Policy Research Working Paper 10715 Environmental Hazards, Climate, and Health in Cambodia The Shield of Sanitation Sandra Aguilar-Gomez Sandra Baquie Paul Jacob Robyn Health, Nutrition and Population Global Practice March 2024 Policy Research Working Paper 10715 Abstract Environmental degradation is the largest public health trends, climate simulations are employed to forecast the challenge of the century and is likely to be exacerbated by incidence of child diarrhea in Cambodia under different climate change. This study undertakes a comprehensive climate and development scenarios. The projections indi- examination of the health implications of environmental cate that diarrhea incidence could increase to 19 percent by hazards in Cambodia, simultaneously addressing extreme 2040 without significant adaptation measures that would temperatures, precipitation patterns, and air pollution. It lessen the adverse impact of weather. For instance, the accel- leverages data from the Demographic and Health Surveys eration in toilet ownership over the last decade reduced and satellite-derived metrics on temperature, precipitation, diarrhea incidence by at least 1.2 to 1.4 percentage points. and fine particulate matter. The analysis identifies a pos- Nevertheless, the path ahead requires proactive efforts to itive association between temperature and the occurrence improve sanitation and hygiene. The forecasts suggest that, of diarrhea and cough among children and a nonlinear without additional strategies to counter climate change’s relationship between precipitation and these health out- adverse effects, only universal toilet ownership would comes. Furthermore, the study demonstrates that pollution contain the climate-driven increase in diarrhea incidence significantly impacts cough incidence. To anticipate future expected by 2040. This paper is a product of the Health, Nutrition and Population Global Practice. 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 sbaquie@worldbank.org. 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 Environmental Hazards, Climate, and Health in Cambodia: The Shield of Sanitation Sandra Aguilar-Gomez∗ Sandra Baquie† Paul Jacob Robyn† Keywords: climate change, environmental health, pollution, Cambodia JEL Codes: Q54, I1, O53 ∗ Department of Economics, Universidad de los Andes. Email: s.aguilargomez@uniandes.edu.co † World Bank. Corresponding authors. Email: sbaquie@worldbank.org The team is grateful for the excellent comments provided by Ajay Tandon, Marion Cros, Javier E. Baez, Katherine Anne Stapleton, Lander Bosch, Saroeun Bou, Stephen Geoffrey Dorey, and Ronald Upenyu Mu- tasa. Valentina Castilla-Gutierrez provided excellent research assistance. 1 Introduction Over recent decades, the escalating climate crisis has become increasingly apparent. This crisis is characterized by extreme conditions such as a continuous rise in Earth’s temper- ature and extended periods of both rainfall and drought. Greenhouse gas emissions that cause climate change are part of a broader process of environmental degradation, generating unhealthy levels of air pollutants, including particulate matter. The repercussions of these hazards —extreme temperature, extreme rainfall, and hazardous air— on health outcomes, including disease spread and long-term outcomes, have been examined for some countries in the scientific literature, although their health impacts are often studied in isolation from one another. Climate change will likely increase the frequency and severity of extreme weather events and, consequently, their impact on health and human capital. The latest report from the Intergovernmental Panel on Climate Change (IPCC) highlights that even under the 2°C sce- nario, there will be a definitive rise in average temperatures across all regions (IPCC, 2022). This warming trend is accompanied by projected changes in temperature and precipitation patterns, resulting in an amplified risk of either drought or intense precipitation events (Arias et al., 2021). This escalating health burden caused by climate change has become a grow- ing concern for health agencies and policymakers worldwide. As extreme conditions persist and intensify, it is crucial for researchers and policy makers to produce regional evidence of weather impacts on health and to assess the effectiveness of adaptation actions. Recent estimates show that nearly 1,000 children under five die every day as a result of diarrhea caused by a lack of access to clean water and sanitation (World Health Organization, 2014). Providing water and sanitation, with associated sustained behavior change, is a critical arm of health policy. However, public health studies quantifying the role of sanitation in mitigating climate-related disease are scarce and often mixed (Howard et al., 2016; Carlton et al., 2014).1 In this study, we quantify the potential impact of climate change on a selected health outcome, diarrhea incidence in children, and assess the effectiveness of a sanitation policy, increasing toilet ownership, in mitigating the impact of weather shocks and adapting to climate change. 1 For the specific case of Cambodia, Howard et al. (2016) suggested that poor water quality and sanitation were responsible for the statistically significant impact of floods on diarrheal disease in 2 of 16 provinces in Cambodia. Similarly, Carlton et al. (2014) found that rainfall was associated with diarrhea incidence in Ecuador, with drinking water treatments decreasing the impact of rainfall but sanitation and hygiene having no impact. 1 The context of our study is Cambodia, a country with a tropical monsoon climate en- compassing distinct wet and dry seasons and a high incidence of floods and droughts. In Cambodia, national-level climate change models predict a considerable increase in mean an- nual temperature, ranging between 0.7° and 2.7° by 2060 and 1.4° and 4.3° by 2100. Mean annual rainfall is expected to rise, particularly during the wet season (Thoeun, 2015). The associated extreme weather is likely to influence the prevalence and transmission of various diseases, particularly water-borne illnesses2 like diarrhea (Davies et al., 2014). We study how weather shocks impact diarrhea incidence in this paper. Although Cambodia’s share of global greenhouse gas (GHG) emissions was only 0.14 percent in 2019, they grew at an av- erage annual rate of 8 percent from 2010 to 2019 (World Bank, 2023b). The country’s rapid industrialization and urbanization have increased emissions and the other face of the same coin, air pollution levels, creating a pressing need to investigate the relationship between pollution and health outcomes.3 In what follows, we choose air pollution’s impact on cough in children since lower respiratory infections are the second cause of death in Cambodia (IHME, 2023). Studying the link between climate change and health is even more important in Cambodia, where increased morbidity is likely to have significant consequences on house- holds’ finances, particularly for the poorest. Indeed, out-of-pocket expenditure represents 60.6% of current health expenditure (WHO, 2023; World Bank, 2024). This study leverages high-resolution satellite data measuring environmental hazards, as well as geo-referenced individual-level indicators from the Demographic and Health Surveys. We use flexible binned specifications that simultaneously consider the impact of temperature, precipitation, and air pollution. The latter is a crucial dimension of environmental degra- dation and one of the negative consequences of transport and industrial emissions. We also assess the effectiveness of adaptation actions for children’s diarrhea incidence by studying their effectiveness in mitigating the adverse effect of weather on diarrheal diseases. Finally, we employ climate simulations to project the incidence of child diarrhea in Cambodia under different climate and development scenarios. We consider two extreme climate scenarios, also called Shared Socioeconomic Pathways (SSPs): SSP1.19, characterized by very low green- house gas emissions, and SSP3.70, marked by high emissions. Combining our estimates of the impact of weather and toilet ownership on diarrhea incidence with the IPCC climate scenarios, we forecast the effectiveness of different toilet ownership scenarios in reducing the 2 There are several water-related diseases with a moderate-to-high incidence in Cambodia: diarrheal diseases, typhoid fever, leptospirosis, melioidosis, viral hepatitis, and schistosomiasis (McIver et al., 2016a). 3 The share of people in urban areas in Cambodia has grown steadily since the 1980s. In 2022, the percentage of people living in urban areas was 25.1% in Cambodia; it was only 10% in 1980 (World Bank, 2023a). 2 effects of climate change on diarrhea incidence. Our findings reveal substantial impacts of weather on child health outcomes in Cambodia. Our flexible binned approach shows a linear relationship between temperature and diarrhea incidence in children. Specifically, a 13.25 percentage point increase in diarrhea incidence is observed at temperatures above 35◦ C (90% of the sample mean), using temperatures lower than 29◦ C as the comparison group. Previous literature suggests this can be attributed to the accelerated proliferation of disease-causing agents in warmer conditions and the increased risk of food contamination (Semenza et al., 2012; Zhou et al., 2013). Similarly, our analysis uncovers a non-linear association between precipitation and diarrhea, with high precipitation levels leading to substantial increases in incidence rates. When examining cough incidence, we find suggestive evidence that temperature exhibits a non-linear effect, with only the highest temperature range significantly impacting the incidence of cough among children. Consistently with the literature (IHME, 2023), our results also highlight the substantial influence of air pollution, particularly a 12.4 percentage point increase in cough incidence, equivalent to 50% of the sample mean, in areas with higher pollution levels. Our study delves into potential adaptation measures to mitigate the adverse effects of projected temperature and precipitation shifts on health outcomes. First, we find that households with access to toilets and refrigerators experience a considerable reduction in diarrhea incidence. Specifically, the accelerated ownership of toilets between 2014 and 2021 contributed to reducing diarrhea incidence by at least 1.2-1.4 percentage points, representing 20% of the overall reduction in diarrhea incidence over 2014-2021 (6.4 percentage points). Additionally, households with higher education levels demonstrate a reduced vulnerability to temperature-related diarrhea incidence. Our simulations indicate that the 2014-2021 acceleration in toilet ownership has been as- sociated with a significant reduction in diarrhea incidence. However, even under the lowest- emission scenario, if toilet ownership does not increase or is not complemented by further improvements in water and sanitation, diarrhea incidence will increase due to climate change. The incidence rate could reach nearly 19% of the population if no adaptation actions are taken, and the estimated rate is even higher in the more concerning high-emissions climate scenario. In both climate scenarios, increasing toilet ownership reduces the impact of climate change on diarrhea incidence by 1.7-2.1 percentage points by 2040. Crucially, without com- plementary investments in water and sanitation, reaching universal toilet ownership by 2025 is the only scenario among others assuming lower ownership rates that contain increased diarrhea incidence due to climate change until 2040. 3 This study contributes to three strands of literature. First, it expands the knowledge of the health consequences of environmental hazards in the tropics. Second, it examines the potential role of household resources and practices in mitigating these impacts. Third, the analysis evaluates the past and future role of sanitation infrastructure, particularly toilet ownership, in reducing climate-related health vulnerabilities. Regarding our first contribution, we add to the environmental health literature by exam- ining the link between weather extremes and the prevalence of diarrhea, respiratory symp- toms, and the spread of various diseases. The correlation between diarrhea and temperature has been studied in diverse contexts, finding that an increase in temperature, or prolonged dry seasons, is associated with a higher number of reported diarrhea cases (Amqam et al., 2019; Mertens et al., 2019; Xu et al., 2014; Zhou et al., 2013). At the other extreme, heavy rainfall has demonstrated similar effects. In countries such as India, Mozambique, Indonesia, and Peru, extended periods of intense rainfall have been associated with a higher incidence of diarrhea (Mertens et al., 2019; Horn et al., 2018). We expand this literature by examining child morbidity in Cambodia, an understudied country. The literature assessing the health impacts of weather shocks in Cambodia is scarce. Choi et al. (2016) conducted a study examining the influence of weather factors on dengue fever incidence in Cambodia, revealing a significant connection between higher temperatures and increased precipitation with elevated dengue fever cases. On the other hand, Saulnier et al. (2018) analyzed the impact of seasonal floods on health using six years of national health data and flood maps. Their study focused on the health consequences of flooding events. While these two studies examined separate hazards, our analysis jointly considers the effects of weather conditions, including temperature and precipitation (potentially flooding), on health outcomes in Cambodia. We add to the literature on the impacts of fine particulate matter (P M2.5 ), a common form of air pollution, which has been linked to respiratory infections, asthma, and low birth weight. Airborne particles are shown to cause inflammation and exacerbate pre-existing respiratory conditions such as asthma and chronic obstructive pulmonary disease (D’Amato et al., 2014; Tecer et al., 2008; Braun-Fahrl¨ ander et al., 1992). Additionally, elevated levels of particulate matter have been associated with a higher incidence of lung cancer and cardio- vascular diseases (Li et al., 2018; Shu et al., 2016). We show that symptoms associated with particulate pollution exposure can be significant and include cough, which can deteriorate the quality of life of children in already disadvantaged contexts such as many regions of Cambodia. 4 Our main contribution is to shed light on the solutions to environmental health challenges of the upcoming decades. We quantify the role of household-level tools for adaptation, underscoring the positive consequences of recent development policies in mitigating climate- induced disease. Our findings underscore the critical role of sanitation as an adaptation strategy in the face of escalating environmental challenges. However, the results also shed light on the urgency of a diversified portfolio of adaptation strategies to address the complex health risks posed by environmental factors in Cambodia. Without other strategies, only universal toilet ownership would temporarily contain the climate-driven increase in diarrhea incidence. The remainder of this paper is organized as follows: In Section 2, we delve into the dataset used for our analysis and provide descriptive statistics. Section 3 outlines our baseline empirical strategy. In Section 4 we present the impacts of weather on health and explore the heterogeneity of these impacts and potential climate adaptation actions. Section 5 shifts our focus to assessing the impact of climate change on diarrhea incidence and evaluating the effectiveness of toilet ownership as a climate adaptation strategy. Section 6 summarizes our key findings and discusses their broader implications. 2 Data The Demographic and Health Surveys (DHS) have been conducted in low- and middle-income countries worldwide for over three decades, providing nationally representative household survey data on a wide range of population, health, and nutrition indicators (Cambodia NIS, MOH, and ICF, 2023). The DHS is a repeated cross-section of geo-referenced data, allowing researchers to track spatial changes over time. We use the 1998, 2000, 2005, 2014, and 2021 Cambodia DHS rounds as our main source of information on individual-level health outcomes for children. The DHS questionnaire separately asks whether children have experienced these symptoms in the past two weeks, allowing us to assess recent health status accurately. To measure exposure to environmental hazards, we use two datasets, the Climate Haz- ards Group InfraRed Precipitation with Station data (CHIRPS) and its analogous version for temperature (CHIRTS), published by Funk et al. (2015). These data products have better resolution and are better calibrated for regions near the equator than existing alternatives, including ERA5r (Copernicus Climate Data, 2024), which is crucial for our study in Cam- bodia. Moreover, their daily frequency enables us to estimate the acute impacts of extreme 5 temperature and precipitation events accurately. In addition to climate data, we incorporate monthly pollution data from the Atmospheric Composition Analysis Group at Washington University in St. Louis (Van Donkelaar et al., 2021), which has a spatial resolution of 0.1 degrees and provides annual and monthly estimates of ground-level fine particulate matter (P M2.5 ). To establish hazard exposure, we leverage the geographic coordinates collected by the DHS. We create 2km and 5km buffers around urban and rural households, respectively, and then extract the average temperature, precipitation, and pollution levels from the satellite data within these predefined radii.4 We employ larger buffers for rural households, taking into account the random displacement of coordinates conducted by the DHS to ensure the privacy of respondents in sparsely populated areas. Our comprehensive dataset for analysis encompasses household-level covariates and health outcomes spanning the years 1998, 2000, 2005, 2014, and 2021, along with pollution exposure data covering the period from 1997 to 2021. We merge the DHS data with the temperature and precipitation data at the cluster, day, month, and year levels. The associated weather variables represent the average temperature and precipitation in the month leading to the interview. Air pollution data is at the monthly level. We merge it at the cluster, month, and year levels. The resulting variable measures air pollution in the month of the survey.5 Table 1 furnishes an extensive overview of key variables essential to our empirical analy- sis. In Panel A, environmental data, including mean daily maximum temperature, average monthly precipitation, and average monthly ground-level fine particulate matter (P M2.5 ), are presented. Panel B sheds light on health outcomes, specifically, diarrhea and cough prevalence. Notably, data on these health indicators were exclusively available for children, constituting 25.28% of the overall sample. For children aged 0-4, data on diarrhea was ac- cessible for nearly 99.81% of the 30,003 children in the sample, while data on cough was available for approximately 99.84% of the same group. Panel C offers insights into relevant covariates and household characteristics, encompassing toilet access (available for 100% of the sample), refrigerator ownership (accessible for approximately 99.97% of the sample), ed- ucation level, household size, and urban residence. Education in Cambodia DHS is defined for six levels: people with no, incomplete primary, complete primary, incomplete secondary, complete secondary, and higher education. Our dummy variable encompasses the last three. 4 The DHS includes a displacement of coordinates to preserve anonymity. It is standard practice in the literature to include buffer sizes corresponding to the DHS displacement to account for it. We apply this method in our study and set the size of the buffers to equal the size of the maximum displacement. 5 We focus on average air pollution in this paper. Air pollution peaks could also have acute consequences on health, but their study requires temporally disaggregated data that we do not have. 6 Table 1: Descriptive Statistics Mean SD Min Max N Panel A. Hazards Temperature (◦ C) 31.97 1.36 26.98 36.39 117,802 Precipitation (l/m2 ) 6.19 4.45 0 41.15 117,843 P M2.5 (µg/m3 ) 19.57 6.91 4.19 48.45 117,843 Panel B. Outcomes Diarrhea 0.16 0.37 0 1 29,946 Cough 0.26 0.43 0 1 29,956 Panel C. Covariates and household characteristics Toilet 0.41 0.49 0 1 118,669 Refrigerator 0.06 0.25 0 1 118,644 Education: higher than primary 0.31 0.46 0 1 118,669 Household size 2.96 1.49 1 21 118,669 Urban 0.74 0.43 0 1 118,669 Notes: The average daily maximum temperature, and average precipita- tion and average pollution during the month leading to the interview are used. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014 Our dataset spans four DHS rounds conducted in Cambodia from 2000 to 2014, enabling a comprehensive exploration of the intricate relationships between these variables and health outcomes. 3 Empirical Strategy Our regression model investigates the relationship between environmental hazards and child health in Cambodia, focusing on diarrhea and cough incidence. To that end, we create temperature, precipitation, and pollution bins using each month’s average maximum tem- perature, average precipitation, and average pollution. This binned specification is standard in the environmental economics literature (as reviewed in Deschenes, 2014; Dell et al., 2014). A binned specification allows us to flexibly assess the hazard-morbidity link, accounting 7 for non-linearities. Causal identification in this modeling framework hinges on the assump- tion that, once we account for baseline average climate conditions —through geographic fixed effects– and seasonality —through the inclusion of time fixed effects—, any residual fluctuations in environmental conditions can be considered unexpected and, consequently, exogenous to the system. In other words, we assume that these weather and shocks are not associated with unobserved factors that could potentially influence morbidity outcomes. While this assumption is more demanding without individual-level panel data, our access to the exact coordinates allows us to precisely impute exposure in the past month, and our estimates are robust to different sets of fixed effects. Equation 1 describes a linear probability model where the dependent variable, Yidt = 1 if child i in district t and date t experienced the symptom of interest (cough or diarrhea) in the past two weeks. 7 Yidt = βj,temp 1(tmaxidt ∈ (tb , tb ]) b=1 7 + βj,prcp 1(prcpidt ∈ (pb , pb ]) b=1 (1) 7 + βj,PM2.5 1(PM2.5idt ∈ (PM2.5b , PM2.5b ]) b=1 + ρt + τd + εidt where tb and tb are the lower and upper bounds for each temperature bin b = 1, . . . , 7. Similarly, pb and pb are the limits for each precipitation bin, and PM2.5 b and PM2.5 b are the thresholds for each PM2.5 bin. ρt represents month-year fixed effects, and τd represents district fixed effects. Our results are robust to the inclusion of province-specific yearly shocks. We use OLS estimation with cluster-robust standard errors to account for clustering at the village level. Appendix A displays the temperature, precipitation, and pollution bins used for all our specifications. The distributions for each hazard are also plotted in that appendix, with the 5th and 9th percentiles marked in red. We use these thresholds to delimit the lowest and highest bins, and then produce roughly-equal-sized intermediate exposure bins. The Poisson estimator has been proposed as an alternative to model dependent variables with non-negative values, including probabilities and count variables.6 However, it is in- 6 The PPML estimator performs well with a large number of zeros and over- or under-dispersion in the 8 compatible with fixed effects, which we include to tackle potential confounders. Thus, as a robustness check, we use a Poisson Pseudo-Maximum likelihood model (PPML), as proposed by Correia (2016) and Correia et al. (2019). In the results section, we present our baseline linear probability models. Figures in Appendix A.2 show that our estimates are robust to using linear probability models fixed-effects or PPML estimators. We build on specification 1 to assess the effectiveness of individual adaptation actions in mitigating diarrheal diseases in section 4.2. Potential actions influencing the incidence of diarrhea for which we have data in the DHS are toilet and fridge ownership, urban/rural indicators, and education level. For all of them, we create a dummy variable equal to 1 when the household owns the asset, lives in an urban area, or has at least some education, respectively. Then, we interact the bins of significant weather variables in equation 1 with indicators of adaptation actions. The coefficient on the interaction term assesses the effec- tiveness of the considered adaptation action in decreasing the impact of weather on diarrhea incidence in children. It is important to keep in mind that we do not have exogenous vari- ation in these characteristics when interpreting the results. As a result, households owning a specific asset or characteristic may disproportionately adopt other adaptation strategies that increase their resilience to climate-related risks. In Appendix A.4, we examine the long-term consequences of in-utero environmental haz- ard exposure on child health. We achieve this by imputing environmental variables – maxi- mum temperature, average pollution levels, and average precipitation – for each trimester of pregnancy. Our findings reveal no substantial impacts on reported birthweight or measured indicators of stunting or underweight among our dataset’s children. However, these findings are subject to limitations from the relatively small number of observations available for these specific variables and the challenges in precisely assigning exposure levels. Indeed, the pollu- tion exposure assignment during pregnancy is only accurate if households have not relocated too far since birth. Otherwise, the measurement error introduces a bias of the coefficient towards zero. In light of these considerations, we exercise caution as the DHS dataset may not be the optimal resource for examining the long-term effects of environmental hazards on child health. Nevertheless, we present these results transparently, aiming to offer insights and guidance for future researchers in this field. data (Silva and Tenreyro, 2011). 9 4 Baseline Results 4.1 Environmental hazards and health Figures 1 and 2 show the coefficients of Equation 1 applied to diarrhea incidence in children. Figure 1 shows temperature has a positive and significant relationship with diarrhea incidence in children. While we used a binned specification to test for non-linearity, results show that diarrhea incidence increases linearly with heat. All coefficients are highly significant and have a high magnitude. For instance, a 13.25 percentage point increase in diarrhea incidence, the coefficient of the 35◦ C bin in Figure 1, amounts to a 90% surge with respect to the sample mean. These results are consistent with insights from the epidemiology literature suggesting that some vectors and bacteria cause diarrhea to proliferate faster under warm temperatures. Under these conditions, food contamination increases as food spoils more quickly. Dietary and hygiene patterns may also change depending on the temperature (Semenza et al., 2012; Zhou et al., 2013). Figure 2 illustrates the relationship between precipitation and diarrhea incidence among children. This relationship exhibits nonlinearity, with increases at both ends of the precipita- tion distribution. It is worth mentioning that while low precipitation levels are theoretically associated with degraded water quality and an elevated risk of water-borne diseases, our data does not provide statistically significant evidence of this association at conventional signif- icance levels. However, high levels of precipitation significantly impact diarrhea incidence. This effect could be attributed to flooding, which can lead to sewage system overflow and contamination of water sources (McIver et al., 2016b; Davies et al., 2014). Temperature, precipitation, and air pollution also increase on the incidence of cough (Figures 3 to 5). The relationship between cough and temperature is non-linear; only the highest temperature bin significantly affects incidence. The average incidence of cough is 23.4%. Hence, the 16 percentage point coefficient of the > 35◦ C bin entails a change equiv- alent to 68% of the sample mean. High precipitation levels increase cough incidence, but the coefficient’s statistical significance is not robust to the inclusion of temperature and pollution controls (Figure 4, Panel B). On the other hand, the impact of air pollution on cough is significant even when controlling for the other hazards, as shown in Figure 5. The relationship is piecewise linear and less steep than the one with temperature. Still, the 12.4 percentage point coefficient of the highest pollution bin entails a change in cough incidence 10 Figure 1: Temperature and diarrhea incidence in children Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced diarrhea in the past two weeks. The average daily maximum tem- perature during the month leading to the interview is used to construct the treatment bins. Controls include month-by-year and province-fixed effects, as well as pollution and precipitation levels in the past month. Standard errors are clustered at the DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014. equivalent to 50% of the sample mean. 11 Figure 2: Precipitation and diarrhea incidence in children Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced diarrhea in the past two weeks. Average precipitation in the month leading to the interview is used to construct the treatment bins. Controls include month-by- year and province-fixed effects and temperature and pollution levels in the past month. Standard errors are clustered at the DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014. 12 Figure 3: Temperature and cough incidence in children Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced a cough in the past two weeks. The treatment bins are constructed with the average daily maximum temperature during the month leading to the interview. Controls include month-by-year and province-fixed effects and pollution and precipitation levels in the past month. Standard errors are clustered at the DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014. 13 Figure 4: Precipitation and cough incidence in children (a) Panel A. without temperature and pollution controls (b) Panel B. with temperature and pollution controls Notes: Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced a cough in the past two weeks. Average precipitation in the month leading to the interview is used to construct the treatment bins. Panel A.: Controls include month-by-year and province-fixed effects. Panel B. Controls include month-by-year and province-fixed effects and temperature and pollution levels in the past month. Standard errors are clustered at the DHS cluster level. 14 Figure 5: Pollution and cough incidence in children Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced a cough in the past two weeks. Average P M2.5 in the month leading to the interview is used to construct the treatment bins. Controls include month-by-year and province-fixed effects and temperature and precipitation controls. Standard errors are clustered at the DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014. 15 4.2 Heterogeneous effects and potential climate adaptation ac- tions In this section, we explore whether protective actions taken by households can reduce the adverse impact of weather on short-term outcomes. We previously showed that air pollution is the main driver of cough incidence. Unfortunately, the DHS has no data on adaptation actions related to air pollution health impacts. As a result, we focus on diarrhea incidence in what follows. Indeed, the DHS contains several variables describing individual actions that could influence diarrhea incidence: ownership of a toilet or refrigerator and education level. Figures 6 to 8 present the coefficient on the interaction of the dummy variables describing these actions and the temperature and precipitation bins. Figure 6 shows that households owning toilets face milder impacts of temperature on diarrhea cases, although the estimates’ precision varies with the bins. The reduction in diarrhea incidence is significant for high temperatures and of a similar magnitude. Owning toilets is also associated with a decrease in diarrhea incidence for high precipitation levels, although the estimates are too noisy to achieve statistical significance at any conventional level. Results on refrigerator ownership, presented in Figure 7, produce similar conclusions. Having a refrigerator is associated with lower impacts of temperature on diarrhea incidence, although in this case, the effects are proportionately larger at milder temperatures. The available data and our specifications do not allow us to recover the causal effect of ownership per se. It could be that owners are also more likely to display other behaviors, for instance, hand-washing, that also increase their resilience to climate shocks. Figure 8 illustrates the results of interacting temperature and precipitation with an in- dicator of having at least higher education. We observe that the impact of temperature is attenuated among educated households, aligning with findings in the literature that suggest education may enhance knowledge about health practices and potentially mitigate the ef- fects of extreme temperatures (higher levels of education have been robustly associated with better health, preventive behaviors are one of many proposed mechanisms (Davies et al., 2018)). Conversely, the impact of precipitation is not moderated by education. This result is consistent with the notion that the primary mechanisms behind the adverse effects of high precipitation, such as sewage systems’ overflow, may operate at a larger scale, beyond the individual’s control or education level. Our findings collectively underscore the significance of sanitation, hygiene, and food 16 safety practices in elucidating the complex relationship between weather conditions and diarrhea incidence. Education appears to play a pivotal role in mitigating the temperature- related risks associated with diarrhea, likely through improved information access and health- related behaviors. However, the persistent impact of precipitation suggests that broader infrastructure and environmental factors may continue to contribute to diarrhea incidence, emphasizing the importance of public health interventions to address climate-sensitive health outcomes. Figure 6: Impact of toilet ownership on the adverse effect of temperature and precipitation on diarrhea incidence. Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced diarrhea in the past two weeks. The average daily maximum temperature and average precipitation during the month leading to the interview are used to construct the treatment bins. We interact a dummy variable equal to 1 when the household owns a toilet with the weather bins. The coefficients of these interactions are presented in the figure. Controls include month-by-year and province-fixed effects. Standard errors are clustered at the DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014. 17 Figure 7: Impact of refrigerator ownership on the adverse effect of temperature and precip- itation on diarrhea incidence. Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced diarrhea in the past two weeks. The average daily maximum temperature and average precipitation during the month leading to the interview are used to construct the treatment bins. We interact a dummy variable equal to 1 when the household owns a fridge with the weather bins. The coefficients of these interactions are presented in the figure. Controls include month-by-year and province-fixed effects. Standard errors are clustered at the DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014. 18 Figure 8: Impact of having at least higher education on the adverse effect of temperature and precipitation on diarrhea incidence. Sample includes children aged 0-5 years old. The dependent variable is whether the child experienced diarrhea in the past two weeks. The average daily maximum temperature and average precipitation during the month leading to the interview are used to construct the treatment bins. We interact a dummy variable equal to 1 when the household has at least higher education with the weather bins. The coefficients of these interactions are presented in the figure. Controls include month-by-year and province-fixed effects. Standard errors are clustered at the DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, and 2014. In Cambodia, toilet and refrigerator ownership increased exponentially from 56% and 10% in 2014 to 90% and 30% in 2021. In the meantime, GDP per capita grew by 12.7%, potentially driving an important portion of the increases in toilet and fridge ownership in the country. The government’s commitment has also played a role in this development. For instance, the National Strategy for Rural Water Supply, Sanitation and Hygiene (2011-2025) (Food et al., 2011) aims to reach 100 percent coverage of rural sanitation services by 2025. To reach this goal, several Water, Sanitation, and Hygiene (WASH) programs have been launched across the country to improve access, the use of clean water, and the adoption of sanitation and hygiene practices. Cambodia’s National Sanitation Strategy and Plan guides efforts to improve sanitation and hygiene practices by raising awareness, promoting 19 behavior change, and providing access to sanitation facilities. As shown in this section, these programs could contribute to climate adaptation in Cambodia since they are likely to decrease the impact of temperature and precipitation on children’s diarrhea incidence. We assess this potential in the following section. 5 Assessing the effectiveness of toilet ownership for cli- mate change adaptation This section examines the potential effectiveness of increased toilet ownership in reducing the impact of climate change in Cambodia. First, we calculate the expected health burden of diarrhea on children in the country due to climate change. To do so, we combine our estimates of the impact of temperature and precipitation on diarrhea incidence with the outcomes of all climate models participating in the Coupled Model Intercomparison Project Phase 6 (CMIP6) of the World Climate Research Programme. CIMP6 is a global collaborative effort among climate modeling centers to simulate and project future climate conditions. Participating centers agreed on common climate scenarios, SSPs (Shared Socioeconomic Pathways), describing broad socioeconomic trends that could shape future society. These are intended to span the range of plausible futures. For our simulations, we consider two extreme scenarios: 1. SSP1.19: Very low GHG emissions and CO2 emissions declining to net zero around or after 2050, followed by varying levels of net negative CO2 emissions. 2. SSP3.70: High GHG emissions and CO2 emissions that roughly double from current levels by 2100. We calculate future precipitation and temperature by adding the climate model’s pre- dicted changes in precipitation or temperature between future years and the baseline (2015) to the observed weather data in that same year. This is a common approach in the economics literature (Auffhammer et al., 2013; Burke et al., 2015; Baquie and Foucault, 2023). Then, we calculate expected diarrhea incidence in future years with the estimated relationship between diarrhea, temperature, and precipitation. For that purpose, we use a continuous version of the binned specification presented in Figure 6 to allow all temperature and pre- cipitation variations to impact the outcome. Regression results are presented in Table A.2 20 and are consistent with the above results. Three different scenarios are considered to assess the effectiveness of increased toilet ownership in terms of climate adaptation: 1. Toilet ownership remains constant at the 2014 level (56%) 2. Toilet ownership plateaus after reaching its 2021 level (90%) 3. Toilet ownership continues to increase at the average increase rate over 2014-2021, implying 100% coverage by 2025. 5.1 Climate change impact on diarrhea incidence Figures 9a and 9b depict the outcomes of the diarrhea incidence simulations in the SSP1 and SSP3 climate scenarios. In these simulations, we assume that toilet ownership remains equal to the 2014 rate. We use these projections as a baseline in what follows to assess the impact of accelerated toilet ownership scenarios. Overall, both climate scenarios forecast an increase in diarrhea incidence due to climate change of approximately 2% by 2040. If no adaptation actions are taken, the incidence rate could be near 19% of the population. As expected, this predicted increase is more rapid and steady in the high GHG emission scenario (SSP3.70). It is important to note that these forecasts are a counterfactual scenario of diarrhea incidence without the intentional and substantial improvements in sanitation that Cambo- dia has implemented since 2014. In the future, other adaptation investments could further decrease the impact of climate change on diarrhea incidence, such as fridge ownership or access to clean water. Nevertheless, our findings emphasize the critical importance of devel- oping adaptation strategies and enhancing the capacities of healthcare systems. Countries, particularly those with already burdened health care systems, must be prepared to bear a heightened disease burden amid the challenges posed by climate change. 21 Figure 9: Projected diarrhea incidence in children under two climate scenarios (SSP1.19 and SSP3.70) assuming constant toilet ownership at the 2014 level 23.0 22.0 Projected diarrhea incidence (%) 21.0 20.0 19.0 18.0 17.0 16.0 2015 2020 2025 2030 2035 2040 Year (a) SSP1.19 23.0 22.0 Projected diarrhea incidence (%) 21.0 20.0 19.0 18.0 17.0 16.0 2015 2020 2025 2030 2035 2040 Year (b) SSP3.70 Note: Each small transparent grey line represents the projection obtained using the output from a given climate model. The ensemble allows us to measure climate uncertainty stemming from the limitations of climate models. The black line is the ensemble median, and the dashed grey lines represent the 10-90% confidence interval. 22 5.2 The effectiveness of increased toilet ownership for climate adaptation Figures 10a and 10b present climate simulations in three different toilet ownership scenarios: the 2014 baseline (black), a scenario assuming a plateau after 2021 (blue), and a scenario in which toilet ownership reaches 100% by 2025 (green). For clarity, only the medians and 10%-90% confidence intervals are plotted. Comparing the values of the black and blue/green curves in 2021, we show that the acceleration in toilet ownership between 2014 and 2021 likely contributed to reducing diarrhea incidence by at least 1.2-1.4 percentage points despite climate change. This represents 20% of the overall reduction in diarrhea incidence over 2014- 2021 (6.4p.p). In both climate scenarios, accelerating toilet ownership reduces the impact of climate change on diarrhea incidence by 1.7-2.1 percentage points by 2040. The blue curves show that, in the absence of other adaptation actions, maintaining the 2021 toilet ownership level will not be enough for diarrhea incidence in 2040 to be lower than the 2014 level, which is already high compared to desirable standards. Reaching universal coverage (green) by 2025 is the only scenario in which toilet adaptation is sufficient to contain the increase in diarrhea incidence due to climate change by 2040. Finally, the effectiveness of toilet ownership for climate adaptation varies across space. Indeed, each region has a share of people owning toilets, and they are differently impacted by adverse weather events influencing diarrhea rates. Figure 11 presents the spatial distribution of the projected decrease in the 2040 diarrhea rate due to increased toilet ownership under climate scenario SSP3.70. Spatial disparities stem from initial spatial differences in toilet ownership and geographical differences in precipitation and temperature increases under climate change. Panel A shows how the fast increase in ownership of toilets between 2014 and 2021 has strengthened climate adaptation in some provinces. Expected diarrhea incidence in 2040 was reduced by 0.5 p.p in Kandal and 3.1 p.p in Preah Sihanouk. Reaching 100% toilet ownership could further reduce diarrhea rates in 2040 by 0.68 p.p in Preah Sihanouk. 23 Figure 10: Projected diarrhea incidence in children under two climate scenarios for three different levels of toilet ownership (2014, 2021, and 100%) 21 Projected diarrhea incidence (%) 20 19 18 17 16 2015 2020 2025 2030 2035 2040 Year SSP1.19 - 2014 toilet adoption SSP1.19 - 2021 toilet adoption SSP1.19 - 100% toilet adoption (a) SSP1.19 21 Projected diarrhea incidence (%) 20 19 18 17 16 15 2015 2020 2025 2030 2035 2040 Year SSP3.70 - 2014 toilet adoption SSP3.70 - 2021 toilet adoption SSP3.70 - 100% toilet adoption (b) SSP3.70 Note: Lines of the same color represent the same climate ensemble. They allow us to communicate climate uncertainty stemming from the limitations of climate models. The plain line is the ensemble median, and the dashed ones represent the 10-90% confidence interval. Black refers to the first scenario (2014 toilet ownership), blue to the second (2021 toilet ownership), and green to the third (100% ownership by 2025). 24 Figure 11: Spatial distribution of the decrease in projected diarrhea rate by 2040 in SSP3.70 under different scenarios of toilet ownership (a) 2021 level of toilet ownership compared to the 2014 level (b) 100% toilet ownership compared to the 2021 level Note: The maps represent the spatial distribution of the projected decrease in diarrhea rate in 2040 due to toilet ownership under climate scenario SSP3.70. Panel A represents the decrease due to the 2014-2021 toilet ownership scenario. Panel B shows the expected additional decrease in the 100% toilet ownership scenario compared to 2021 ownership levels. 25 6 Conclusion Environmental conditions play a crucial role in health. Vulnerable groups, such as children and the elderly, are particularly susceptible to the impacts of environmental hazards. This paper examines the impacts of three common environmental hazards, extreme temperature, extreme precipitation, and air pollution, in the context of Cambodia, a tropical country with expected increases in temperature and precipitation under most climate scenarios. We combine the Demographic and Health Surveys (DHS) with satellite-based information on temperature, precipitation, and fine particulate matter (P M2.5 ) to uncover the impact of environmental hazards on children’s diarrhea and cough. We find a positive impact of temperature on diarrhea and cough among children. Results also show a significant impact of extreme precipitation on diarrhea. Furthermore, our study demonstrates that pollution significantly impacts cough incidence. In Cambodia, out-of-pocket expenditure is high and amounts to 60.6% (WHO, 2023). As a consequence, the increase in morbidity due to weather shocks is likely to represent a significant cost for households that could be estimated in future work. Simulations are employed to forecast the incidence of diarrhea in Cambodia under dif- ferent climate and development scenarios. Our projections suggest that future diarrhea incidence among children could reach 19% by 2040 unless significant adaptation measures are taken. The acceleration in toilet ownership until 2021 has significantly decreased diarrhea incidence by at least 1.2-1.4 percentage points. However, results show that more remains to be done. Indeed, without further investments in water and sanitation or complementary climate adaptation, only a universal toilet ownership scenario would be sufficient to contain the increase in diarrhea incidence due to climate change by 2040. Hence, our findings un- derscore the need for proactive measures to mitigate the projected climate-driven increase in diarrhea incidence. More generally, there is a need to adopt comprehensive strategies to mitigate the health impacts of climate change in countries like Cambodia. 26 References H. Amqam et al. Relationship between climate and diarrhoea. Proceedings of the Proceedings of the 3rd International Conference on Environmental Risks and Public Health, ICER-PH 2018, 26-27, October 2018, Makassar, Indonesia, 2019. URL https://doi.org/10.4108/ eai.26-10-2018.2288628. Preprint. P.A. Arias, N. Bellouin, E. Coppola, R.G. Jones, G. Krinner, J. Marotzke, V. Naik, M.D. Palmer, G.-K. Plattner, J. Rogelj, M. Rojas, J. Sillmann, T. Storelvmo, P.W. Thorne, B. Trewin, K. Achuta Rao, B. Adhikary, R.P. Allan, K. Armour, G. Bala, R. Barimalala, S. Berger, J.G. Canadell, C. Cassou, A. Cherchi, W. Collins, W.D. Collins, S.L. Connors, S. Corti, F. Cruz, F.J. Dentener, C. Dereczynski, A. Di Luca, A. Diongue Niang, F.J. Doblas-Reyes, A. Dosio, H. Douville, F. Engelbrecht, V. Eyring, E. Fischer, P. Forster, B. Fox-Kemper, J.S. Fuglestvedt, J.C. Fyfe, N.P. Gillett, L. Goldfarb, I. Gorodetskaya, J.M. Gutierrez, R. Hamdi, E. Hawkins, H.T. Hewitt, P. Hope, A.S. Islam, C. Jones, D.S. Kaufman, R.E. Kopp, Y. Kosaka, J. Kossin, S. Krakovska, J.-Y. Lee, J. Li, T. Maurit- sen, T.K. Maycock, M. Meinshausen, S.-K. Min, P.M.S. Monteiro, T. Ngo-Duc, F. Otto, I. Pinto, A. Pirani, K. Raghavan, R. Ranasinghe, A.C. Ruane, L. Ruiz, J.-B. Sall´ee, B.H. Samset, S. Sathyendranath, S.I. Seneviratne, A.A. S¨orensson, S. Szopa, I. Takayabu, A.- M. Tr´eguier, B. van den Hurk, R. Vautard, K. von Schuckmann, S. Zaehle, X. Zhang, and K. Zickfeld. Technical Summary, page 33144. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 2021. doi: 10.1017/9781009157896.002. Maximilian Auffhammer, Solomon M Hsiang, Wolfram Schlenker, and Adam Sobel. Using weather data and climate model output in economic analyses of climate change. Review of Environmental Economics and Policy, 2013. Sandra Baquie and Guillem Foucault. Background note on bringing climate change into vulnerability analysis. World Bank, EFI Notes, 2023. ander et al. Air pollution and respiratory symptoms in preschool children. C. Braun-Fahrl¨ American Review of Respiratory Disease, 145(1):42–47, 1992. URL https://doi.org/ 10.1164/ajrccm/145.1.42. Marshall Burke, John Dykema, David B Lobell, Edward Miguel, and Shanker Satyanath. Incorporating climate uncertainty into estimates of climate change impacts. Rev. Econ. Stat., 97(2):461–471, May 2015. Cambodia NIS, MOH, and ICF. Cambodia demographic and health surveys, 2023. Title of the publication associated with this dataset: Cambodia Demographic and Health Surveys (2023). Phnom Penh, Cambodia, and Rockville, Maryland, USA: NIS, MoH, and ICF. Elizabeth J Carlton, Joseph NS Eisenberg, Jason Goldstick, William Cevallos, James Trostle, and Karen Levy. Heavy rainfall events and diarrhea incidence: the role of social and environmental factors. American journal of epidemiology, 179(3):344–352, 2014. Youngjo Choi, Choon Siang Tang, Lachlan McIver, Masahiro Hashizume, Vibol Chan, Ra- bindra Romauld Abeyasinghe, Steven Iddings, and Rekol Huy. Effects of weather factors 27 on dengue fever incidence and implications for interventions in cambodia. BMC Public Health, 16:241, March 2016. Copernicus Climate Data. ERA5, 2024. Sergio Correia. Linear models with high-dimensional fixed effects: An efficient and feasible estimator. Working Paper, 2016. Sergio Correia, Paulo Guimar˜aes, and Thomas Zylkin. ppmlhdfe: Fast poisson estimation with high-dimensional fixed effects. arXiv preprint arXiv:0000.0000, 2019. Gennaro D’Amato et al. Climate change and respiratory diseases. European Respiratory Review, 23(132):161–169, 2014. Grace I Davies, Lachlan McIver, Yoonhee Kim, Masahiro Hashizume, Steven Iddings, and Vibol Chan. Water-borne diseases and extreme weather events in cambodia: review of impacts and implications of climate change. Int. J. Environ. Res. Public Health, 12(1): 191–213, December 2014. Neil M Davies, Matt Dickson, George Davey Smith, Gerard J Van Den Berg, and Frank Windmeijer. The causal effects of education on health outcomes in the uk biobank. Nature human behaviour, 2(2):117–125, 2018. Melissa Dell, Benjamin F Jones, and Benjamin A Olken. What do we learn from the weather? the new climate-economy literature. Journal of Economic literature, 52(3):740–798, 2014. Olivier Deschenes. Temperature, human health, and adaptation: A review of the empirical literature. Energy Economics, 46:606–619, 2014. Food, Agriculture Organization of the United Nations, Food, and [FAO] Agriculture Or- ganization of the United Nations, 2011. URL https://faolex.fao.org/docs/pdf/ cam159126.pdf. Chris Funk, Pete Peterson, Martin Landsfeld, Diego Pedreros, James Verdin, Shraddhanand Shukla, and Joel Michaelsen. The climate hazards infrared precipitation with stations—a new environmental record for monitoring extremes. Scientific data, 2(1):1–21, 2015. Lindsay M Horn, Anjum Hajat, Lianne Sheppard, Colin Quinn, James Colborn, Maria Fer- nanda Zermoglio, Eduardo S Gudo, Tatiana Marrufo, and Kristie L Ebi. Association between precipitation and diarrheal disease in mozambique. International journal of en- vironmental research and public health, 15(4):709, 2018. Guy Howard, Roger Calow, Alan Macdonald, and Jamie Bartram. Climate change and water and sanitation: Likely impacts and emerging trends for action. Annual Review of Environ- ment and Resources, 41(1):253–276, 2016. doi: 10.1146/annurev-environ-110615-085856. URL https://doi.org/10.1146/annurev-environ-110615-085856. IHME. Cambodia: Top 10 causes of death. https://www.healthdata.org/ research-analysis/health-by-location/profiles/cambodia, 2023. Accessed: 2023- 11-29. 28 IPCC. Climate change 2022: Impacts, adaptation, and vulnerability. contribution of working group II to the sixth assessment report of the intergovernmental panel on climate change. Technical report, IPCC, Cambridge University Press, Cambridge, UK and New York, NY, USA, 2022. Ruyi Li, Rui Zhou, and Jiange Zhang. Function of pm2. 5 in the pathogenesis of lung cancer and chronic airway inflammatory diseases. Oncology letters, 15(5):7506–7514, 2018. Lachlan J McIver, Vibol S Chan, Kathyrn J Bowen, Steven N Iddings, Kol Hero, and Piseth P Raingsey. Review of climate change and Water-Related diseases in cambodia and findings from stakeholder knowledge assessments. Hawaii J. Med. Public Health, 28(2):49S–58S, 2016a. Lachlan J McIver, Chisato Imai, Petra G Buettner, Paul Gager, Vibol S Chan, Masahiro Hashizume, Steven N Iddings, Hero Kol, Piseth P Raingsey, and K Lyne. Diarrheal diseases and climate change in cambodia. Asia. Pac. J. Public Health, 28(7):576–585, October 2016b. Andrew Mertens, Kalpana Balakrishnan, Padmavathi Ramaswamy, Paramasivan Rajkumar, Prabhakar Ramaprabha, Natesan Durairaj, Alan E Hubbard, Ranjiv Khush, John M Col- ford Jr, and Benjamin F Arnold. Associations between high temperature, heavy rainfall, and diarrhea among young children in rural tamil nadu, india: a prospective cohort study. Environmental health perspectives, 127(04):047004, 2019. Dell D Saulnier, Claudia Hanson, Por Ir, Helle M¨ olsted Alvesson, and Johan von Schreeb. The effect of seasonal floods on health: Analysis of six years of national health data and flood maps. Int. J. Environ. Res. Public Health, 15(4), April 2018. Jan C Semenza, Susanne Herbst, Andrea Rechenburg, Jonathan E Suk, Christoph H¨ oser, Christiane Schreiber, and Thomas Kistemann. Climate change impact assessment of food- and waterborne diseases. Crit. Rev. Environ. Sci. Technol., 42(8):857–890, April 2012. Yang Shu, Liucun Zhu, Fei Yuan, Xiangyin Kong, Tao Huang, and Yu-Dong Cai. Analysis of the relationship between pm2. 5 and lung cancer based on protein-protein interactions. Combinatorial chemistry & high throughput screening, 19(2):100–108, 2016. Jose S Silva and Silvana Tenreyro. Further simulation evidence on the performance of the poisson pseudo-maximum likelihood estimator. Economics Letters, 112(2):220–222, 2011. Lokman Hakan Tecer, Omar Alagha, Ferhat Karaca, G¨ urdal Tuncel, and Nilufer Eldes. Particulate matter (pm2. 5, pm10-2.5, and pm10) and children’s hospital admissions for asthma and respiratory diseases: a bidirectional case-crossover study. Journal of Toxicology and Environmental Health, Part A, 71(8):512–520, 2008. Heng Chan Thoeun. Observed and projected changes in temperature and rainfall in cambo- dia. Weather and Climate Extremes, 7:61–71, 2015. 29 Aaron Van Donkelaar, Marc S Hammer, Lyndsey Bindle, Michael Brauer, Jeffrey R Brook, Michael J Garay, and Randall V Martin. Monthly global estimates of fine particulate matter and their uncertainty. Environmental Science & Technology, 55(22):15287–15300, 2021. WHO. Out-of-pocket expenditure as percentage of current health expenditure (CHE) (%) - data by country, April 2023. World Bank. Urban population (%). https://genderdata.worldbank.org/indicators/ sp-urb-totl-in-zs/?gender=total, 2023a. Accessed: 2023-10-17. World Bank. Cambodia country climate and development report. Washington, D.C. : World Bank Group., 2023b. World Bank. Cambodia Public Finance Review: From Spending More to Spending Better. Washington, DC: World Bank., 2024. [WHO] World Health Organization. Preventing diarrhoea through better water, sanitation and hygiene: exposures and impacts in low-and middle-income countries. In Preventing diarrhoea through better water, sanitation and hygiene: exposures and impacts in low-and middle-income countries. 2014. Zhiwei Xu, Yang Liu, Zongwei Ma, Ghasem Toloo, Wenbiao Hu, and Shilu Tong. Assessment of the temperature effect on childhood diarrhea using satellite imagery. Scientific reports, 4(1):5389, 2014. Xiaodan Zhou, Yanbing Zhou, Renjie Chen, Wenjuan Ma, Haiju Deng, and Haidong Kan. High temperature as a risk factor for infectious diarrhea in shanghai, china. Journal of Epidemiology, 23(6):418–423, 2013. 30 A Appendix A.1 Distribution of environmental hazards Table A.1: Temperature, Precipitation, and P M2.5 Bins Bin Temperature Precipitation P M2.5 (1 month average) Lower Upper Lower Upper Lower Upper bound bound bound bound bound bound 1 27ºC 29ºC 0 l/m2 0.5 l/m2 5.1 µg/m3 6.5 µg/m3 2 29ºC 31ºC 0.5 l/m2 1 l/m2 6.5 µg/m3 11 µg/m3 3 31ºC 33ºC 1 l/m2 5 l/m2 11 µg/m3 15 µg/m3 4 33ºC 35ºC 5 l/m2 10 l/m2 15 µg/m3 20 µg/m3 5 35ºC 36.4ºC 10 l/m2 15 l/m2 20 µg/m3 25 µg/m3 6 15 l/m2 20 l/m2 25 µg/m3 30 µg/m3 7 20 l/m2 41 l/m2 30 µg/m3 49 µg/m3 Notes: For the estimation of short-term models, the bins have these ranges. For temperature, the measurement unit is Celsius degrees (ºC), for precipitation the mea- surement unit is liters per square meters and for Particulate Matter (P M2.5 ) the measurement units are micrograms per cubic meter. 31 Figure A.1: Temperature distribution Distribution of the average daily maximum temperature during the month leading to the interview. Temperature is estimated using 5km and 2km buffers for rural and urban household coordinates, respectively. Gridded estimates (ºC) obtained from Funk et al. (2015). 32 Figure A.2: Precipitation distribution Distribution of the average daily total precipitation during the month leading to the interview. Precipitation is estimated using 5km and 2km buffers for rural and urban household coordinates, respectively. Gridded estimates obtained from Funk et al. (2015). 33 Figure A.3: Pollution distribution Distribution of the average P M2.5 concentrations during the month leading to the interview. Pollution is estimated using 5km and 2km buffers for rural and urban household coordinates, respectively. Gridded estimates obtained from Funk et al. (2015). 34 A.2 Robustness check: Poisson specification Figure A.4: Temperature and diarrhea incidence in children. Sample includes children aged 0-5 years old. Dependent variable is whether the child experienced diarrhea in the past two weeks. The average of the daily maximum tem- perature during the month leading to the interview is used to construct the treatment bins. Controls include month-by-year and province fixed effects, as well as pollution and precipitation levels in the past month. Standard errors at clustered at DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, 2014 and 2021. 35 Figure A.5: Temperature and cough incidence in children Sample includes children aged 0-5 years old. Dependent variable is whether the child experienced cough in the past two weeks. The average of the daily maximum tempera- ture during the month leading to the interview is used to construct the treatment bins. Controls include month-by-year and province fixed effects, as well as pollution and pre- cipitation levels in the past month. Standard errors at clustered at DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, 2014 and 2021. 36 Figure A.6: Precipitation and diarrhea incidence in children Sample includes children aged 0-5 years old. Dependent variable is whether the child experienced diarrhea in the past two weeks. Average precipitation in the month leading to the interview is used to construct the treatment bins. Controls include month-by-year and province fixed effects, as well as temperature and pollution levels in the past month. Standard errors at clustered at DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, 2014 and 2021. 37 Figure A.7: Precipitation and cough incidence in children (a) Panel A. (b) Panel B. Sample includes children aged 0-5 years old. Dependent variable is whether the child experienced cough in the past two weeks. Average precipitation in the month leading to the interview is used to construct the treatment bins. Panel A.: Con- trol include month-by-year and province fixed effects. Panel B.: Controls include month-by-year and province fixed effects, as well as temperature and pollution lev- els in the past month. Standard errors at clustered at DHS cluster level. 38 Figure A.8: Pollution and cough incidence in children Sample includes children aged 0-5 years old. Dependent variable is whether the child experienced cough in the past two weeks. Average P M2.5 in the month leading to the interview is used to construct the treatment bins. Controls include month-by-year and province fixed effects, as well as temperature and precipitation controls. Standard errors at clustered at DHS cluster level. The sample has information from 4 DHS rounds for Cambodia: 2000, 2005, 2010, 2014 and 2021. 39 A.3 Continuous specification Building on results from Figure 6, we replace bins by the following variables in the continuous specification: • Average monthly temperature • Its interaction with toilet ownership • A variable capturing an extreme specification. It is equal to 0 if monthly precipitation is below 15mm and to the precipitation amount when it is above it. • Its interaction with toilet ownership Table A.2: Continuous specification of the impact of temperature and precipitation on diar- rhea incidence Diarrhea incidence (=1) Temp 0.024*** (0.005) Toilet X Temp -0.001*** (0.000) Precip (if ≥ 15mm) 0.004*** (0.001) Toilet X Precip -0.002 (if ≥ 15mm) (0.002) Observations 28730 40 A.4 Long Run We also estimate the long-run impacts of environmental hazards on child health by estimating the impacts of maximum temperature, average pollution, and average precipitation during pregnancy7 as follows: 3 3 3 Yimdt = δq tmaxq i + αq prcpq i + θq P M2.5 q i + ρt + τd + µm + εimdt (2) q =1 q =1 q =1 The health outcome of interest is denoted by Yimdt and is measured for child i, son/daughter of mother m, in district d and time t. The independent variables of interest are the average temperature, precipitation, and pollution (P M2.5 ) levels during each trimester of pregnancy, denoted by tmaxq q q i , prcpi , and P M2.5 i , respectively, where q is an index that ranges from 1 to 3, representing the trimesters of pregnancy. The model also includes a battery of time-fixed effects (ρt ), as well as district (τd ), and mother (µm ) to control for unobserved heterogeneity, as well as an error term εimdt . Where standard errors are clustered at the DHS cluster level. The coefficients of interest are denoted by δq , αq , and θq , which represent the linear effects of temperature, precipitation, and pollution, respectively, during each trimester on the health outcome of interest.8 Table A.3 shows the results of our long-run analysis. The model finds no statistically significant effects of exposure to the three environmental hazards during pregnancy on rel- evant health measures. Analysis was also conducted using bins, but the results were not statistically significant either. There are many potential reasons why our short-run findings do not translate into long-run health effects. First of all, there are several sources of atten- uation bias due to measurement error. Birth weight is self-reported, and we observed a lot of anomalous values (for instance, mothers reporting live births weighing either 8kg or 1kg). There is also measurement error in assigning treatment because, as described above, we cannot observe whether women lived in their current location when the child was born. Our working assumption is that the woman’s location during pregnancy is the same as the loca- tion when interviewed and asked about the child. Finally, the sample size gets significantly reduced due to a large percentage of missing values for the dependent variables. Hence, we 7 We can only correctly assign pollution exposure in utero to children whose household did not migrate or move too far from their initial location, which represents an important limitation to our results. 8 Non-linear and non-parametric specifications were also analyzed, but since we found overall no impacts in the long run, here we report the most parsimonious econometric model. 41 conclude that due to data limitations, it is not possible to quantify the long-run impacts of pollution, temperature, or precipitation using the Cambodia DHS. For the estimations, we used dichotomous variables to accurately capture effects on ges- tation and child development. Low birth weight is defined as a variable indicating whether the reported birth weight of the child is below 2500 grams, according to the WHO. Stunting indicates whether the child, based on the reported age in months, falls below the WHO growth projections. Finally, Underweight indicates whether the child, based on the reported age in months, falls below the WHO weight-for-age projections. As mentioned earlier, the effect lacks statistical significance for most cases. 42 Table A.3: Long run effects of exposure to environmental hazards during pregnancy (1) (2) (3) (4) (5) (6) VARIABLES low birth weight stunting underweight Avg. Precipitation, third trimester 0.017 -0.005 -0.015 0.037 -0.011 0.020 (0.037) (0.022) (0.047) (0.030) (0.037) (0.027) Std. Precipitation, third trimester -0.005 -0.004 -0.012 -0.024 -0.004 -0.003 (0.027) (0.018) (0.042) (0.026) (0.027) (0.020) Avg. TMax, third trimester 0.034 -0.058 0.063 0.021 0.152 0.113* (0.087) (0.039) (0.083) (0.053) (0.096) (0.059) Std. TMax, third trimester -0.170 -0.138 0.044 0.041 -0.130 0.064 (0.152) (0.088) (0.204) (0.135) (0.187) (0.120) Avg. P M2.5 , third trimester 0.000 0.003 0.026 0.021 0.017 -0.001 (0.013) (0.008) (0.019) (0.016) (0.018) (0.011) Avg. P M2.5 , second trimester 0.011 0.006 -0.029* -0.018 -0.009 -0.015 (0.013) (0.008) (0.017) (0.012) (0.019) (0.013) Avg. P M2.5 , first trimester -0.004 -0.007 0.007 0.011 0.023 0.012 (0.015) (0.008) (0.018) (0.016) (0.018) (0.012) Observations 3,210 5,169 2,506 3,988 2,506 3,988 R2 0.854 0.822 0.918 0.865 0.920 0.880 Mother FE Yes No Yes No Yes No Household FE No Yes No Yes No Yes MOB-YOB-Province FE Yes Yes Yes Yes Yes Yes Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Sample includes children aged 0-5 years old. The creation of low birth weight, stunting, and underweight is based on the variables of birth weight, height-for-age, and weight-for-age. Birth weight is winsorized to account for anomalous reports. Height-for-age and weight-for-age are standard deviations from the mean. Low birth weight takes a value of 1 if a birth weight less than 2500 grams is reported, stunting takes a value of 1 if height-for-age is reported as negative, and underweight takes a value of 1 if weight-for-age is reported as negative. Standard errors at clustered at DHS cluster level. 43