آیا فقر بهداشت روانی را کاهش می دهد؟تجزیه و تحلیل متغیرهای ابزاری

That poverty and mental health are negatively associated in developing countries is well known among epidemiologists. Whether the relationship is causal or associational, however, remains an open question. This paper aims to estimate the causal effect of poverty on mental health by exploiting a natural experiment induced by weather variability across 440 districts in Indonesia (N = 577,548). Precipitation anomaly in two climatological seasons is used as an instrument for poverty status, which is measured using per capita household consumption expenditure. Results of an instrumental variable estimation suggest that poverty causes poor mental health: halving one's consumption expenditure raises the probability of suffering mental illness by 0.06 point; in terms of elasticity, a 1% decrease in consumption brings about 0.62% more symptoms of common mental disorders. This poverty effect is approximately five times stronger than that obtained prior to instrumenting and is robust to alternative distributional assumption, model specification, sample stratification and estimation technique. An individual's mental health is also negatively correlated with district income inequality, suggesting that income distribution may have a significant influence upon mental health over and above the effect of poverty. The findings imply that mental health can be improved not only by influencing individuals' health knowledge and behaviour but also by implementing a more equitable economic policy.

The negative association between poverty and mental health in developing countries has been increasingly documented. Research from various parts of the world generally shows that low levels of income, education, and assets as well as low social class are correlated with a higher probability of having common mental disorders (Lund et al., 2010). However, empirical evidence regarding the causal effect of the association remains scarce. Few studies have investigated the strength or the direction of causality between poverty and mental health in developing countries, although such study clearly benefits the formulation of public policy aimed at improving the health of the population. In encouraging study of this topic in the United States, Stowasser et al. (2011, p.2) note that ‘ … if causal links between wealth and health were confirmed, society would likely benefit from more universal access to health care and redistributive economic policy. Yet, if such causal links were rebutted, resources would be better spent on influencing health knowledge, preferences, and ultimately the behavior of individuals.’ Considering both the growing burden of disease attributed to mental illness (IHME, 2013) and tightly constrained health budgets (Patel, 2007), it is important to understand whether poverty reduces mental health in developing countries. The fact that poverty is negatively associated with mental health in low- and middle-income countries is hardly surprising, but to reach a convincing estimate of its causal effect is certainly not an easy task. Two-way or simultaneous causation may come into play (Smith, 1999), inflating the estimated effect and making it impossible for researchers looking at observational data to separate the effect of wealth on mental health (social causation hypothesis) from that of the reverse (social selection hypothesis). Secondly, the observed wealth–health relationship may be confounded by unobserved common causes that accidentally induce a spurious correlation. Genetic frailty, early childhood environment, family background and preference or taste for lifestyle may impact both an individual's ability to work (and hence accumulate wealth) and his or her susceptibility to mental illness (Stowasser et al., 2011). The study on the mental health effect of poverty may also suffer from what is generally known as the attenuation bias. More often than not, wealth is measured with error, as a noisy, low signal-to-noise ratio variable which could trivially result in a downward-biased parameter estimate (Cameron and Trivedi, 2005). Because these endogeneity problems might be working at the same time, it is difficult to predict the magnitude and direction of the potential bias resulting from their presence a priori. In addition, the small number of population data available in developing countries remains a major obstacle for public health research. The aim of this paper is therefore to address these issues. We apply instrumental variable and control function estimators to a large (N = 987,205), nationally representative dataset from Indonesia, namely the Riset Kesehatan Dasar (Riskesdas) 2007. We use seasonal precipitation anomaly, defined as the average deviation of monthly precipitation from its half-century (1951–2000) normals in all 440 kabupaten (districts) in Indonesia, as an instrument for poverty status. The identifying assumptions are that precipitation anomaly strongly predicts per capita household expenditure in a largely agricultural economy (relevance condition), is randomly assigned hence unrelated to any potential unobserved confounders (validity condition), and is exerting its influence upon mental health only through its effect on consumption expenditure (exclusion restriction). Conditional on these partially testable assumptions, the instrumental variable approach allows the analyst to isolate the exogenous variation of poverty, thus allowing for the derivation of a consistent estimate of the mental health effect of poverty in the presence of endogeneity. This study is one of the few population-based studies that attempts to look beyond the simple correlation between poverty and mental health in the context of low- and middle-income countries.

Despite the claim that poverty causes mental illness in low- and middle-income countries, empirical evidence remains scarce. Little has been done to address the question of whether the observed wealth–health relationship is causal or just associational. The present study attempts to fill this gap by exploiting seasonal precipitation anomaly as a form of natural experiment that randomly determines poverty status in Indonesia. Results suggest that poverty causes poor mental health. Holding all other covariates constant, halving one's consumption expenditure raises the probability of having mental illness by 0.06 point, or, in terms of elasticity, a 1% decrease in consumption brings about 0.62% more symptoms of common mental disorders. This study finds that the effect of poverty on mental health is approximately five times stronger than is conventionally estimated, which may be indicative of the fact that measurement error rather than reverse causality was the main source of bias (Ettner, 1996). The effect is robust to varying distributional assumption, model specification, estimation technique and sample stratification. This supports the general finding in social epidemiology (Lund et al., 2010). The present study also investigates the association between district-level income inequality and mental health. It is consistently estimated that income inequality correlates negatively with mental health over and above the effect of poverty. Individuals living in unequal districts are found to have a higher probability of suffering from mental illness than those who live in more egalitarian districts. This is consistent with the recent finding of Filho et al. (2013), who conducted a multilevel study in the Brazilian context. This also weakly supports the broader idea of the income inequality hypothesis put forward by Wilkinson and Pickett (2010). Additionally, the present study found that women, older people and those who are divorced or widowed tend to have a higher probability of suffering common mental disorders. This is, again, consistent with the existing literature on mental health in developing countries. Finally, negative health behaviours such as less physical activity, frequent smoking and heavy drinking are all related to lower levels of mental health. This study has a number of limitations. The first pertains to the core assumption of instrumental variable estimation. For this method to work properly, one must maintain three strong assumptions, namely the relevance condition, the validity condition and the exclusion restriction. Not all of these are testable. While it has been shown through the weak identification test that seasonal precipitation anomaly strongly predicts per capita household expenditure (hence satisfying the relevance condition), there is no empirical test capable of examining the exclusion restriction (Freedman, 2005, Freedman, 2010 and Hernán and Robins, 2006). This must be established a priori. The quality of an instrumental variable estimation is only as good as its story; here it rests ultimately on the untestable assumption that precipitation anomaly is indeed a random variate perfectly uncorrelated with any determinants of mental health, and that it does not affect an individual's mental health except through its influence upon consumption expenditure. The second limitation relates to the possible interpretation of the causal parameter recovered by instrumental variable estimation, namely as a local average treatment effect (LATE) ( Angrist and Pischke, 2008). Under the LATE interpretation, the causal parameter obtained in this study is simply the average effect of poverty on mental health for individuals whose income fluctuates in accordance with the randomisation provided by the natural experiment (the average treatment effect of the compliers). Of course, generalising this causal effect to the entire population of Indonesia requires additional layers of assumption, but given that a large proportion of the Indonesian workforce is employed in the largely rain-dependent agriculture sector, we believe that even the LATE parameter is worthy of consideration. This study is also limited by the cross-sectional nature of the data. Future studies may take advantage of a longitudinal design so that temporal order can be incorporated into the model. Despite its limitations, the present study contributes to the literature on mental health in developing countries in several ways. First, this study is among the few studies that attempt to address the endogeneity problem in the estimation of the mental health effect of poverty. Second, using a large and representative data from Indonesia, this study demonstrates that the adverse effect of poverty on mental health is not merely attributed to the self-selection bias that threatens small-sample community or facility studies. Third, considering both the use of a standard mental health and poverty measure and the fact that Indonesia is the most populous developing country after China and India, this study provides a finding that is suitable for cross-national comparison. Finally, the present study shows that poverty remains an important determinant of mental health irrespective of whether it is treated as an exogenous or as an endogenous variable. Indonesian policy makers now have reason to believe that poverty alleviation efforts can have considerable impact on the mental health of the population. Mental health can be improved not only by influencing individuals' health knowledge and behaviour but also by implementing a more equitable economic policy. Policy makers may also want to consider a greater public investment in the long-neglected mental health service sector, which would certainly benefit the nation as a whole given that the burden of mental illness is borne not only by the patients but also by their family members. Additionally, research has shown that mental illness is costly for a nation's economy (Lund et al., 2013). Furthermore, according to the referral scheme of Indonesia's recently launched version of the universal health care system (The Lancet, 2014), every prospective patient is required to report to the nearest primary care centre prior to visiting a hospital; mental health care service, then, must be surely made available at the lowest level of the referral hierarchy. Unless such a provision is available, the mental health of Indonesians will continue to be overlooked.