Air Pollution and Long Term Mental Health

: We study the long-term consequences of air pollution on mental health, using a natural experiment in Indonesia. We ﬁnd that exposure to severe air pollution has signiﬁcant and persistent consequences on mental health. An extra standard deviation in the pollution index raises the probability of clinical depression measured 10 years past exposure by almost 1%. Women in particular seem to be more affected, but some effects persist for men as well. Pollution exposure increases the likelihood of clinical depression for women and also the severity of depressive symptoms for both sexes. It is not clear if men are more resistant to pollution or they simply recover faster from it. Education, perceived economic status, and marriage seem to be the best mitigators for these negative effects.


Introduction
Air pollution, especially fine particulate matter, has immediate negative consequences on health and causes a range of additional economic costs to society. These effects are well documented by the medical and economic literature. Air pollution is linked to respiratory problems in both infants and adults (Emmanuel 2009 It is also worth mentioning studies like Wang et al. (2014) or Zijlema et al. (2016) who found no consistent evidence for an association between air pollution and depressive symptoms. It is important to note first that some associations were found on certain sub-samples, and second that these studies are also purely correlational and lack any causal identification. This is further complicated by the fact that they used proxies for pollution exposure that are less than ideal, such as distance to the nearest major roadway, which is arguably correlated with many other confounders.
Our study contributes to the existent literature in a few distinct ways. We study the persistent (long term) effects of air pollution on mental health, focus on a short-lived but intense pollution shock, use longitudinal data and, most importantly, provide a credible identification source for the causal effects of air pollution on mental health: a natural experiment that occurred in Indonesia in 1997.
From September to November of 1997, large parts of Indonesia were blanketed in thick smoke caused by massive forest fires. The fires originated with slash-and-burn practices used by local farmers as a cheap way of clearing land, but got out of control due to the especially dry and windy season caused by El Niño. Since the exposure of people living in different parts of Indonesia to this pollution was due to natural phenomena such as lack of rain or wind direction and intensity, it can be considered as good as a random assignment. In fact, this episode has been widely used in the social science literature as an identification source for a range of studies involving air pollution (see for instance Heil  We find significant effects of pollution on depression and depressive-like symptoms that persist even 10 years after exposure. Women seem to be more affected than men, or at least they recover slower than men do. The pollution shock of 1997 was found to increase the severity of mild depression-like symptoms in both sexes and also to increase the likelihood of experiencing severe clinical depression in women. Both these effects are observed in 2007, ten years after the pollution shock, which suggests not only that pollution has long-lasting effects, but also that the actual near-term effects and associated costs are much larger in magnitude. Unfortunately, due to data limitations, we cannot test the recovery hypothesis nor establish a clear significant link between pollution and clinical depression for men. The pattern observed in the results however, coupled with results from the existent literature, is highly suggestive of the fact that men are not immune to pollution, but they might recover faster from it.

Data and Methodology
We exploit data from the Indonesia Family Life Survey (IFLS). IFLS is a longitudinal survey spanning over 20 years and containing a representative 1 sample of Indonesian households. IFLS collects a vast number of demographic and socio-economic indicators at the individual, household, and community level. The attrition rates are also very small in IFLS, which is ideal for our study, as it avoids potential attrition-related biases.  Radloff (1977). Based on respondents' answers to a series of 10 questions, and following the CES-D guidelines, we first computed a depression score for all individuals in our sample. This score ranges from 0 to 30, with larger numbers representing more severe depressive symptoms. A well-established cutoff in the medical literature (see for instance Andresen et al. 1994) is that a score higher or equal to 10 represents severe depressive symptoms or clinical depression. We therefore coded an indicator variable equal to 1 if a respondent's score was higher or equal to 10 and zero otherwise. We used this indicator and the raw depression score as our dependent variables in our study. The CES-D questionnaire was introduced in IFLS during the 2007 wave and so all our analyses refer to the persistent (long-term) effects of air pollution on depressive symptoms.
Since our focus is studying the effect of air pollution on depression, the main explanatory variable we use is the pollution level that respondents were exposed to during the 1997 fires. The  using these GPS coordinates. We then compute the pollution variable we use in all our regressions as the average monthly pollution over the September, October, and November months of 1997. For the monthly pollution we use the median of the daily values. Table 1 contains simple summary statistics describing the pollution levels and also the incidence of clinical depression among respondents. The In terms of depressive symptoms, about 6.15% of the surveyed respondents are clinically depressed, with values of the CES-D score above 10. The incidence of depression seems to be higher for women than for men: About 7.2% of the women are significantly depressed, compared to only 4.7% of the men.
In addition, we also collect a number of socio-economic factors at the individual and household level that we use as controls in all our regressions. Although we argue that the exposure to pollution was due to a natural phenomenon and is therefore not likely correlated to any socio-economic indicators that could affect depression, these controls allow for a more robust estimation. We include controls such as respondents' age and age squared (to allow for non-linearities with respect to age), years of formal education, per capita expenditures (PCE), which is used as a proxy for household income 3 , whether the household kitchen and water source are inside or outside the household (as this might affect pollution exposure), indicators for sex and marital status, indicators for subjective (perceived) economic status 4 , and initial general health status (before the pollution event). Ideally, we would like to also control for the pre-pollution depressive symptoms, but this data was not collected by IFLS prior to 2007. However, since the pollution was an unprecedented and exogenous shock, it is unlikely to be correlated with the initial depressive symptoms and, furthermore, the initial general health status proxies to some degree for mental health too. The general health status (GHS) is a self-reported binary health measure which takes value 1 (Poor GHS) if respondents classify their own health as either "unhealthy" or "somewhat unhealthy", and value 0 (Good GHS) if respondents classify their health as "healthy" or "somewhat healthy". In spite of being somewhat vague and possibly suffering from subjective biases, GHS is an aggregate measure that proxies for a large variety of health issues and has been found to be a good 3 Results Table 2 presents the estimated effects of pollution on the raw CES-D depression score. As mentioned before, this score takes values from 0 to 30, with larger figures representing more severe depressive symptoms. We control for a number of socio-economic indicators and for base-level health. Since the incidence of depression in women was found to be larger than in men, we estimated these effects for 4 A series of perceived income ladder indicators are constructed, based on respondents' answer to the following question: "Please imagine a six-step ladder where on the bottom (the first step) stand the poorest people, and on the highest step (the sixth step) stand the richest people. On which step are you today?" Since very few respondents chose the sixth step, we combined the fifth and sixth step.
the entire sample and disaggregating by gender. The results in the table are estimated using ordinary least squares (OLS) with robust standard errors. For robustness purposes, ordered probit and ordered logit regressions were also performed, which yielded similar results.  We find that air pollution has negative consequences on the CES-D score that persist 10 years after exposure. These effects are statistically significant for both men and women, with slightly larger impacts for women. Besides the direct effect of pollution, our estimations also point to men being overall less depressed than women, to marriage and education being mitigating factors for depression, and somewhat interestingly, to the effects of subjective and objective economic well-being. While objective economic well-being (proxied by PCE) seems to have no significant effect, perceived economic well-being (proxied by the income ladder) is clearly and consistently mitigating depressive symptoms for both sexes. Perceiving yourself to be on a higher income ladder is generally associated with a lower CES-D score.
We further investigate the relationship between air pollution and depression by estimating the effects of pollution exposure in 1997 on the presence of severe (or clinical) depression in 2007, controlling for the same set of socio-economic indicators and base-level health status. The presence of clinical depression is signaled by a CES-D score higher or equal to 10 so the new dependent variable is an indicator that takes value 1 if the respondent's CES-D score is higher or equal to 10 and 0 otherwise. We again performed this estimation first for the entire sample, then separately for men and women. The results in Table 3 are that of a linear probability model estimated using ordinary least squares (OLS) with robust standard errors. For robustness purposes, we also performed logit and probit estimations and found similar results.  year effect is much larger. While theirs is not a study on mental health, the results signaling recovery are suggestive for a larger context. In addition, our previous results show that men have higher CES-D scores as a result of pollution, which proves that men are not immune to pollution. An important conclusion under this recovery hypothesis is that, while our estimation shows that women still suffer from depression caused by air pollution even 10 years after the event, the impact of this pollution could have arguably been much larger in magnitude in the immediate months and years after the fires.
As a potential mitigator of depression, it is worth nothing that some previous research (see for instance Helgadotir et al. 2017) found that exercise and physical activity has short-and long-term benefits for depression. Since men are generally more likely to engage in physical activities, this might be the channel responsible for their faster recovery. To that end, we coded an indicator variable equal to 1 for those respondents claiming their job requires them to engage in physical labor and included it in the estimations. Physical labor does not seem to be the channel responsible for men's recovery, although engaging in physical labor lowers the probability of depression. Specifically, when including the physical labor indicator in the men regression as an additional control, we find a statistically significant negative effect on depression (lower incidence of depression for men engaging in physical labor). However, when further disaggregating the sub-sample of men into those who performed physical labor and those who did not, we find the same insignificant effects of pollution on clinical depression for both sub-samples of men. This implies that even those men who did not engage in physical labor seem to have been recovered from clinical depression. So while exercise and physical labor are important mitigators for depression, they do not seem to be the responsible channels for men's faster recovery from clinical depression.
For robustness purposes, since the vast majority (over 93%) of the respondents in the sample cannot be classified as experiencing clinical depression according to the CES-D guidelines, we extend our analysis to study the effects of air pollution exposure on the intensity of mild depressive symptoms.
To that end, we first restrict the sample to only those respondents with a CES-D score below 10, then estimate the effects of pollution exposure on the CES-D score itself, while controlling for the same socio-economic variables from our main regression. These are nothing more than a reiteration of the estimations from Table 2, performed for those respondents who experience only mild depressive symptoms (CES-D scores below 10). Table 4 presents the results of these OLS estimations, but ordered probit and ordered logit estimations were also employed and the results were found to be similar. We again estimated these for the whole sample first, then for men and women separately.  The results are robust to our previous estimations and show significant effects of pollution for both men and women, with slightly higher magnitudes for women. This strengthens the idea that men are not immune to pollution, but the effects are somewhat smaller. Whether that is because men are more resistant to depressive symptoms or because they recover faster from such symptoms remains an important question. Since the analysis is based on voluntary survey responses, it is also possible that men are simply less likely to report such symptoms even in the absence of any actual psychological or physiological differences between sexes.
While the results presented here are hard to interpret quantitatively due to the lack of measurement units for the aerosol index measuring pollution, they are highly suggestive from a qualitative perspective.
The estimates are robust to alternative estimation strategies and show significant and persistent negative effects of exposure to air pollution on depression and depression-like symptoms. The fact that we are able to estimate effects that persist over a ten years period, suggests that the actual negative effects of pollution in the short-term are even larger and can have serious social and economic consequences.

Conclusions
In this paper, we have provided evidence of significant and persistent negative effects of air pollution on mental health. Using a natural experiment in Indonesia, we find that exposure to severe air pollution significantly increases the incidence of depressive symptoms for both men and women and the incidence of clinical depression among women, even ten years after the pollution exposure. We also find robust significant effects of pollution on the severity of mild depressive symptoms that persist over time for both sexes. This pattern is consistent with a hypothesis of faster recovery times for men, but more research needs to be done to address this issue.
Another important avenue for future research is investigating the linkages and causal chains between mental health, general health, and economic outcomes in a context of air pollution exposure. In light of the results presented in this paper and those from previous research that link pollution to general health, labor supply, and earnings, it is important to understand the degree to which general health affects mental health (or vice-versa) and the degree to which both affect economic outcomes. If there are strong causal links, policies to mitigate the negative effects of pollution can be targeted towards the base of the causal chain. If however, these links are weak and pollution has separate and distinct negative effects on general health, mental health, and labor activities, then multiple policies need to be implemented to target these issues individually.
Also, since using aerosol index data to proxy for pollution exposure cannot provide meaningful quantitative estimates, better efforts have to be made to establish robust correlations of the aerosol index data with the ground-based pollution monitors. The qualitative implications of ours and other emerging results involving the persistence of the negative consequences of pollution over time are troubling and the economic costs that are often ignored by the medical literature are very high. But in the absence of robust quantitative measures of ground-level pollution, proper cost-benefit analyses cannot be performed and mitigating policies cannot be properly evaluated. Alternatively, studies that do use data generated by ground-level monitors need to be better formulated to control for many potential confounders to ensure the magnitudes of the estimates are not biased and represent true causal inference.
Air pollution, particularly particulate matter pollution, is clearly affecting humans in a negative way and these negative effects seem to linger for years. It is therefore extremely important the scientific debate continues and more studies address the issue to better pinpoint the effects and best possible policies. In the meantime, education and economic stability seem to be the best and most consistent mitigators for these negative consequences and they need to be continuously improved, especially in developing countries that are resource constrained.