Impact of Air Pollution on Mental Health in India

Abstract There is extensive evidence linking air pollution exposure to physical health. Less is known about the mental health impacts of poor air quality, especially in developing countries. We use data from India and estimate the causal impact of air pollution exposure on feeling sad, experiencing cognitive difficulties, and feeling unable to control and cope with important things in life. We instrument for air pollution exposure using the annual number of nighttime thermal inversions and show that air pollution exposure in the previous calendar year significantly worsens mental health in the current year. We examine potential mechanisms and find that air pollution exposure negatively impacts self-reported physical health, worsens respiratory conditions, and increases the likelihood of experiencing sleeping difficulties. Accounting for mental health impacts of pollution exposure is critical to accurately estimating the true health costs of air pollution and designing optimal environmental policy.


Introduction
The link between air pollution and physical health has been studied extensively with research showing effects on mortality (Lelieveld, Evans, Fnais, Giannadaki, & Pozzer, 2015;Pope, Ezzati, & Dockery, 2009;Pope et al., 1995) and cardiovascular problems (Hoek et al., 2013;Miller et al., 2007). Relatively little is known about the causal impacts of air pollution on mental health. This evidence gap is surprising given that mental disorders account for 7-13 percent of disabilityadjusted life-years (DALYs) and are one of the largest contributors to the global disease burden (Vigo, Thornicroft, & Atun, 2016). 1 The economic cost of mental health disorders including health care costs and lost productivity was $2. 5 trillion in 2010(World Health Organization, 2011. According to the World Health Organization (WHO), worldwide, more than 300 million people were impacted by mental disorders in 2015: an 18 percent increase since 2005.
Epidemiological research has shown that exposure to air pollution for a prolonged time can lead to the onset of depressive symptoms or a worsening of depression (Braithwaite, Zhang, Kirkbride, Osborn, & Hayes, 2019;Fan et al., 2020;Kim et al., 2016;Song et al., 2017).
In India, Lin et al. (2017) found associations between PM2. 5 and depression, while in China Chen, Oliva, and Zhang (2018) found that poor air quality significantly increased the likelihood of developing severe mental illness. Studies from South Korea and China have demonstrated that exposure to high levels of PM2.5 levels is positively correlated with the number of suicides (Kim et al., 2010;Lin et al., 2016). A study from the U.S. found that exposure to other pollutants like NO2 also impacted the likelihood of self-harm (Bakian et al., 2015). However, most of these epidemiological studies focus only on correlations between air pollution exposure and mental health and are unable to establish causality. This study fills this gap by examining the causal impacts of air pollution exposure on mental health in the large developing country context of India.
Mental health is a significant public health and economic concern in India. The total value of the economic loss due to mental health conditions in India between 2012 and 2030 is estimated to be approximately $1 trillion USD. 2 According to the WHO, the average suicide rate in 2015 in India was 15.7 for every 100,000 peoplesignificantly higher than the global average of 10.6. Suicide is one of the main causes of death in the 15-29 age group (Patel et al., 2012). Sagar et al. (2020) also find that the prevalence of mental disorders during adulthood increased between 1990 and 2017. Approximately 197 million people were diagnosed with mental health conditions in 2017, with 22 and 23 percent of them having depressive and anxiety disorders, respectively.
While WHO recommended safe air quality guidelines for PM2.5 are 10 mg/m 3 for maximum yearly average, between 2010 and 2017, India averaged about 90 mg/m 3 . 3 The economic cost of air pollution in terms of poor health and higher mortality is almost 8 percent of the GDP of India (World Bank, 2016). Air pollution accounts for almost 13 percent of all deaths in India (Greenstone & Hanna, 2014) with crop burning, vehicular emissions, dust storms due to construction activity, and industrial emissions being the most common sources of pollution (Bikkina et al., 2019). Given the existing mental health burden in India and the fact that pollution exposure in the past decade has been consistently above the recommended levels, it is critical to examine the extent to which air pollution exposure could be a risk factor for mental health in India.
In this context, we examine the causal impact of pollution exposure on mental health in India using survey data from the World Health Survey (WHS) and the Study on Global AGEing and Adult Health (SAGE). Isolating the causal impact of air pollution on mental health is challenging. Pollution monitoring is uneven in India with monitoring stations located mostly in urban areas (Dey et al., 2012). To address this concern, we use annual PM2.5 estimates from satellite monitoring data from Van Donkelaar et al. (2016). Further, it is likely that ordinary least square (OLS) estimates are downward biased because of omitted variable bias since both mental health and air pollution are correlated with economic activity. OLS estimates could also be downward biased because of reverse causality since mental health could itself impact economic productivity which in turn impacts air pollution levels (Haushofer & Fehr, 2014). To address these concerns, we use an instrumental variable (IV) strategy with the annual number of nighttime inversions as an instrument for air pollution exposure.
We find that exposure to air pollution in the year before the survey significantly impacts mental health. Respondents are significantly more likely to report feeling sad, low, or depressed in the last 30 days and having experienced cognitive difficulties in the last 30 days. Further, respondents are also more likely to report not being able to control the important things in their lives and cope with everything they have to do. We examine potential mechanisms through a formal mediator analysis of four mediator variables: overall physical health, poor respiratory conditions, changes in physical activity, and current work status. We find that overall physical health has the largest mediating impact on mental health. We also find that all results for the relationship between air pollution and mental health remain statistically significant even after controlling for these key mediator variables. This suggests that while air pollution impacts mental health by worsening physical health, it also likely has a direct effect.
This study addresses three main gaps in the literature. First, most of the current epidemiological literature focuses on the impacts of air pollution exposure on mental health in developed countries. This lacuna is striking since the impacts of air pollution exposure on mental health have important implications for developing countries. Developing countries have worse health outcomes, to begin with, and a higher baseline disease burden has been shown to increase susceptibility to adverse physical health impacts of air pollution (Clougherty & Kubzansky, 2009). Social safety nets in these countries are also often inadequate which leaves households more vulnerable to the physical and mental health impacts of air pollution exposure. Further, households in developing countries often do not have access to adequate information on pollution exposure. For example, while India has air pollution monitoring systems in urban areas, there is little evidence of these in rural areas. This lack of information impedes the ability of households to shield themselves from high levels of pollution. 4 Given that developing countries are increasingly likely to have high levels of particulate matter emissions (Alpert, Shvainshtein, & Kishcha, 2012), it is imperative to understand how pollution could impact mental health in these countries. Second, our study contributes to the limited empirical evidence on the mechanisms through which air pollution exposure can impact mental health. Previous studies have shown physiological responses to air pollution exposure including oxidative stress and inflammation in the brain (Power et al., 2015), which can increase the likelihood of depression, anxiety, and cognitive dysfunction (Salim, Chugh, & Asghar, 2012). In this study, we examine several indirect channels through which pollution can impact mental health including through impacts on physical health, physical activity, and labor productivity. Finally, most of the evidence on the impacts of pollution exposure on mental health comes from epidemiological studies which explore only correlations. Using an instrumental variable estimation strategy, this study adds to the literature on the causal impacts of air pollution exposure on mental health.
The rest of the paper is organized as follows -Section 2 describes the data and Section 3 outlines our main methodology. Section 4 discusses the main findings, while Section 5 explores potential mechanisms. Section 6 provides concluding remarks.

Mental health data
In this study, we combine data from the Study on Global AGEing and Adult Health (SAGE) and the World Health Survey (WHS) to create a panel dataset. The WHS was administered in 2003 and the SAGE in 2007 in six states in India: Assam, Karnataka, Maharashtra, Rajasthan, Uttar Pradesh, and West Bengal. These states were selected to be nationally representative. Both the WHS and the SAGE interviewed adults older than 18 years and include information on mental, physical, and emotional health in addition to information on demographics and household consumption and assets, with the 2007 SAGE survey being more comprehensive. In terms of respondent overlap between the two surveys, 4,600 adults aged 18-49 years from the WHS surveyed households were also included in the 2007 SAGE sample. Additionally, a cohort of WHS respondents aged 50 years and above was included in the SAGE survey.
We use four different measures of mental health as reported by individuals. We examine if an individual felt low, sad, or depressed in the past 30 days. We also estimate impacts on cognitive difficulties in the last 30 days (defined as being unable to concentrate or remember things); agency (measured through an indicator for being able to control important things in one's life); and stress (measured through inability to cope with all the things one has to do). 5,6 We restrict the sample to primary sampling units (PSUs) included in both the SAGE and WHS datasets (369). PSUs in urban areas are city wards and in rural areas are villages. Table 1 presents summary statistics. Our final sample includes 21,203 individuals older than 18 years. Respondents in 2003 and 2007 have a similar likelihood of feeling low, sad, or depressed in the last 30 days. They also have a similar likelihood of experiencing cognitive difficulties and being Impact of air pollution on mental health in India 135 unable to cope with all the things they have to do. Individuals in 2007 are significantly more likely to report being unable to control the important things in their life (p-value ¼ 0.02). The differences in the two survey rounds are likely because of differences in respondent selection in the two waves as described above. In all regression models, we control for a variety of demographic characteristics to account for these differences.

Pollution and weather data
Pollution is routinely monitored by the Central Pollution Control Board (CPCB) in India through nationwide pollution monitors. Most of these pollution monitors are placed in urban areas leading to low variation in pollution data (Dey & Di Girolamo, 2010). To counter the low spatial variation, we use satellite data on air pollution exposure using annual PM2.5 Ground-based PM2.5, at a resolution of 0.01degree grid cells is estimated using the geophysical relationship between surface PM2.5 and Aerosol Optical Depth (AOD) using geographically weighted regressions (Van Donkelaar et al. (2016)). Many studies have used these estimates to proxy for air quality, including in India (Hammer et al., 2020;Hansen-Lewis, 2018). We calculate annual PSU-level PM2.5 in micrograms per cubic meter (mg/m 3 ) using the coordinates for each PSU in our sample and merge the PM2.5 data to the SAGE and WHS datasets. Table 2 presents descriptive statistics for pollution, overall and separately by year. The mean PM2.5 in our sample is 46.05 mg/m 3 with a standard deviation of 18.53. 8 Out of the six states, Uttar Pradesh is the most polluted with a mean PM2.5 of 72. Karnataka is the least polluted with a mean PM2.5 of 21.10 (see Appendix Figure 1 in Supplementary Materials). This is unsurprising since the northern states of India are landlocked and, hence, more polluted than other parts due to high levels of crop burning (Hansen-Lewis, 2018).
Finally, in all regression models, we control flexibly for weather controls including precipitation, temperature, wind speed, and relative humidity and the squares of these variables. Data on precipitation and temperature are obtained in 0.5-degree grids from the University of Delaware, and data on wind speed and relative humidity are obtained from ECMWF-Interim. Descriptive statistics are reported in Table 2. Impact of air pollution on mental health in India 137

Empirical methodology
Our main goal is to estimate the causal relationship between air pollution and mental health. We first estimate an ordinary least squares (OLS) specification: where Y ipt is the outcome of interest for respondent i in PSU p in year t ¼ 2003, 2007. AirQuality ptÀ1 is ambient air quality defined by PM2.5 in PSU p in the year before the survey year t. We include PSU-level fixed effects, g p , to control for time-invariant PSU-level heterogeneity. We also control for survey year fixed effects, g t , to account for common trends in mental health, and month of interview fixed effects, g m , to account for seasonal effects correlated with both air pollution and mental health. X ipt includes individual and household level controls including dummies for ages 18-30, 30-40, 40-50, and older than 50; years of education; dummies for marital status (married, single, or widowed); a dummy for female; household size; and dummies for owning a mobile phone, bike/moped, and a computer. PSU-level controls, Z pt , include precipitation, temperature, wind velocity, relative humidity, and the squares of these variables. Finally, standard errors are clustered at the PSU level to allow for correlation of the error term within a PSU. OLS estimates are most likely downward biased because of omitted variable bias since both mental health and air pollution are correlated with economic activity. That is, it is possible that households in more polluted areas have higher incomes due to local economic conditions which could be positively correlated with mental health. Additionally, it is also possible that OLS estimates are downward biased because of reverse causality. Specifically, it is possible that mental health itself impacts economic productivity which in turn impacts air pollution levels (Haushofer & Fehr, 2014;Schilbach, Schofield, & Mullainathan, 2016). Finally, the validity of the OLS estimates is also threatened due to possible measurement error. Specifically, we observe ambient air quality and not individual exposure to pollution. 9 Measurement error in air pollution exposure can thus further attenuate OLS estimates.
To address these issues, we use annual nighttime thermal inversions as an instrument for air pollution exposure. A growing literature uses thermal inversions as an instrument for air pollution exposure (Jans, Johansson, & Nilsson, 2018;Molina, 2021). We use the NCEP/NCAR reanalysis temperature data at the two pressure levels closest to the ground, 1000 hPa and 925 hPa, to define thermal inversions. 10 The temperature difference between these two layers identifies an inversion episode, with the difference being negative under normal conditions and positive when there is an inversion episode. We focus only on nighttime inversions to ensure that households are less likely to change their exposure levels to pollution by observing inversion episodes. This is less likely to happen during the night. In other words, the concern is that households might limit the time they spend outside based on their private information about thermal inversions. To minimize the possibility of this, we focus on nighttime inversions similar to Molina (2021) and Jans et al. (2018). Our main inversion variable is the annual number of nighttime inversion episodes for each PSU in our sample. The average number of inversions in our sample is 89, with a standard deviation of 94.57 (Table 2). Using the annual number of nighttime inversions as our main IV, we estimate the following two-stage least squares (2SLS) specification: where "Inversions ptÀ1 " is equal to the annual number of nighttime inversions in PSU p in the year prior to the survey year t.
The main identifying assumption of the IV is that conditional on controls for precipitation, temperature, wind speed, and relative humidity and the squares of these variables, the annual number of nighttime thermal inversions impacts mental health only through their impact on air pollution exposure. Controlling for these weather variables is crucial to satisfy the exclusion restriction of the instrumental variable strategy. This is because inversions are correlated with temperature (see Appendix Figure 2 in Supplementary Materials), with inversions more likely in colder months, but also in some warmer months such as April and October. Since the temperature has been shown to independently impact mental health outcomes in India (Carleton, 2017;Pailler & Tsaneva, 2018), it is important to control for these weather variables to ensure that the IV estimates are valid.
The identifying assumption could be violated if individuals could predict their air pollution exposure by observing inversion episodes. However, since we include only nighttime inversions as described above, it is unlikely that households observe inversion episodes at midnight. Following Moretti and Neidell (2011) we check if the information on inversion episodes induces people to reduce their exposure to air pollution by limiting their time outside. In Appendix Table 1 in Supplementary Materials, we examine the impact of inversion episodes on the likelihood of spending time outdoors at the extensive and intensive margins. Specifically, we examine the likelihood of walking or biking outside in a typical week and, for those respondents who walk or bike, the number of days in a typical week the respondent walks or bikes outside for at least 10 minutes. There is no impact of inversion episodes on the likelihood of walking or biking outside at either the extensive or intensive margins. We also conduct a placebo test by regressing our main outcome variables on air pollution exposure in the year after the survey was conducted. To the extent that thermal inversions are isolating truly exogenous variation in PM2.5, there should be no correlation between air pollution exposure in the future and current year mental health. Appendix Table 2 in Supplementary Materials estimates the instrumental variable specification in Equation 3 with air pollution in the year following the survey year as the main independent variable of interest. We find small and statistically insignificant impacts of next year PM2.5 on our measures of mental health.
To test the assumption of ignorable treatment assignment (Angrist, Imbens, & Rubin, 1996), in Appendix Table 3 in Supplementary Materials we test for covariate balance at baseline in 2002 between PSUs with above and below the median number of inversions. We do not find any statistically significant differences in the baseline covariates between PSUs with above and below the median number of nighttime inversions.
Finally, migration in response to pollution could bias our estimates as well. Yet, selective migration is not a concern in our sample. Specifically, several studies have documented that India witnesses very low levels of inter-district rural migration (Munshi & Rosenzweig, 2009;Topalova, 2010). Further, Balakrishnan and Tsaneva (2021) use the 64 th round of the National Sample Survey to directly examine the impact of current year pollution and past pollution on two main outcomes: (a) a dummy variable indicating whether the household migrated in the past year to another district; and (b) a dummy variable for whether the reason for migration was in search of better employment or for a business. The authors do not find any statistically significant impacts of current year pollution and past pollution on migration.

OLS estimates
We first present the OLS estimates in Panel A in Table 3. Every coefficient represents a different regression. Results show the effect of a one mg/m 3 increase in PM2.5 and the coefficients are multiplied by 100 to reflect the percentage change in the outcomes of interest. Overall, the Impact of air pollution on mental health in India 139 likelihood of feeling low, sad, or depressed in the last 30 days decreases by 0.13 percentage points for a one mg/m 3 increase in last year's PM2.5 but this effect is not statistically significant. We also find a small and not statistically significant effect on experiencing cognitive difficulties. The probability that respondents feel unable to control and cope with the important things in life increases by 0.92 and 1.13 percentage points respectively (significant at the 1 percent level). As discussed above, OLS estimates are potentially downward biased because of omitted variable bias and reverse causality. Consequently, we present results from IV estimates next.

IV estimates
We first test the first stage of the IV specification from Equation 2. There is a positive and statistically significant (at the 1 percent level) impact of inversions on PM2.5. One additional nighttime inversion, on average, results in an increase in PM2.5 by 0.10 mg/m 3 . The first-stage effective F-statistic is 70.53. 11 Panel B in Table 3 presents the IV estimates of air pollution on mental health. Overall, there are significant adverse impacts on all measures of mental health. Specifically, for a one mg/m 3 increase in PM2.5 in the last year, the probability of feeling low, sad, or depressed in the past 30 days increases by 2.47 percentage points. There is a similar impact on cognitive difficulties: respondents are 2.31 percentage points more likely to experience cognitive difficulties in the last 30 days because of a one mg/m 3 increase in PM2.5. We also find that the inability of respondents to control or cope with the important things in life increases by 5.68 and 5.48 percentage points, respectively. 12,13 Our findings are consistent with Chen et al. (2018) the one other paper that estimates the causal impacts of air pollution exposure on mental health. 14 In Appendix Table 5 in Supplementary Materials, we present heterogeneous effects by worker type. It is possible that impacts are larger for agricultural workers due to potentially higher levels of air pollution exposure as they engage in more outdoor work (Saglan et al., 2020). In Panel A, we report estimates for the proportion of the sample who reported their main occupation in the last 12 months as agricultural or fishery workers, and in Panel B estimates are reported for the remaining occupation types. While we find that the magnitude of most impacts on agricultural workers is larger than impacts on non-agricultural workers, the Notes: [1] Individual and household controls include dummies for age between 18 -30, 30 -40, 40 -50, and older than 50; years of education; dummies for marital status being married, single, or widowed; dummy for female; household size; and dummies for owning a mobile phone, bike/moped; and a computer. Weather controls include controls for annual temperature, precipitation, windspeed, relative humidity, and their squared values. We also include survey year, PSU level, and month of interview fixed effects.
[2] Standard errors in parentheses, clustered at the PSU level. Number of PSUs ¼ 369.
differences are not statistically significant. 15 This is also consistent with our analysis of heterogeneity by urban/rural area of residence, where we find that both rural and urban populations experience consistently negative effects of air pollution exposure on mental health with similar effect sizes (Appendix Table 6 in Supplementary Materials). Ravishankara, David, Pierce, and Venkataraman (2020) also find that air pollution due to particulate matter is almost equally harmful across rural and urban India in terms of health impacts including heart disease, stroke, lung cancer, and other respiratory infections.

Robustness checks
The main identifying assumption of the estimation strategy is that conditional on weather controls thermal inversions impact mental health only through their impact on air pollution exposure. There is a possibility that the instrument may affect the outcome via pathways other than air pollution, for example through changes in indoor fuel use or other pathways. Below we show that the results are robust to using different instruments and different estimation strategies. All robustness checks are presented in Appendix Table 7 in Supplementary Materials. First, we restrict the sample to respondents present in both survey rounds (Panel A) to ensure similar observable characteristics across both survey rounds. Results are qualitatively similar to our main results. In Panel B, we also use the first difference estimator to estimate the results for respondents with information in both survey rounds and find that results are robust. Next, we apply district-level fixed effects instead of PSU fixed effects in the main model. This helps account for broader district-level events and policies. We find that effects are qualitatively similar to the main results (Panel C). We also test the robustness of our instrumental variable estimation using two alternative instruments. First, in Panel D we use an alternate definition of inversions. Specifically, we define inversion strength as the difference in temperatures between the 1000 hPa and 925 hPa pressure levels. 16 In Panel E, we also use wind speed as an alternate instrument. 17,18 We find that air pollution has a significant negative effect on mental health, irrespective of the instrument used. Finally, in Panel F, we examine if our results are impacted by including total cloud cover and its square as additional control variables (Penckofer, Kouba, Byrn, & Estwing Ferrans, 2010;Papadopoulos et al., 2005). We find that results are robust to the inclusion of these control variables.

Mechanisms
There are several mechanisms through which air pollution exposure could affect mental health. Direct impacts could occur through oxidative stress and inflammation in the brain which are common physiological responses to air pollution exposure (Power et al., 2015). This can significantly increase the likelihood of depression, anxiety, and cognitive dysfunction (Salim et al., 2012). The impact of air pollution exposure on mental health can also be caused by several indirect channels including impacts on physical health, physical activity, and labor productivity. While we are unable to examine the direct impacts on brain function, we conduct a mediator analysis to examine the extent to which the following four factors mediate the impact of air pollution exposure on mental health: (a) overall physical health, (b) poor respiratory conditions 19 , (c) changes in physical activity 20 , and (d) current work status 21 .
We first start by examining the impact of air pollution exposure on overall physical health, poor respiratory conditions, changes in physical activity, and current work status in Appendix Table 8 in Supplementary Materials. In columns [1][2], we examine the impact of air pollution exposure on physical health. Poor overall physical health in column [1] is coded as an indicator variable taking the value of 1 if individual rates their health as either being bad/very bad, reports overall difficulty in performing work or household activities in the last 30 days, or has sleeping difficulties in the past 30 days. 22 We also examine poor respiratory conditions in Impact of air pollution on mental health in India 141 column [2] measured by an indicator variable for either experiencing chest pains, attacks of wheezing or whistling breathing, or shortness of breath. We find that for a one mg/m 3 increase in PM2.5 exposure in the past year, respondents were 0.81 percentage points and 0.22 percentage points more likely to have poor overall physical health and poor respiratory conditions, respectively. In column [3], we examine if air pollution exposure impacts physical activityproxied by an indicator variable for the likelihood of walking or biking in a typical week. It is possible that air pollution induces people to remain indoors and reduces outside exposure, thereby limiting physical activity. 23 We find that air pollution exposure does not impact physical activity. Finally, in column [4] we examine the impacts of air pollution exposure on labor productivity. While we are unable to directly measure labor productivity, we examine the impact of air pollution exposure on an indicator for currently working. We find a negative impact on the probability of currently working.
Next, we conduct a formal mediator analysis. As a first step in the mediator analysis in Appendix Table 9 in Supplementary Materials, we examine if there is any link between mental health and these four factors in our data using OLS regressions. In Panels A and B, we examine the correlation between poor physical health and poor mental health. As expected, we find a large positive correlation between both overall physical health and respiratory conditions and mental health. Next, in panel C, we study the relationship between physical activity and poor mental health and find a negative correlation. Finally, we examine the correlation between poor mental health and an indicator for currently working in Panel D and again find a negative correlation.
As a second step in the mediator analysis, we examine the extent to which these four factors mediate the impact of air pollution exposure on mental health in Table 4. Following VanderWeele (2016), we report the direct effect of air pollution exposure on mental health (i.e. the effect that remains after accounting for the mediator). We compare coefficients in Table 4 with the IV estimates in Table 3. Results in Table 3 are presented without the mediator variables. Comparing the results in Table 4 to those in Table 3, we see that controlling for respiratory conditions, physical exercise and currently working has little effect on the coefficient estimate for the impact of air pollution exposure on the probability of feeling sad, low or depressed. Controlling for overall physical health, however, reduces the coefficient from 2.47 in Table 3 to 1.14 in Table 4 a 54% reduction. We find similar patterns for other outcomes, where overall physical health has the largest mediating impact compared to the other mediator variableswith the exception of the effect of respiratory conditions on feelings of being unable to control the important things in life. Nevertheless, all results for the relationship between air pollution and mental health remain statistically significant even after controlling for these key mediator variables. This suggests that while air pollution impacts mental health by worsening physical health, it also likely has a direct effect.

Conclusion
In this paper, we causally estimate the impact of air pollution exposure on mental health in India. We show that higher levels of pollution have significant impacts on feeling low, sad, or depressed; cognitive difficulties; and being able to control or cope with everything in life. There are important economic implications for the mental health effects of air pollution exposure. Poor mental health results in lower labor force participation and higher healthcare utilization and could perpetuate poverty. Developing countries like India are increasingly vulnerable to the adverse impacts of air pollution exposure due to higher pollution levels, higher disease burdens, and lack of social safety nets. Given the large proportion of people living in these countries, it is imperative to understand the impacts of pollution exposure on mental health to design optimal environmental and social policy which accounts for both physical and mental health costs of pollution exposure. Notes: [1] Individual and household controls include dummies for age between 18-30, 30-40, 40-50, and older than 50; years of education; dummies for marital status being married, single, or widowed; dummy for female; household size; and dummies for owning a mobile phone, bike/moped; and a computer. Weather controls include controls for annual temperature, precipitation, windspeed, relative humidity and their squared values. We also include survey year, PSU level, and month of interview fixed effects. In Panels A, B, C, and D, we also include one additional control variable each: overall physical health, poor respiratory conditions, physical activity, and an indicator for currently working as controls, respectively.
[2] Standard errors in parentheses, clustered at the PSU level. Number of PSUs ¼ 369.
Impact of air pollution on mental health in India 143