bank. http://documents.worldbank.org/curated/en/099530010212241754/P17840200ca7ff098091b7014001a 08952e. Blom, S., A. Ortiz-Bobea, and J. Hoddinott. 2022. “Heat Exposure and Child Nutrition: Evidence from West Africa.” Journal of Environmental Economics and Management 115 (September): 102698. . Chaplin-Kramer, R., Neugarten, R.A., Sharp, R.P. et al. 2023. Mapping the planet’s critical natural assets. Nat Ecol Evol 7, 51–61. https://doi.org/10.1038/s41559-022-01934-5 Christian, P., S. E. Lee, M. D. Angel, L. S. Adair, S. E. Arifeen, P. Ashorn, F. C. Barros, et al. 2013. “Risk of Childhood Undernutrition Related to Small-for-Gestational Age and Preterm Birth in Low- and Middle- Income Countries.” International Journal of Epidemiology 42 (5): 1340. https://doi.org/10.1093/IJE/DYT109. Cohen, A. J., M. Brauer, R. Burnett, H. R. Anderson, J. Frostad, K. Estep, K. Balakrishnan, et al. 2017. “Estimates and 25-Year Trends of the Global Burden of Disease Attributable to Ambient Air Pollution: An Analysis of Data from the Global Burden of Diseases Study 2015.” The Lancet 389 (10082): 1907–18. https://doi.org/10.1016/S0140-6736(17)30505-6. Cohen, F., C. J. Hepburn, and A. Teytelboym. 2019. “Is Natural Capital Really Substitutable?” Https://Doi.Org/10.1146/Annurev-Environ-101718-033055 44 (October): 425–48. https://doi.org/10.1146/ANNUREV-ENVIRON-101718-033055. Cunha, D. G. F., Sabogal-Paz, L. P., and Dodds, W. K. (2016). Land use influence on raw surface water quality and treatment costs for drinking supply in São Paulo State (Brazil). Ecol. Eng. 94, 516–524. doi: 10.1016/j.ecoleng.2016.06.063 Currie, J. and Vogl, T. 2013. “Early‐life health and adult circumstance in developing countries”, Annual Review of Economics, Annual Reviews, 5(1): 1–36, 05 24 Danaei, G., K. G. Andrews, C. R. Sudfeld, G. Fink, D. C. McCoy, E. Peet, A. Sania, M. C. Smith Fawzi, M. Ezzati, and W. W. Fawzi. 2016. “Risk Factors for Childhood Stunting in 137 Developing Countries: A Comparative Risk Assessment Analysis at Global, Regional, and Country Levels.” PLoS Medicine 13 (11). https://doi.org/10.1371/JOURNAL.PMED.1002164. Deaton, A., 2013. The Great Escape: Health, Wealth and the Origins of Inequality. Princeton University Press. https://doi.org/10.2307/j.ctt3fgxbm Domingo, N. G.G., S. Balasubramanian, S. K. Thakrar, M. A. Clark, P. J. Adams, J. D. Marshall, N. Z. Muller, et al. 2021. “Air Quality-Related Health Damages of Food.” Proceedings of the National Academy of Sciences of the United States of America 118 (20). https://doi.org/10.1073/PNAS.2013637118/-/DCSUPPLEMENTAL. Escamilla, V., B. Wagner, M. Yunus, P. K. Streatfield, A. van Geen, and M. Emch. 2011. “Effect of Deep Tube Well Use on Childhood Diarrhoea in Bangladesh.” Bulletin of the World Health Organization 89 (7): 521. https://doi.org/10.2471/BLT.10.085530. FAO. 2020. Global Forest Resources Assessment 2020: Main report. Rome. https://doi.org/10.4060/ca9825en FAO, and SERFOR. 2017. “Nuestros Bosques En Números. Primer Reporte Del Inventario Nacional Forestal y de Fauna Silvestre.” https://sinia.minam.gob.pe/documentos/nuestros-bosques-numeros. FAO, UNEP, WHO, and WOAH. 2022. “One Health Joint Plan of Action, 2022–2026.” One Health Joint Plan of Action, 2022–2026, October. https://doi.org/10.4060/CC2289EN. Filoso, S., M. O. Bezerra, K. C.B. Weiss, and M. A. Palmer. 2017. “Impacts of Forest Restoration on Water Yield: A Systematic Review.” PLOS ONE 12 (8): e0183210. https://doi.org/10.1371/JOURNAL.PONE.0183210. Galasso, E., and A. Wagstaff. 2018. “The Aggregate Income Losses from Childhood Stunting and the Returns to a Nutrition Intervention Aimed at Reducing Stunting.” The Aggregate Income Losses from Childhood Stunting and the Returns to a Nutrition Intervention Aimed at Reducing Stunting, August. https://doi.org/10.1596/1813-9450-8536. Garg, T. 2019. “Ecosystems and Human Health: The Local Benefits of Forest Cover in Indonesia.” Journal of Environmental Economics and Management 98 (November): 102271. Garg, T., Hamilton, S. E., Hochard, J. P., Kresch, E. P., & Talbot, J. (2018). (Not so) gently down the stream: River pollution and health in Indonesia. Journal of Environmental Economics and Management, 92, 35-53. Giani, P. Castruccio, S., Anav, A., Howard, D., Hu, W., and Crippa, P. 2020. “Short-term and long-term health impacts of air pollution reductions from COVID-19 lockdowns in China and Europe: a modelling study”, The Lancet Planetary Health 4(10): e474-e482, . Glassman, Amanda, and Miriam Temin. 2016. Millions Saved : New Cases of Proven Success in Global Health. Brookings Institution Press. Grammatikopoulou, I., and Vačkářová, D. (2021) “The value of forest ecosystem services: A meta-analysis at the European scale and application to national ecosystem accounting.” Ecosystem Services 48 101262, https://doi.org/10.1016/j.ecoser.2021.101262. 25 Halpenny, C. M., K. G. Koski, V. E. Valdés, and M. E. Scott. 2012. “Prediction of Child Health by Household Density and Asset-Based Indices in Impoverished Indigenous Villages in Rural Panamá.” The American Journal of Tropical Medicine and Hygiene 86 (2): 280. https://doi.org/10.4269/AJTMH.2012.11-0289. Hansen, M. C., P. V. Potapov, R. Moore, M. Hancher, S. A. Turubanova, A. Tyukavina, D. Thau, et al. 2013. “High-Resolution Global Maps of 21st-Century Forest Cover Change.” Science 342 (6160): 850–53. https://doi.org/10.1126/SCIENCE.1244693/SUPPL_FILE/HANSEN.SM.PDF. Herrera, D., A. Ellis, B. Fisher, C. D. Golden, K. Johnson, M. Mulligan, A. Pfaff, T. Treuer, and T. H. Ricketts. 2017. “Upstream Watershed Condition Predicts Rural Children’s Health across 35 Developing Countries.” Nature Communications 2017 8:1 8 (1): 1–8. https://doi.org/10.1038/s41467-017-00775-2. Hoddinott, J. , Behrman, J., Maluccio, J.A., Melgar, P., Quisumbing, A. R., Ramirez-Zea, M., Stein, A. D., Yount, K. M., and Martorell, R.. “Adult consequences of growth failure in early childhood.” The American journal of clinical nutrition, 98(5):1170–1178, 2013. Huangfu, P., and R. Atkinson. 2020. “Long-Term Exposure to NO2 and O3 and All-Cause and Respiratory Mortality: A Systematic Review and Meta-Analysis.” Environment International 144 (November). Huicho, Luis, Elisa Vidal-Cárdenas, Nadia Akseer, Samanpreet Brar, Kaitlin Conway, Muhammad Islam, Elisa Juarez, et al. 2020. “Drivers of Stunting Reduction in Peru: A Country Case Study.” The American Journal of Clinical Nutrition 112 (Suppl 2): 816S. Institute for Health Metrics and Evaluation (IHME). GBD Compare Data Visualization. Seattle, WA: IHME, University of Washington, 2020. Available from http://vizhub.healthdata.org/gbd-compare. (Accessed June 21, 2023) Jenkins, M. and Schaap, B., 2018, April. Forest ecosystem services. In United Nations Forum on Forests. Johnson, J.; Ruta, G. Baldos, U.; Cervigni, R.; Chonabayashi, S., Corong, E.; Gavryliuk, O.; Gerber, J.; Hertel, T.; Nootenboom, C.; Polasky, S. 2021. The Economic Case for Nature: A Global Earth-Economy Model to Assess Development Policy Pathways. World Bank, Washington, DC. Jones, B. A. 2019. “Infant Health Impacts of Freshwater Algal Blooms: Evidence from an Invasive Species Natural Experiment.” Journal of Environmental Economics and Management 96 (July): 36–59. https://doi.org/10.1016/J.JEEM.2019.05.002. Juginović, A., Vuković, M., Aranza, I., and Biloš, V. 2021. “Health impacts of air pollution exposure from 1990 to 2019 in 43 European countries.” Scientific Reports 11, 22516. https://doi.org/10.1038/s41598-021-01802-5 Karayalcin ,C., and Onder, H. 2023. “The Fragility and Resilience of Nations”. Policy Research Working Paper no. WPS 10362 Washington, D.C. : World Bank Group. http://documents.worldbank.org/curated/en/099717103132331477/IDU03f0080c40ac2f0496308b2a03d 516ab9a9a6 Kateb, H. el, H. Zhang, P. Zhang, and R. Mosandl. 2013. “Soil Erosion and Surface Runoff on Different Vegetation Covers and Slope Gradients: A Field Experiment in Southern Shaanxi Province, China.” CATENA 105 (June): 1–10. https://doi.org/10.1016/J.CATENA.2012.12.012. 26 Keiser, D.A., and J.S. Shapiro. 2018. “Consequences of the Clean Water Act and the Demand for Water Quality.” Quarterly Journal of Economics 134 (1): 349–96. . Lehner, B., K. Verdin, and A. Jarvis. 2008. “New Global Hydrography Derived From Spaceborne Elevation Data.” EOSTr 89 (10): 93–94. https://doi.org/10.1029/2008EO100001. Lelieveld, J., J. S. Evans, M. Fnais, D. Giannadaki, and A. Pozzer. 2015. “The Contribution of Outdoor Air Pollution Sources to Premature Mortality on a Global Scale.” Nature 2015 525:7569 525 (7569): 367–71. https://doi.org/10.1038/nature15371. Lipscomb, M., and A. M. Mobarak. 2017. “Decentralization and Pollution Spillovers: Evidence from the Re- Drawing of County Borders in Brazil.” The Review of Economic Studies 84 (1): 464–502. https://doi.org/10.1093/RESTUD/RDW023. Liu, L., H. L. Johnson, S. Cousens, J. Perin, S. Scott, J. E. Lawn, I. Rudan, et al. 2012. “Global, Regional, and National Causes of Child Mortality: An Updated Systematic Analysis for 2010 with Time Trends since 2000.” Lancet (London, England) 379 (9832): 2151–61. . Muñoz-Sabater, Joaquín, et al. “ERA5-Land: A state-of-the-art global reanalysis dataset for land applications.” Earth System Science Data 13.9 (2021): 4349-4383. Mallett, L. H., and R. A. Etzel. 2018. “Flooding: What Is the Impact on Pregnancy and Child Health?” Disasters 42 (3): 432–58. https://doi.org/10.1111/DISA.12256. Manisalidis I., Stavropoulou E., Stavropoulos A., and Bezirtzoglou E. 2020. “Environmental and Health Impacts of Air Pollution: A Review.” Frontier in Public Health 8:14. doi: 10.3389/fpubh.2020.00014 Mapulanga, A. M., and H. Naito. 2019. “Effect of Deforestation on Access to Clean Drinking Water.” Proceedings of the National Academy of Sciences of the United States of America 116 (17): 8249–54. . Marini, Alessandra, Claudia Rokx, and Paul Gallagher. 2017. “Standing Tall : Peru’s Success in Overcoming Its Stunting Crisis,” June. https://doi.org/10.1596/28321. Marta-Pedroso, C., Laporta, L., Proença, V., Azevedo, J.C., Domingos, T. 2014. “Changes in the ecosystem services provided by forests and their economic valuation: a review.” In: Azevedo, J., Perera, A., Pinto, M. (eds) Forest Landscapes and Global Change. Springer, New York, NY. https://doi.org/10.1007/978-1- 4939-0953-7_5 MIDIS. 2012. “A Policy for Development and Social Inclusion in Peru.” https://www.unpei.org/files/files/doc/midis_politicas_desarrollo_en.pdf. Myers, S. S., L. Gaffikin, C. D. Golden, R. S. Ostfeld, K. H. Redford, T. H. Ricketts, W. R. Turner, and S. A. Osofsky. 2013. “Human Health Impacts of Ecosystem Alteration.” Proceedings of the National Academy of Sciences of the United States of America 110 (47): 18753–60. https://doi.org/10.1073/PNAS.1218656110. Narayan, S., P. K. Narayan, and S. Mishra. 2010. “Investigating the Relationship between Health and Economic Growth: Empirical Evidence from a Panel of 5 Asian Countries.” Journal of Asian Economics 21 (4): 404– 11. https://ideas.repec.org/a/eee/asieco/v21y2010i4p404-411.html. 27 Ogundari, K., and T. Awokuse. 2018. “Human Capital Contribution to Economic Growth in Sub-Saharan Africa: Does Health Status Matter More than Education?” Economic Analysis and Policy 58 (June): 131–40. https://doi.org/10.1016/j.eap.2018.02.001. Pattanayak, S. K., R. A. Kramer, and J. R. Vincent. 2017. “Ecosystem Change and Human Health: Implementation Economics and Policy.” Philosophical Transactions of the Royal Society B: Biological Sciences 372 (1722). https://doi.org/10.1098/RSTB.2016.0130. Pattanayak, S. K., and A. Pfaff. 2009. “Behavior, Environment, and Health in Developing Countries: Evaluation and Valuation.” Https://Doi.Org/10.1146/Annurev.Resource.050708.144053 1 (1): 183–217. https://doi.org/10.1146/ANNUREV.RESOURCE.050708.144053. Piu, H.C., and M. Menton. 2014. “The Context of REDD+ in Peru: Drivers, Agents and Institutions.” The Context of REDD+ in Peru: Drivers, Agents and Institutions. https://doi.org/10.17528/CIFOR/004438. Prüss-Ustün, A., J. Bartram, T. Clasen, J. M. Colford, O. Cumming, V. Curtis, S. Bonjour, et al. 2014. “Burden of Disease from Inadequate Water, Sanitation and Hygiene in Low- and Middle-Income Settings: A Retrospective Analysis of Data from 145 Countries.” Tropical Medicine & International Health : TM & IH 19 (8): 894–905. https://doi.org/10.1111/TMI.12329. Rasolofoson, R. A., T. H. Ricketts, K. B. Johnson, A. Jacob, and B. Fisher. 2021. “Forests Moderate the Effectiveness of Water Treatment at Reducing Childhood Diarrhea.” Environmental Research Letters 16 (6): 064035. https://doi.org/10.1088/1748-9326/ABFF88. Ritchie, H., and M. Roser. 2013. “Indoor Air Pollution.” Our World in Data, November. https://ourworldindata.org/indoor-air-pollution. Robertson, R. C., A. R. Manges, B. B. Finlay, and A. J. Prendergast. 2019. “The Human Microbiome and Child Growth – First 1000 Days and Beyond.” Trends in Microbiology 27 (2): 131–47. Santos, Maria P, Beja Turner, and M Pia Chaparro. 2021. “The Double Burden of Malnutrition in Peru: An Update with a Focus on Social Inequities.” The American Journal of Clinical Nutrition 113 (4): 865–73. https://doi.org/10.1093/ajcn/nqaa387. Sigman, H. 2002. “International Spillovers and Water Quality in Rivers: Do Countries Free Ride?” American Economic Review 92 (4): 1152–59. https://doi.org/10.1257/00028280260344687. ———. 2004. “Transboundary Spillovers and Decentralization of Environmental Policies,” August. https://doi.org/10.3386/W10717. Siikamäki, J., Piaggio, M., da Silva, N., Álvarez, I., and Chu, Z. 2021. Assessment of Non-Wood Forest Ecosystem Services: A Revision of a Spatially Explicit Meta-Analysis and Benefit Transfer. The World Bank Group. Washington, DC. https://documents1.worldbank.org/curated/en/099850110202253173/pdf/P17727806732bb09d0a33f04 16b38f3c966.pdf Smulders, S., and van Soest, D. 2023. “Natural Capital Substitution: Implications for growth, shadow prices, and natural capital accounting” Working paper Souza Jr, Carlos M., et al. “Reconstructing three decades of land use and land cover changes in brazilian biomes with landsat archive and earth engine.” Remote Sensing 12.17 (2020): 2735. 28 Swenson, J.J., C. E. Carter, J. C. Domec, and C. I. Delgado. 2011. “Gold Mining in the Peruvian Amazon: Global Prices, Deforestation, and Mercury Imports.” PLOS ONE 6 (4): e18875. https://doi.org/10.1371/JOURNAL.PONE.0018875. Tan-Soo, J. S., and S. K. Pattanayak. 2019. “Seeking Natural Capital Projects: Forest Fires, Haze, and Early-Life Exposure in Indonesia.” Proceedings of the National Academy of Sciences of the United States of America 116 (12): 5239–45. https://doi.org/10.1073/PNAS.1802876116/SUPPL_FILE/PNAS.1802876116.SAPP.PDF. Waite, T. D., K. Chaintarli, C. R. Beck, A. Bone, R. Amlôt, S. Kovats, M. Reacher, et al. 2017. “The English National Cohort Study of Flooding and Health: Cross-Sectional Analysis of Mental Health Outcomes at Year One.” BMC Public Health 17 (1): 1–9. https://doi.org/10.1186/S12889-016-4000-2/TABLES/5. Weil, D. N. 2007. “Accounting for the Effect Of Health on Economic Growth.” The Quarterly Journal of Economics 122 (3): 1265–1306. https://doi.org/10.1162/QJEC.122.3.1265. ———. 2014. “Health and Economic Growth.” Handbook of Economic Growth 2: 623–82. https://doi.org/10.1016/B978-0-444-53540-5.00003-3. Whitmee, S., A. Haines, C. Beyrer, F. Boltz, A. G. Capon, B. F. de Souza Dias, A. Ezeh, et al. 2015. “Safeguarding Human Health in the Anthropocene Epoch: Report of the Rockefeller Foundation-Lancet Commission on Planetary Health.” The Lancet 386 (10007): 1973–2028. WHO, and UNICEF. 2021. “Progress on Household Drinking Water, Sanitation and Hygiene 2000‒2020: Five Years into the SDGs.” Geneva. . World Bank. 2016. The Cost of Air Pollution: Strengthening the Economic Case for Action. Washington, DC: World Bank Group. World Bank. 2021. The Changing Wealth of Nations 2021: Managing Assets for the Future. Washington, DC: World Bank Group. https://doi.org/10.1596/978-1-4648-1590-4. Wulder, Michael A., et al. 2016 “The global Landsat archive: Status, consolidation, and direction.” Remote Sensing of Environment 185: 271-283. Zaveri, E. D., Russ, J. D., Desbureaux, S. G., Damania, R., Rodella, A. S., & Ribeiro Paiva De Souza, G. 2020. The nitrogen legacy: the long-term effects of water pollution on human capital. World Bank Policy Research Working Paper, (9143). Zheng, F. L. 2006. “Effect of Vegetation Changes on Soil Erosion on the Loess Plateau1 1 Project Supported by the Chinese Academy of Sciences (No. KZCX3-SW-422) and the National Natural Science Foundation of China (Nos. 9032001 and 40335050).” Pedosphere 16 (4): 420–27. https://doi.org/10.1016/S1002- 0160(06)60071-4. 29 Annex Table A1: List of countries with global DHS data for rural non-migrant children, 2000-2020 Countries Region Sample Year Sample Size Average DD (%) Average Stunting (%) Albania Europe and Central Asia 2017 1,431 6.58 4.29 Angola Sub-Saharan Africa 2015 6,210 13.33 19.17 Armenia Europe and Central Asia 2015 661 4.89 5.30 Bangladesh South Asia 1999; 2004; 2007; 2017 2,298; 3,928; 3,225; 2,598 7.93 14.60 Benin Sub-Saharan Africa 2001; 2017 3,345; 7,196 12.50 12.61 Bolivia Latin America & the Caribbean 2008 3,539 28.19 11.39 Burkina Faso Sub-Saharan Africa 2003 7,974 19.92 Burundi Sub-Saharan Africa 2016 9,756 22.57 26.75 Cambodia East Asia and Pacific 2000; 2005 7,280; 5,754 20.18 19.34 Cameroon Sub-Saharan Africa 2004; 2018 3,908; 4,128 14.99 18.43 Colombia Latin America & the Caribbean 2010 5,118 17.14 5.35 Congo, Democratic Republic Sub-Saharan Africa 2007 4,582 16.29 27.93 Dominican Republic Latin America & the Caribbean 2007 4,298 17.18 3.32 Egypt, Arab Rep. Middle East and North Africa 2000, 2003, 2005, 2008 6,549; 3,395; 8,066; 6,324 13.90 12.77 Eswatini Sub-Saharan Africa 2006 1,555 14.92 9.51 Ethiopia Sub-Saharan Africa 2016 7,808 11.22 18.69 Gambia, The Sub-Saharan Africa 2019 3,982 20.31 3.64 Ghana Sub-Saharan Africa 2003; 2008 2,345; 1,580 19.56 10.54 Guinea Sub-Saharan Africa 2005; 2018 4,515; 5,282 14.55 15.69 Guyana Latin America & the Caribbean 2009 1,313 10.55 9.15 Haiti Latin America & the Caribbean 2000, 2005, 2016 4,101; 3,056; 3,840 22.98 8.08 India South Asia 2015, 2019 163,193; 114,699 8.67 17.82 Jordan Middle East and North Africa 2002; 2007. 2017 1,777; 2,983; 2,126 14.36 7.76 30 Countries Region Sample Year Sample Size Average DD (%) Average Stunting (%) Kenya Sub-Saharan Africa 2003, 2008; 2014 3,438; 3,747; 5,513 15.81 11.58 Lesotho Sub-Saharan Africa 2004; 2009 2,535; 2,382 13.53 15.44 Liberia Sub-Saharan Africa 2007; 2009; 2019 3,130; 2,004; 3,388 19.55 17.51 Madagascar Sub-Saharan Africa 2008 9,079 7.96 27.25 Malawi Sub-Saharan Africa 2000, 2004, 2010, 2015 8,049; 7,934; 15,201; 19.67 15.12 12,367 Mali Sub-Saharan Africa 2001; 2006; 2018 616; 9,412; 6,765 15.57 17.98 Moldova Europe and Central Asia 2005 608 4.18 Morocco Middle East and North Africa 2003 3,187 12.80 Namibia Sub-Saharan Africa 2000; 2006 2,014; 2,391 12.62 11.66 Nepal South Asia 2001; 2006; 2016 5,441; 3,787; 1,784 15.50 22.14 Nigeria Sub-Saharan Africa 2003; 2008; 2018 3,422; 18,224; 20,302 13.69 25.90 Peru Latin America & the Caribbean 2000; 2004; 2009 6,493; 5,907; 3,923 16.42 13.71 Philippines East Asia and Pacific 2003; 2008; 2017 3,110; 3,022; 5,742 8.34 Rwanda Sub-Saharan Africa 2005; 2019 5,739; 4,905 14.70 10.44 Senegal Sub-Saharan Africa 2005; 2008 6,197; 10,456 22.90 Sierra Leone Sub-Saharan Africa 2008; 2019 3,238; 5,894 8.75 15.50 South Africa Sub-Saharan Africa 2016 1,362 11.80 11.27 Tajikistan Europe and Central Asia 2017 3,875 13.48 5.15 Tanzania Sub-Saharan Africa 2007; 2015 4,977; 6,577 10.76 12.74 Timor-Leste East Asia and Pacific 2009; 2016 7,272; 4,626 11.85 30.70 Uganda Sub-Saharan Africa 2000; 2006; 2016 3,455; 4,857; 9,141 22.40 11.86 Zambia Sub-Saharan Africa 2007; 2013; 2018 3,315; 6,501; 5,138 15.44 17.10 Zimbabwe Sub-Saharan Africa 2005; 2015 3,398; 2,831 15.05 11.12 Total 6 Regions 91 Sample Surveys 724,092 12.8 17.3 Notes: The sample only includes children residing in rural clusters, whose households have either always lived in the current location or those that have migrated to their current location before the child was conceived (non-migrant sample). Region is classified based on World Bank country and lending groups. Sample year is when the survey is recorded and may not necessarily equal survey year as the year of the actual survey may vary between households. 31 Table A2: Summary Statistics for Global DHS data on Health outcomes and Forest Loss, 2000-2020 Variable Mean Std. Dev. N Main Variables Diarrheal Disease (0: no, 1: yes) 0.1283 0.3344 663,214 Severe Stunting (0: no, 1: yes) 0.1726 0.3779 423,234 Upstream Forest Loss (20-year average) 0.0012 0.0028 724,092 Downstream Forest Loss (20-year average) 0.0011 0.0027 724,092 Within 5km Forest Loss (20-year average) 0.0012 0.0030 724,092 Control Variables Average Temperature (Degree Celsius) 23.3099 4.6739 721,002 Average Rainfall (Meters) 0.003706 0.002485 721,002 Improved Water (0: no, 1: yes) 0.7326 0.4426 724,092 Birth order (1-4) 1.3261 0.5487 724,092 Age of the mother when child born 26.6808 6.3558 724,092 Age of the child (in months) 28.9828 17.2385 724,092 Sex (0: male, 1: female) 0.4884 0.4999 724,092 Notes: This table shows the general summary statistics for the variables used in the regression analysis for rural non-migrant children under 60 months. Diarrheal Disease is a dummy variable for whether a child had diarrhea within two weeks of the survey, Severe Stunting is a dummy variable for a height-for-age Z score below 3 standard deviations, and forest loss represents the 20-year average of forest shares lost. For the control variables, Average Temperature shows the mean annual temperature in Celsius, Average Rainfall shows the total annual precipitation in meters, Improved Water is a dummy based on WHO/JMP definition of improved water sources, Sex is a dummy variable for a female child, and the age of the mother is reported in years while the age of the child is reported in months. 32 Table A3: Peru DHS data per year, 2004-2020 Number of clusters Non-migrant children of the study Year Rurala Urbanb Total Rural Urban Total 2004 89 129 218 3,136 2,956 6,092 2005 91 140 231 3,712 3,280 6,992 2006 85 144 229 2,598 2,958 5,556 2007 94 136 230 2,010 1,716 3,726 2008 33 43 76 317 293 610 2009 371 568 939 3,094 3,752 6,846 2010 367 570 937 2,730 3,359 6,089 2011 367 566 933 2,568 3,358 5,926 2012 371 807 1,178 3,030 4,292 7,322 2013 365 779 1,144 2,385 3,498 5,883 2014 388 892 1,280 2,436 4,008 6,444 2015 677 1,951 2,628 4,779 11,704 16,483 2016 672 1,941 2,613 3,935 10,521 14,456 2017 662 1,942 2,604 3,644 10,922 14,456 2018 682 2,013 2,695 4,513 11,629 16,142 2019 668 2,005 2,673 3,692 10,757 14,449 2020 633 1,834 2,467 3,441 8,636 12,077 Total 6,615 16,460 23,075 52,020 97,639 149,659 Notes: Rural clusters refer to localities with less than two thousand inhabitants. Urban clusters are localities with more than two thousand inhabitants. Data collected by the authors from “Peru Demographic and Health Survey” data. 33 Table A4: Summary Statistics for Peru DHS data on Health outcomes and Forest loss, 2004-2020 Variable Mean Std. Dev. N Main Variables Diarrheal Disease (0: no, 1: yes) 0.1262 0.3321 38,652 Severe Stunting (0: no, 1: yes) 0.0696 0.2544 33,683 Upstream Forest Loss (17-year average) 0.0020 0.0046 38,652 Downstream Forest Loss (17-year average) 0.0025 0.0057 38,652 Within 5km Forest Loss (17-year average) 0.0024 0.0057 38,652 Control Variables Average Temperature (Degree Celsius) 15.5561 7.0350 38,652 Average Rainfall (Meters) 0.0045 0.0025 38,652 Improved Water (0: no, 1: yes) 0.9122 0.2831 38,652 Birth order (1-4) 1.1761 0.4084 38,652 Age of the mother when child born 28.2991 7.2678 38,652 Age of the child (in months) 29.6639 17.1171 38,652 Sex (0: male, 1: female) 0.4963 0.4999 38,652 Notes: See notes to Table A2 for details on variable construction. Table A5: Percent share of household’s source of drinking water by urban/rural and wealth in Peru Water source Urban Rural Urban Rich Urban Poor Rural Rich Rural Poor Improved drinking water 99.85 91.07 99.99 98.53 98.41 87.80 Unimproved drinking water 0.15 8.93 0.01 1.47 1.59 12.20 Total 100 100 100 100 100 100 Notes: See notes to Table 2 for details on variable construction. 34 Table A6: Impact of upstream forest loss on DD incidence (Logit, Probit), 2004-2020 (1) (2) (3) (4) Regression model (Logit) (Logit) (Probit) (Probit) Dependent variable: Incidence of Diarrheal Disease Upstream Forest Loss Year t-1 to t 6.809** 3.683** (2.947) (1.685) Year t-2 to t-1 -3.539 -2.209 (3.608) (1.969) Downstream Forest Loss Year t-1 to t 4.209 2.097 (2.713) (1.523) Year t-2 to t-1 0.647 0.344 (2.786) (1.559) N 496,025 496,025 496,025 496,025 Fixed Effects Y Y Y Y Climate Controls Y Y Y Y Demographic Controls Y Y Y Y Improved Water Y Y Y Y Notes: This table presents logit and probit estimates of equation (1) where the dependent variable is the DD incidence for non-migrant children under the age of 5 between 2004-2020. See notes to Table 3 for details on variable construction. Robust standard errors clustered at the province level are reported in parentheses. * p<0.10 ** p<0.05 *** p<0.01. 35 Figure A1: Diarrheal Disease and Stunting rates for non-migrant children by age and urban/rural area; Peru Sample Notes: See notes to Figure 2 for details on variable construction. 36