ARTICLE IN PRESS

Social Science & Medicine 65 (2007) 467480 www.elsevier.com/locate/socscimed

Mental health and poverty in developing countries: Revisiting the relationship

Jishnu Das, Quy-Toan Do, Jed Friedman, David McKenzie, Kinnon Scott
World Bank, Development Economics Research Group, MSN MC 3-311, 1818 H Street, Washington, DC, USA Available online 25 April 2007

Abstract The relationship between poverty and mental health has received considerable attention in the recent literature. However, the associations presented in existing studies typically rely on limited samples of individuals and on proxy indicators for poverty such as education, the lack of tap water, or being unemployed. We revisit the relationship between poverty and mental health using data from nationally representative household surveys in Bosnia and Herzegovina, Indonesia and Mexico, along with special surveys from India and Tonga. As in previous studies, we nd that individuals who are older, female, widowed, and in poor health are more likely to report worse mental health outcomes. Individuals living with others with poor mental health are signicantly more likely to report worse mental health themselves. The size of the coefcients and their signicance are comparable across the ve countries. In contrast to previous studies, the relationship between higher education and better mental health is weak or non-existent. Furthermore, there is no consistent association between consumption poverty and mental health in two countries mental health measures are marginally worse for the poor; in two countries there is no association; and in one country mental health measures are better for the poor compared to the non-poor. Moreover, the sizes of the coefcients for both education and consumption poverty are small compared to other factors considered here. While the lack of an association between consumption poverty and mental health implies that poor mental health is not a disease of afuence, neither is it a disease of poverty. Changes in life circumstances brought on, for instance, by illness may have a greater impact on mental health than levels of poverty. Effective public health policy for mental health should focus on protecting individuals and households from adverse events and on targeted interventions following such adverse changes. r 2007 Elsevier Ltd. All rights reserved.
Keywords: Mental health; Poverty; Household concordance; Socioeconomic gradients; Developing countries; Comparative

Introduction The relationship between poverty and mental health holds great interest for both health and
Corresponding author. Tel.: +1 0202 473 2781.

E-mail addresses: jdas1@worldbank.org (J. Das), qdo@worldbank.org (Q.-T. Do), jfriedman@worldbank.org (J. Friedman), dmckenzie@worldbank.org (D. McKenzie), kscott@worldbank.org (K. Scott). 0277-9536/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.socscimed.2007.02.037

economic policy makers. Empirical ndings from developed countries suggest that, for most mental health disorders, the association between low socioeconomic status and psychiatric morbidity is strong and signicant (Kessler, Chiu, Demler, & Walters, 2005; WHO International Consortium in Psychiatric Epidemiology, 2000). This relationship has been found to hold, in some cases even more strongly, in low-income countries.

ARTICLE IN PRESS
468 J. Das et al. / Social Science & Medicine 65 (2007) 467480

In 11 developing-country community-based studies, signicant associations between poverty indicators and common mental disorders were found in all but one study (Patel & Kleinman, 2003). Univariate odds ratios (henceforth OR) predicting prevalence of a common mental disorder include low education (OR 3.3), earning less than one-quarter of the minimum wage (OR 3.9), and not working (OR 3.6) in Brazil; having no electricity (OR 1.47) or tap water (OR 2.2) in Indonesia; arguing with spouse for economic reasons (OR 10) in Pakistan; and being unemployed (OR 2.9) or living in an overcrowded situation (OR 2.1) in Zimbabwe. Another multi-country review of studies in Zimbabwe, two sites in Brazil, and in Chile showed strong associations between income terciles and the prevalence of common mental disorders, with OR for those in the highest terciles ranging from 0.46 to 0.50 relative to those in the lowest terciles (Patel, Araya, de Lima, Ludermir, & Todd, 1999). Higher monthly income and formal education were also associated with reduced odds of mood disorders in rural Ethiopia (Awas, Kebede, & Alem, 1999). This paper revisits the association between mental health and socio-economic outcomes in a number of low and middle-income countries through the analysis of recently available household survey data from Bosnia and Herzegovina, Indonesia, India, Mexico and Tonga. These data differ from those used in previous studies in important ways. First, the samples are drawn from a sampling frame of households rather than (for instance) a sampling frame of patients in health clinics (Patel et al., 1998). The latter may result in biased estimates of population-wide morbidity and the association with socio-economic characteristics if the use of health clinics is different among the general population compared to those with mental disorders. Second, mental health measures are collected for all adults in the household; these allow us to examine the concordance of mental health outcomes among different members of the household with important implications for treatment. Third, detailed expenditure modules in each of these multi-purpose surveys can be used to construct household consumption measures. Household per capita consumption is the preferred monetary-based welfare measure for poverty analysis among economists and hence is particularly germane for discussions of mental health and poverty in the developing world (Deaton & Zaidi, 2002; Ravallion, 1994).

Why mental health and poverty might be associated with one another? Conceptually, there are a number of potential channels that may lead to a higher prevalence of mental health disorders among the poor. Under the social causation hypothesis, poverty may lead to mental health disorders through pathways such as stress or deprivation (Johnson, Cohen, Dohrenwend, Link, & Brook, 1999; Miech, Caspi, Moftt, Wright, & Silva, 1999), or lower the likelihood of individuals receiving effective treatment (WHO, 2001). Under the drift or selection hypothesis, causation may run the other way, as poor mental health can impoverish people through lower employment and higher health costs (Bartel & Taubman, 1986; Miranda & Patel, 2005). The relationship may also simply reect third factors related to both poverty and mental health. For example, poor people are more vulnerable and may be more likely to experience stressful life experiences such as exposure to violence and poor physical health, which are recognized risk factors for mental health disorders (Patel & Kleinman, 2003). This points to the importance of multivariate analysis, which allows one to control for the inuence of many of these third factors when examining the povertymental health relationship. This general framework suggests several ways through which the mental healthpoverty relationship may differ in developing countries compared to developed countries. The rst concerns the relative availability of mental health services in developing countries compared with the developed world the proportion of individuals with mental health disorders receiving treatment is much higher in developed countries. WHO World Mental Health Survey Consortium (2004) reports that 49.764.5% of serious cases are treated in developed countries, compared to 14.623.7% in Mexico, Colombia, and Lebanon, while only 0.510% of mild cases received treatment in developing countries. The relative availability of competent yet costly mental health services can result, as in the US, in a situation where middle-class individuals with mild disorders in rich countries receive treatment, whereas the poor with severe disorders do not. This differential access to treatment will then lead to an association between mental health and poverty. In contrast, with equally poor access over at least the bottom three-quarters of the income distribution in developing countries, the

ARTICLE IN PRESS
J. Das et al. / Social Science & Medicine 65 (2007) 467480 Table 1 Context GDP per capita US$ Health expenditure per capita (US$) 168 27 30 372 102 3509 29 115 Hospital beds per 1000 Physicians per 1000 469

Bosnia India Indonesia Mexico Tonga OECD Low income average Middle income average

1325 512 872 5876 1615 28055 428 1937

3.1 0.9 n.a. 1.0 3.2 3.0 n.a. 2.0

1.3 0.6 0.1 1.5 0.3 6.0 n.a. n.a.

Source: Latest year available from 2001 to 2003 from World Bank Central Database.

relationship between mental health and poverty should be less strong. The relationship between mental health and poverty may also be weaker in developing countries due to the more exible nature of employment, especially in the informal sector in developing countries. Self-employment, agricultural work, and other jobs with less rigid attendance requirements are more common in developing countries. If depression, anxiety, and other mental health disorders make it more difcult for individuals to keep regular working hours, mental health problems would be expected to lead to more of an unemployment-related fall into poverty in developed countries than in developing countries. Additionally, larger family and village social support systems in developing countries may act to both lower the risk of developing mental health disorders and help insure individuals against poverty should they develop a disorder. Data and methods Data and context Multi-purpose surveys combining a mental health component with extensive socio-economic measures and information on all household members were elded in Bosnia and Herzegovina (hereafter Bosnia), Indonesia, India, Mexico and Tonga. Interviews were conducted face-to-face by trained interviewers in the local languages. These household surveys are multi-stage probability samples representative of the national population in Bosnia, Indonesia and Mexico with data on over 5400

households in Bosnia and over 10000 households in Indonesia and Mexico. The Indian and Tongan surveys were special purpose surveys. The data in India are from a longitudinal study of 300 households (1600 individuals) in the capital, Delhi; the sample of households is no different in observable attributes from a representative sample of house nchez-Pa ramo, 2003). holds in the city (Das & Sa The Tongan respondents in 230 households were chosen randomly from villages in which some individuals had applied for an emigration lottery (Stillman, McKenzie, & Gibson, 2006). These ve countries span a range of continents, levels of development, and cultural settings, allowing us to determine the extent to which associations with mental health are similar in very different contexts. Table 1 provides the GDP per capita and available health infrastructure indicators for each country, and compares these with the OECD, low income, and middle income averages.1 India and Indonesia have similar health care expenditure per capita to the low income country average, at less than 1% of the OECD average. Bosnia and Tonga are slightly poorer than the average middle income country and have health care expenditure equal to 35% of the OECD average, while Mexico is an upper-middle income country, with health expenditure still only 10% of the OECD average. The prevalence of mental health disorders is likely to be particularly high in Bosnia, due to lingering
The World Bank classies 54 countries as low income, based on GNI per capita of less than $875 in 2005, and a further 98 countries as middle income, with GNI per capita between $876 and $10,725 in 2005.
1

ARTICLE IN PRESS
470 Table 2 Overview of datasets employed Country Year of survey Number of Obs. Level of representation Mental health survey instrument Mental health measure Mean Bosnia India Indonesia Mexico Tonga 2001 2003 2000 2002 2005 12956 784 25470 19798 714 National 7 neighborhoods in New Delhia National National Special sample of migrant sending villages CESD SCL-90R GHQ derived GHQ derived MIH-5 1.495 1.535 1.413 1.341 1.745 SD 0.502 0.416 0.508 0.358 0.337 J. Das et al. / Social Science & Medicine 65 (2007) 467480

Indistinguishable from a representative sample of the day.

effects of the 199195 war, and possibly in Indonesia due to the 199798 nancial crisis (Friedman & Thomas, 2006), although comparison of mental health prevalence across countries is notoriously difcult. Both Mexico and Tonga have high rates of international migration. However, in both countries there is evidence that migrants have higher levels of education than those who do not migrate (Chiquiar & Hanson, 2005; McKenzie, Gibson, & Stillman, 2006). As a result, it does not seem that the magnitudes of migration among the poor are high enough for any possible non-random migration by mental health status amongst poor households to affect the inferences drawn in this paper. Given the variation in country contexts and cultural backgrounds, our analysis will look at deviations from country means, examining the relationship between different socioeconomic characteristics and mental health status within a country. Each of the surveys elded a widely used mental health screening instrument designed to measure mental health status of the general population. Table 2 lists the survey instrument used by country. The Tonga survey used the Mental Health Inventory (MHI-5) of Veit and Ware (1983), which has been used in over 50 countries as part of the International Quality of Life Assessment project. It has been shown to perform well in a number of settings in detecting major depression, general affective disorders, and anxiety disorders (Berwick et al., 1991). The Mexican and Indonesian surveys used variants of the General Health Questionnaire (GHQ) of Goldberg (1972), which displays similar psychometric properties to the MHI-5 (McCabe, Thomas, Brazier, & Coleman, 1996). The Bosnia survey used the Center for Epidemiological Studies

Depression Scale of Radloff (1977), a 20-question self-reported depression scale. The Indian survey uses the most comprehensive instrument, the 90 question Symptom Checklist 90 Revised.2 These screening surveys were translated and back-translated to ensure accuracy and extensively tested in the eld to ensure comprehension among study subjects. In four countries (Bosnia, Indonesia, Mexico and Tonga), the surveys were elded on the rst visit to the household. In India, the 90 question SLC-90R was elded 1 year after the longitudinal survey was initiated to ensure some degree of comfort between the respondents and eld workers. The mental health modules in each of the utilized surveys ask respondents the frequency in the last month of a similar range of internal states (e.g. feeling sad or blue, feeling anxious or nervous) or related behaviors (e.g. difculty falling asleep, distracted from everyday activities).3 The frequency of such states or behaviors is recorded on a four-point scale that ranges from never or almost never to very often. To score the individuals survey response, a low ordinal value (1 point) is assigned to categorical responses of infrequency and high ordinal values (up to 4 points) to the categories indicating greatest frequency. The average response across all questions constitutes the respondents mental health score, often known as
As opposed to relatively lengthy diagnostic interviews such as the Comprehensive International Diagnostic Interview (Kessler et al., 2005), the more common mental health instruments included in socio-economic surveys attempt to measure general psychological distress and are not intended to diagnose specic manifestations of mental illness per se. 3 The recall period in the Indian survey using the SCL-90R was 1 week instead of 1 month.
2

ARTICLE IN PRESS
J. Das et al. / Social Science & Medicine 65 (2007) 467480 471

the Global Severity Index or GSI, which is higher for those reporting worse mental health.4 The Global Severity Index weights all questions equally, in accordance with the approach widely employed in the existing literature across a number of settings and countries. One question is whether an alternative index could account for more of the variation in mental health status across individuals if a different weighting scheme were used. One natural approach is to reweight with the rst principal component of the different questions used in constructing the index, which weights each question in order to provide maximum discrimination across individuals. This approach will give questions which vary the most across individuals higher weight. We nd the correlation between the mental health score obtained by equal weighting and by principal components to be 0.9932 in India, 0.9996 in Tonga, 0.9959 in Indonesia, 0.9980 in Mexico, and 0.9984 in Bosnia. As a result, any change in the index from re-weighting will be minimal, and so we follow the existing literature in constructing the GSI with equal weights. Table 2 presents the mean of the raw scores across the surveys, as well as various characteristics of each survey. The main mental health scores across all surveys fall in the narrow range of approximately 1.351.50 indicating the average response to any particular mental health measure to be somewhere between almost never and rarely or infrequently. The standard deviation around this mean is also relatively similar in the general range of .35.50. Tonga is the exception where the average question response is higher at 1.75 and with slightly less dispersion at a standard deviation below .35.5 Given similar distributions of the mental health score across the ve countries, the score is
4 Typically a cut-off score indicating the likely presence of psychological disorders is determined by mental health professionals through supplementary validation exercises. These exercises were not available for all studies included here. 5 The exact questionnaire content varies across countries. However for every country but Tonga there are three similar questions: whether the respondent has recently felt sad, felt anxious, and had trouble sleeping. Limiting overall mental health score to these three questions yields similar yet slightly elevated scores and greater dispersion around the mean score. The mean response for the 3-question subset and the overall mean score are highly correlated with coefcients ranging from .84 to .90 across the datasets. Due to these close correlations, and in order to include Tonga in the analysis, the analysis focuses on the comprehensive measure.

standardized around the mean of each country and expressed in units of standard-deviations to enhance comparability and facilitate the interpretation of the results. The standardized GSI, formally dened as GSIindividualmean(GSI)country/standard deviation(GSI)country, is the outcome variable for the analysis in this paper. The relative magnitudes of different factors are then directly compared across countries. Statistical methods The analysis explores the co-variation of the standardized GSI with a range of potentially related factors at the level of the individual and household. In order to parsimoniously explore predictors of poor mental health, similar groups of characteristics measured consistently across each data set are identied. These characteristics vary either at the individual level, such as age, gender, marital status, or education, or the household level, such as household size or total household expenditures.6 For the larger surveys, community level characteristics are accounted for either through a community-level xed effect or by including the average individual mental health score for the entire community as an additional explanatory variable. The predictive power of each of these characteristics is estimated in a separate national level ordinary least square regressions7 using STATA/SE Version 9.0. In each of these regressions, the respondents are restricted to those aged 1580 and standard errors are clustered to correct for possible response dependency at the household level. Selected results are graphically summarized in accompanying gures. Associations between the mental health score and continuous control variables are depicted by gradients estimated with a partial linear model. In this approach, all covariates except the one depicted are modeled in a parametric
Most of the surveys use an extensive expenditure module to capture not just monetary expenditures, but also the value of goods produced for home production, gifts, and the value of owner-occupied housing. This provides a comprehensive indicator of consumption welfare. The Tongan survey uses per capita household income instead of consumption. 7 Qualitatively similar results are obtained using logistic regression to examine the odds that the individual will be in the worst 20%, 10% or 5% of the population in terms of mental health scores; hence, the associations found with mental health outcomes are not purely driven by those with sub-clinical symptoms that are of less important than clinical illness.
6

ARTICLE IN PRESS
472 J. Das et al. / Social Science & Medicine 65 (2007) 467480

Age Mental health score 1 0.5 0 -0.5 -1 20 40 Age Bosnia India Tonga Indonesia Mexico 60 80 Mental healthscoree 1 0.5 0 -0.5 -1 0

Education

5 10 Years of formal schooling Bosnia Mexico Indonesia Tonga

15

Mental health of other household members 2 Mental healthscore 1.5 1 0.5 0 -0.5 -1 -2 0 2 4 Household average mentalhealth score Bosnia India Tonga Indonesia Mexico 6 Mental health scoree 2 1.5 1 0.5 0

Mental health of other community members

-0.5 -1 -4 -2 0 2 4 Community average mental health score Bosnia Mexico Indonesia 6

Fig. 1. Mental health score by selected characteristics (continuous). (A) Age, (B) Education, (C) Mental health of other members and (D) Mental health of other community members.

fashion while the depicted variable is allowed to vary in a non-parametric fashion.8 Associations between mental health and discrete covariates are conveyed graphically by their 95% condence intervals.

Demographic inuences Several of the empirical regularities with regard to demographic inuences identied in previous research are reproduced in the data here (Andrews, Henderson, & Hall, 2001; Awas et al., 1999; Kessler et al., 2005; Patel et al., 1999; Weissman et al., 1996; WHO International Consortium in Psychiatric Epidemiology, 2000). Age, gender, and marital status are all signicant predictors of individual mental health in the direction of inuence found earlier. Mental health measures are positively and signicantly associated with age in every country but Tonga (Fig. 1A). However the magnitude of the age gradient varies substantially across countriesin four of the ve countries mental health worsens with age, while in Tonga the association between mental

Results Figs. 1 and 2 summarize the ndings from the ve countries; Fig. 1 presents mental health associations with continuous variables and Fig. 2 with discrete variables. The parametric version of these relationships using ordinary least squares are presented in Table 3.
See Yatchew (1998) for a description of these semi-parametric methods and Lokshin (2005) for programming implementation.
8

ARTICLE IN PRESS
J. Das et al. / Social Science & Medicine 65 (2007) 467480 473

A
Female 1 0.5 0 -0.5 -1 Bosnia India Indonesia Mexico Tonga

B
Poor physical health 1.5 1 0.5 0 -0.5 -1 Bosnia India Indonesia Mexico Tonga

C
Marital status (compared to non-married) 1 0.5 0 -0.5 -1 Married Married Married Married Widowed Widowed Widowed Widowed Married Widowed

D
Per capita household expenditures 1 0.5 0 -0.5 -1 2nd Quartile 3rd Quartile Top Quartile 2nd Quartile 3rd Quartile Top Quartile 2nd Quartile 3rd Quartile Top Quartile 2nd Quartile 3rd Quartile Top Quartile 2nd Quartile 3rd Quartile Top Quartile Bosnia India Indonesia Mexico Tonga

Bosnia

India

Indonesia

Mexico

Tonga

Fig. 2. Mental health score by selected discrete characteristics. (A) Female, (B) Poor physical health, (C) Marital status (compared to nonmarried) and (D) Per capita household expenditures.

health and age is an inverted-U shape whereby mental health scores improve at advanced ages. Bosnia exhibits the steepest age gradient by far. One of the most pronounced demographic regularities is that the odds of experiencing any disorder and specically of experiencing affective (mood), anxiety, and somatoform disorders are signicantly higher among females.9 Fig. 2A replicates this regularity. In four of the ve national settings, mental health measures are worse among women, although the magnitude of the female penalty varies widely from a low of 0.16 standard deviations in Indonesia to a high of 0.49 standard deviations in Mexico. Tonga is again the exception, where women report signicantly lower level of distress to the order of 0.2 standard deviations. A third consistent nding in the literature is that respondents who are separated, divorced or widowed report worse mental health compared to those who are married (Andrade, Walters, Gentil, & Laurenti, 2002; Andrews et al., 2001; Kessler et al., 2005; Weissman et al., 1996; WHO International Consortium in Psychiatric Epidemiology, 2000). Again, in a majority of countries examined here, widows are indeed worse off, although the relative
In one important exception, males tend to have higher odds of substance use disorders (Andrews et al., 2001).
9

deprivation of widows varies over the national setting and in the India data widows report lower levels of distress than non-married individuals (Fig. 2B). Physical health and mental health Similarities between these data and previously reported results are also evident in the positive association between mental and physical health (Kessler et al., 1994; Kessler et al., 2005; Bijl et al., 2003). Poorer physical health, as measured by a binary variable based on self-assessed general health status (poor health 0), is strongly associated with worse mental health outcomes in all countries with sufcient data. As observed in Fig. 2C, the coefcients are large and precisely estimated. The lowest magnitude is for India at 0.42 standard deviations; while in Bosnia, an individual reporting poor health also reports a mental health score 1.1 standard deviations higher than someone in good physical health. Socioeconomic status and mental health In sharp contrast to these results, which replicate those reported earlier in the literature, associations between socio-economic measures and mental

ARTICLE IN PRESS
474 Table 3 Correlates of mental health Tonga (1) Age Female dummy Married dummy Widowed dummy Poor physical health Years of education HH PCEquartile 2 HH PCEquartile 3 HH PCEquartile 4 Household size Old dependents Young dependents HH mental health Community mental health Primary to high school High school or more District xed effects Constant Observations R2
***

J. Das et al. / Social Science & Medicine 65 (2007) 467480

India (2) 0.00799* (0.0041) 0.356*** (0.078) 0.104 (0.12) 0.421* (0.23) 0.417*** (0.12)

Mexico (3) 0.00269*** (0.0010) 0.488*** (0.020) 0.0329 (0.030) 0.169*** (0.052) 0.940*** (0.060) 0.0197*** (0.0031) 0.520** (0.025) 0.0695*** (0.026) 0.0467* (0.028) 0.00361 (0.0054) 0.0685 (0.054) 0.0706 (0.050) 0.178*** (0.014) 0.119*** (0.0093) (4) 0.00263** (0.9910) 0.486*** (0.020) 0.0250 (0.029) 0.161*** (0.052) 0.943*** (0.060) 0.0219*** (0.0033) 0.0443* (0.026) 0.0617** (0.028) 0.0264 (0.030) 0.00262 (0.0054) 0.0451 (0.055) 0.0841* (0.050) 0.161*** (0.014)

Bosnia (5) 0.0127*** (0.00095) 0.298*** (0.016) 0.123*** (0.030) 0.272*** (0.051) 1.076*** (0.13) 0.00976*** (0.0016) 0.0323* (0.019) 0.0501*** (0.019) 0.0621*** (0.018) 0.0118** (0.0060) 0.355*** (0.035) 0.0758 (0.047) 0.472*** (0.018) 0.187*** (0.011) (6) 0.0123*** (0.00096) 0.300*** (0.016) 0.133*** (0.031) 0.290*** (0.052) 1.115*** (0.12) 0.0113*** (0.0018) 0.0317* (0.017) 0.0460*** (0.018) 0.0332* (0.018) 0.00892* (0.0052) 0.352*** (0.034) 0.0469 (0.046) 0.511*** (0.015)

Indonesia (7) 0.00266*** (0.00067) 0.164*** (0.015) 0.172*** (0.022) 0.0919** (0.042) 0.663*** (0.028) 0.00306 (0.0019) 0.00122 (0.018) 0.00372 (0.019) 0.0224 (0.021) 0.000318 (0.0024) 0.0448 (0.045) 0.0857** (0.038) 0.194*** (0.012) 0.124*** (0.0069) (8) 0.00270*** (0.00070) 0.161*** (0.015) 0.173*** (0.023) 0.0914** (0.042) 0.672*** (0.028) 0.00322 (0.0021) 0.00229 (0.019) 0.00560 (0.020) 0.0232 (0.023) 0.000034 (0.0026) 0.0742 (0.047) 0.115*** (0.040) 0.174*** (0.012)

0.00285 (0.0043) 0.221*** (0.069) 0.449*** (0.11) 0.604** (0.25)

0.0164 (0.019) 0.0119 (0.072) 0.0701 (0.072) 0.156** (0.076) 0.00715 (0.010) 0.504*** (0.17) 0.586*** (0.15) 0.484*** (0.038)

0.00644 (0.094) 0.0396 (0.089) 0.0180 (0.097) 0.0140 (0.016) 0.389 (0.24) 0.0288 (0.16) 0.321*** (0.053)

No 0.155 (0.25) 681 0.32

0.149* (0.084) 0.197* (0.10) No 0.330* (0.17) 747 0.18

No 0.386*** (0.055) 17926 0.19

Yes 0.355*** (0.058) 17926 0.21

No 0.634*** (0.049) 11766 0.61

Yes 0.656*** (0.049) 11766 0.60

No 0.183*** (0.035) 19584 0.15

Yes 0.211*** (0.038) 19584 0.16

Robust standard errors in parentheses clustered at the household level. po0.01, **po0.05, *po0.1.

health deviate considerably from expected patterns. Figs. 1B and 2D depict how mental health scores vary with years of formal schooling (Fig. 1) and quartiles of household per capita consumption (Fig. 2). For either socio-economic measure there is no clear pattern across the ve countries. In three of the ve countriesIndia, Mexico, and Bosniaeducation is signicantly and negatively associated with worse mental health. There

is no association across years of education in Indonesia and in Tonga the estimated relationship is U-shaped, with mental health rst improving as education increases and then worsening. Of equal interest are the relatively small magnitudes of the schooling coefcients compared to the demographic and physical health measures reported abovethe association between education and mental health, where it is signicant at all, is small.

ARTICLE IN PRESS
J. Das et al. / Social Science & Medicine 65 (2007) 467480 475

Any general relationship between mental health and household per capita expenditures is even more tenuous. Across the ve countries only twoTonga and Bosniaexhibit a negative gradient between household per capita consumption and individual mental health. The largest gradient is observed in Tonga, where individuals in the top quartile report on averaging a mental health score that is 0.15 standard deviations below the other three quartiles. However, to claim that the poor report worsens mental health in Tonga one would have to adopt an expansive denition of poverty that includes the bottom three quartiles of the distribution. Bosnia also exhibits a negative gradient, albeit much smaller in magnitude. In Indonesia and India, there is no association between per capita expenditures and mental health outcomes and in Mexico, there is a signicant positive gradient suggesting that mental health outcomes are better for the poor. Furthermore, for all ve countries, the magnitudes of the estimated socioeconomic coefcients are much smaller than the coefcients for any other factor presented here. For example, in Bosnia one more year of education is associated with a 0.01 standard deviation improvement in mental health, while moving from the bottom to the top quartile of the per capita expenditure distribution improves mental health by 0.03 standard deviations (Table 3). These pale in comparison with the associated worsening in mental health from being female (0.30 standard deviations), being widowed (0.29 standard deviations), having poor physical health (1.12 standard deviations), and having a one standard deviation worse mental health of other members in the household (0.51 standard deviations). Similar magnitudes are seen in other countries, suggesting that the relative importance of consumption poverty in the determination of mental health is actually quite slight. Spatial clustering of individuals with poor mental health In addition, the household-based nature of these data enables a unique exploration of the co-location of poor mental health within the household and communityfew other studies in the developing country literature are able to look at the covariation of poor mental health among household and community members. Fig. 1C depicts the nonparametric regression lines of individual mental health on the average mental health score of other

members in the household (excluding the individual herself).10 There is a strong positive association between an individuals psychological well-being and others in his or her constituent household. This association exists at all levels of mental health and is one of the most powerful predictors of the mental health score on the rough order of gender or physical health and certainly more inuential than any socio-economic measure. The community average mental health score also inuences an individuals mental well-being even after adjusting for household average mental health (Fig. 1D). The degree of association is roughly half as large as the association at the household level in the three countries that allow for community level measures (see Table 3).11 Determining the possible reasons for such observed co-location are beyond the scope of inquiry, but can include the negative externalities of poor mental health on family and neighbors, genetic predisposition towards poor mental health within families, and the uneven geographic distribution of mental and other health services.

Discussion Summary of main results Household surveys in ve low and middle-income countries covering Latin America, Eastern Europe, East Asia and the Pacic, and South Asia reveal signicant associations between mental health scores and gender, the physical health of the respondent, his/her marital status, and the mental well-being of other members in the respondents household and community. These relationships hold (with occasional deviations) across all the countries with roughly comparable magnitudes. In contrast, there is no consistent relationship between mental health scores and socio-economic measures such as the respondents education or the per capita expenditure of the household in which the respon10 Hence individuals living alone, representing 3.2% of the pooled data, are not included in the analysis. 11 An important distinction in low-income countries is between urban and rural areas. Rapid urbanization and economic restructuring are dening forces in much of the developing world and may lead to unique stressors (Blue & Harpham, 1996). In these data, urban residents report worse average mental health scores. However, controlling for residence through district xed effects leaves the interpretation of the regression coefcients unchanged. (The Indian data are exclusively an urban sample and the urban/rural distinction in Tonga has less meaning.)

ARTICLE IN PRESS
476 J. Das et al. / Social Science & Medicine 65 (2007) 467480

dent resides. These summary results provide the setting for a discussion centered around: (a) the potential use of mental health modules in multipurpose household surveys and (b) the implications for policy and research on mental health. Measuring mental health in multi-purpose surveys The consistency of the magnitudes and signs obtained across the ve countries suggest that mental health screening questionnaires can be incorporated into large and nationally representative standard household surveys such as the Living Standards Measurement Survey (Scott, Massagli, Kapetanovic, & Mollica Lavelle, 2005). Furthermore, the associations between mental health scores and individual/household characteristics are very similar in surveys where questionnaires were elded on a rst visit to households and where they were elded after a period of acquaintanceship. Finally, shorter modules (such as the GHQ-12) reveal similar associations as the longer SCL-90R, which took one hour to eld for non-literate respondents. Indeed, the depression and anxiety components of the SCL-90R are found to contain most of the relevant information for the nine dimensions covered under the full questionnaire (Das & Das, 2006). Implications for public health interventions On a more substantive note, effective public health policy requires an understanding of the mechanisms that determine poor mental health and, in turn, the implications of poor mental health for the individual and his/her family. The descriptive analysis here provides suggestive evidence for what these mechanisms may be and therefore a potential role for public health interventions. The lack of any relationship between conventional economic welfare measures and mental health outcomes across a diverse sample of developing countries suggests that poverty, per se, is not a strong determinant of poor mental health. A straightforward equity rationale for public investments in mental health is undermined by the frequently higher relative prevalence among the poor of other health problems such as tuberculosis and malaria, as well as continuing nancing gaps for these illnesses. The lack of a relationship between consumption poverty and

mental health is certainly not, however, supportive of arguments that suggest no scope for public interventions towards improving mental health. Instead, we argue that resources should be targeted towards improving the mental health of those who have experienced adverse events, and note also the distinction between severe and common mental disorders. Two of the strongest factors associated with poor mental health are poor physical health and widowhood. Related papers on India, Indonesia and Tonga conrm that, more generally, changes in life circumstances brought on by positive or negative events have long-lasting implications for mental health. In India, women who report childloss (either through miscarriages, abortions or death) are at signicantly higher risk of mental health problems compared to those without; indeed, the female penalty observed in the India data is entirely driven by the difference between men and women in households that experienced the loss of a child (Das & Das, 2006). In Indonesia, the mental health of the population worsened dramatically following the economic crisis of the nineties; however, although consumption levels recovered by 2000, mental health did not (Friedman & Thomas, 2006). Finally, in Tonga, individuals who were selected by a lottery to emigrate (and randomly received a positive income shock) reported signicantly better mental health outcomes after emigration (Stillman et al., 2006). These ndings are also consistent with the studies that report worsened mental health outcomes in populations that have suffered conict or disasters (Mollica et al., 1999; Mollica et al., 2001; Lopes Cardozo et al., 2004; United Nations High Commissioner for Refugees (UNHCR), 2005). The Indian and Indonesian studies suggest that the trauma from adverse events may persist long after the recovery of more traditional measures of welfare and there may very well be real individual and household costs to this persistence. Examples of such costs along the health dimension previously identied in the literature include lower adherence to dietary recommendations and medication regimes among diabetics with depressive symptoms compared to diabetics without (Ciechanowski, Katon, & Russo, 2000) high co-morbidity rates for smoking and psychiatric disorders, with smoking twice as common among the mentally ill compared to the mentally healthy population (Lasser et al.,

ARTICLE IN PRESS
J. Das et al. / Social Science & Medicine 65 (2007) 467480 477

2000) and an association between maternal mental health and child welfare, with maternal depression signicantly increasing the odds that a child will experience growth faltering (Harpham, Huttly, De Silva, & Abramsky, 2005; Patel, Rahman, Jacob, & Hughes, 2004). Strong evidence of such costs in other dimensions of welfare such as education also exist (Kessler, Foster, Saunders, & Stang, 1995). If individuals who have suffered adverse events (or shocks as in the economic literature) are particularly likely to report worse mental health, traditional measures of poverty such as per capita household expenditure are insufcient to fully understand the association between mental health and poverty. These poverty measures do not account for the risk and uncertainty that households face; alternative measures that incorporate risk and vulnerability could yield signicantly different results. Focusing on shocks and mental health outcomes suggests a dual role for policy. First, addressing the causes of poor mental health is a viable policy alternative for which there is already a strong global consensus in place. Few would argue that decreasing child mortality or improving physical health should not be a global priority; that such investments also have an effect on mental health strengthens an already existing case. Second, there may be a role for targeted treatments to the directly affected in the aftermath of an adverse event if such treatments lead to improved outcomes. The clustering of mental health outcomes within households provides evidence that such treatments targeted at the level of the household may have larger benets than those targeted to individuals. An important limitation of this study and of the household-survey-based methodology is our inability to differentiate common from severe mental disorders. A clear distinction has been remarked on in the literature, especially in the context of ndings that the annual prevalence of common mental disorders exceeds 10% in many countries, and is as high as 16.9% in Lebanon, 17.8% in Colombia, 20.4% in Ukraine, and 26.3% in the US (WHO World Mental Health Survey Consortium, 2004). Severe mental health problem (such as schizophrenia), brought on by biogenetic causes and possible interactions with environment, require a separate policy response. In several low-income countries, the institutional

capacity for treating such disorders is very poor with frequent human-rights violations of the severely mentally ill (WHO, 2001). Neither does it appear that the private sector is capable of providing the required responsedoctors tested on the handling of a patient with depression in Delhi had to be above average competence to have a better than even chance of not harming the patient; even those in the highest quintile of competence provided a harmless treatment only 58% of the time (Das & Hammer, 2005). The longterm treatment required for such disorders and the high costs imposed on households suggest that these are the types of disorders where the lack of insurance markets requires clear government intervention. A second limitation of this study is that, in the absence of an experimental setup, the associations presented are consistent with multiple interpretations. For instance, the concordance of mental health outcomes within households could reect unobserved household-level shocks, assortative matching (where those in poor mental health are more likely to marry each other), genetic links between parents and children, or a contagion effect, whereby caring for a mentally ill person in the household in turn affects the mental well-being of others. Longitudinal data and experimental mental health interventions are needed to try and separate these channels. These results ask for a more nuanced understanding of the relationship between poverty and mental health. Two potential avenues for further research suggested by these ndings concern the long- and short-term effects of negative and positive shocks to mental health, as well as the link between mental health outcomes and broader measures of welfare that incorporate risk and vulnerability in their construction. Acknowledgement We would like to thank Alison Buttenheim and Le Dang Trung for expert research assistance. All errors accrue to the authors. The views presented herein do not reect the views of the World Bank or any of its afliates and only reect on the authors. Appendix Correlates to severe mental health (see Table A1).

478

Table A1 Marginal effects from probit estimation of being in the worst 10% or worst 5% of mental health scores Tonga (1) Worst 10% Age Female dummy Married dummy Widowed dummy Years of education HH incomequartile 2 HH incomequartile 3 HH incomequartile 4 Household size Old dependents Young dependents HH mental health Primary to high school High school or more Poor physical health Community mental health Observations 681 747 0.0000237 (0.00049) 0.0120 (0.011) 0.0260* (0.014) 0.0883 (0.13) 0.00127 (0.0011) 0.00108 (0.010) 0.00768 (0.0096) 0.000399 (0.012) 0.00292* (0.0016) 0.0860** (0.042) 000465* (0.025) 0.0326*** (0.0059) India (2) Worst 10% 0.000972 (0.0010) 0.0449** (0.021) 0.0169 (0.032) 0.0431 (0.036) Mexico (3) Worst 10% 0.000463* (0.00026) 0.0791*** (0.0056) 0.0140* (0.0075) 0.0339** (0.016) 0.00442*** (0.00079) 0.00566 (0.0065) 0.0139* (0.0077) 0.00239 (0.0082) 0.000644 (0.0014) 0.0117 (0.015) 0.0142 (0.014) 0.0228*** (0.0025) (4) Worst 10% 0.000103 (0.00017) 0.0371*** (0.0036) 0.0113** (0.0048) 0.0234* (0.012) 0.00213*** (0.00051) 0.00295 (0.0043) 0.00333 (0.0048) 0.00289 (0.0054) 0.000736 (0.00087) 0.0116 (0.0085) 0.00302 (0.0092) 0.0100*** (0.0015) Bosnia (5) Worst 10% 0.000810*** (0.00016) 0.0216*** (0.0039) 0.0131*** (0.0047) 0.0496** (0.024) 0.000281 (0.00024) 0.00369 (0.0032) 0.00470 (0.0031) 0.00824*** (0.0027) 0.00187* (0.00099) 0.0230*** (0.0059) 0.00771 (0.0095) 0.0222*** (0.0021) (6) Worst 10% 0.000235*** (0.000053) 0.00636*** (0.0017) 0.00234 (0.0016) 0.00757 (0.0062) 0.000168 (0.00010) 0.00100 (0.0012) 0.000936 (0.0014) 0.00177* (0.0011) 0.000471 (0.00047) 0.00758*** (0.0022) 0.00756* (0.0041) 0.00653*** (0.00096) Indonesia (7) Worst 10% 0.000278 (0.00018) 0.0206*** (0.0040) 0.0245*** (0.0066) 0.0178 (0.011) 0.000715 (0.00053) 0.00623 (0.0046) 0.00882* (0.0049) 0.0128** (0.0054) 0.000463 (0.00069) 0.00232 (0.013) 0.0138 (0.011) 0.0266*** (0.0020) (8) Worst 10% 0.0000658 (0.000099) 0.00989*** (0.0023) 0.00532 (0.0037) 0.0141* (0.0074) 0.000538* (0.00030) 0.00277 (0.0026) 0.00534** (0.0027) 0.00624** (0.0030) 0.00000528 (0.0030) 0.00225 (0.0068) 0.00169 (0.0061) 0.0106*** (0.0011)

J. Das et al. / Social Science & Medicine 65 (2007) 467480

ARTICLE IN PRESS

0.0244 (0.021) 0.0157 (0.024) 0.0213 (0.027) 0.00647 (0.0045) 0.0262 (0.061) 0.102** (0.050) 0.0486*** (0.0091) 0.0184 (0.024) 0.0276 (0.026) 0.0933** (0.043)

0.176*** (0.019) 0.0150*** (0.0022) 17926

0.0964*** (0.015) 0.00850*** (0.0013) 17926

0.156*** (0.035) 0.0110*** (0.0019) 11766

0.0941*** (0.023) 0.00252*** (0.00071) 11766

0.129*** (0.0087) 0.0169*** (0.0018) 19584

0.0679*** (0.0062) 0.00669*** (0.0011) 19584

Note: Marginal effects are the change in probability associated with a discrete change in dummy variables from 0 to 1 and with an innitesimal change in continuous variables. Robust standard errors in parentheses. *** po0.01, **po0.05, *po0.1.

ARTICLE IN PRESS
J. Das et al. / Social Science & Medicine 65 (2007) 467480 479

References
Andrade, L., Walters, E. E., Gentil, V., & Laurenti, R. (2002). Prevalence of ICD-10 mental disorders in a catchment area in the city of Sao Paulo, Brazil. Social Psychiatry and Psychiatric Epidemiology, 37(7), 316325. Andrews, G., Henderson, S., & Hall, W. (2001). Prevalence, comorbidity, disability and service utilization. Overview of the Australian National Mental Health Survey. British Journal of Psychiatry, 178, 145153. Awas, M., Kebede, D., & Alem, A. (1999). Major mental disorders in Butajira, southern Ethiopia. Acta Psychiatrica Scandinavia (Supplement), 397, 5664. Bartel, A., & Taubman, P. (1986). Some economic and demographic consequences of mental illness. Journal of Labor Economics, 4(2), 243256. Berwick, D. M., Murphy, J. M., Goldman, P. A., Ware, J. E., Jr., Barsky, A. J., & Weinstein, M. C. (1991). Performance of a ve-item mental health screening test. Medical Care, 29(2), 169176. Bijl, R. V., de Graaf, R., Hiripi, E., Kessler, R. C., Kohn, R., Offord, D. R., et al. (2003). The prevalence of treated and untreated mental disorders in ve countries. Health Affairs (Millwood), 22(3), 122133. Blue, I., & Harpham, T. (1996). Urbanization and mental health in developing countries. Current Issues in Public Health, 2(4), 181185. Chiquiar, D., & Hanson, G. (2005). International migration, selfselection, and the distribution of wages: Evidence from Mexico and the United States. Journal of Political Economy, 113(2), 239281. Ciechanowski, P. S., Katon, W. J., & Russo, J. E. (2000). Depression and diabetes: Impact of depressive symptoms on adherence, function, and costs. Archives of Internal Medicine, 160(21), 32783285. Das, J., & Das, V. (2006). Mental health in urban India: Patterns and narratives. The World Bank. Processed. Das, J., & Hammer, J. (2005). Which doctor? Combining vignettes and item response to measure clinical competence. Journal of Development Economics, 78, 348383. nchez-Pa ramo, C. (2003). Short but not sweet: New Das, J., & Sa evidence on short duration morbidities from India. Policy Research Working Paper 2971, World Bank, Development Research Group, Washington, DC. Deaton, A., & Zaidi, S. (2002). Guidelines for constructing consumption aggregates for welfare analysis. Living Standards Measurement Study Working Paper No. 135, The World Bank. Friedman, J., & Thomas, D. (2006). Mental health during a crisis. Mimeo, The World Bank. Goldberg (1972). The detection of psychiatric illness by questionnaire. London: Oxford University Press. Johnson, J. G., Cohen, P., Dohrenwend, B. P., Link, B. G., & Brook, J. S. (1999). A longitudinal investigation of social causation and social selection processes involved in the association between socioeconomic status and psychiatric disorders. Journal of Abnormal Psychology, 108(3), 490499. Kessler, R. C., Chiu, W. T., Demler, O., & Walters, E. E. (2005). Prevalence, severity, and comorbidity of 12-month DSM-IV

disorders in the National Comorbidity Survey Replication. Archives of General Psychiatry, 62(6), 617627. Kessler, R. C., Foster, C. L., Saunders, W. B., & Stang, P. E. (1995). Social consequences of psychiatric disorders, I: Educational attainment. American Journal of Psychiatry, 152(7), 10261032. Kessler, R. C., McGonagle, K. A., Zhao, S., Nelson, C. B., Hughes, M., Eshleman, S., et al. (1994). Lifetime and 12month prevalence of DSM-III-R psychiatric disorders in the US. Results from the National Comorbidity Survey. Archives of General Psychiatry, 51(1), 819. Lasser, K., Boyd, J. W., Woolhandler, S., Himmelstein, D. U., McCormick, D., & Bor, D. H. (2000). Smoking and mental illness: A population-based prevalence study. JAMA, 284(20), 26062610. Lokshin, M. (2005) Semi-parametric difference-based estimation of partial linear regression models. Stata adole with help and documentation. Washington, DC: World Bank [Available at: /http://econ.worldbank.org/programs/ poverty/toolkitS]. Lopes Cardozo, B., Bilukha, O. O., Crawford, C. A., Shaikh, I., Wolfe, M. I., Gerber, M. L., et al. (2004). Mental health, social functioning, and disability in postwar Afghanistan. JAMA, 292(5), 575584. McCabe, C. J., Thomas, K. J., Brazier, J. E., & Coleman, P. (1996). Measuring the mental health status of a population: A comparison of the GHQ-12 and the SF-36 (MHI-5). British Journal of Psychiatry, 169, 516521. McKenzie, D., Gibson, G., & Stillman, S. (2006). How Important is Selection? Experimental vs. non-experimental measures of the income gains from migration, World Bank Policy Research Working Paper no. 3906. Miech, R. A., Caspi, A., Moftt, T. E., Wright, B. R. E., & Silva, P. A. (1999). Low socioeconomic status and mental disorders: A longitudinal study of selection and causation during young adulthood. American Journal of Sociology, 104(4), 10961131. Miranda, J. J., & Patel, V. (2005). Achieving the millennium development goals: Does mental health play a role? PLoS Medicine, 2(10), 962965. Mollica, R. F., McInnes, K., Sarajlic, N., Lavelle, J., Sarajlic, I., & Massagli, M. P. (1999). Disability associated with psychiatric comorbidity and health status in Bosnian refugees living in Croatia. JAMA, 282(5), 433439. Mollica, R. F., Sarajlic, N., Chernoff, M., Lavelle, J., Vukovic, I. S., & Massagli, M. P. (2001). Longitudinal study of psychiatric symptoms, disability, mortality, and emigration among Bosnian refugees. JAMA, 286(5), 546554. Patel, V., Araya, R., de Lima, M., Ludermir, A., & Todd, C. (1999). Women, poverty and common mental disorders in four restructuring societies. Social Science & Medicine, 49(11), 14611471. Patel, V., & Kleinman, A. (2003). Poverty and common mental disorders in developing countries. Bulletin of the World Health Organization, 81(8), 609615. Patel, V., Pereira, J., Coutinho, L., Fernandes, R., Fernandes, J., & Maan, A. (1998). Poverty, psychological disorder, and disability in primary care attenders in Goa, India. British Journal of Psychiatry, 172(June), 533536. Patel, V., Rahman, A., Jacob, K. S., & Hughes, M. (2004). Effect of maternal mental health on infant growth in low income

ARTICLE IN PRESS
480 J. Das et al. / Social Science & Medicine 65 (2007) 467480 Veit, C. T., & Ware, J. E., Jr. (1983). The structure of psychological distress and well-being in general populations. Journal of Consulting and Clinical Psychology, 51, 730742. Weissman, M. M., Bland, R. C., Canino, G. J., Faravelli, C., Greenwald, S., Hwu, H. G., et al. (1996). Cross-national epidemiology of major depression and bipolar disorder. JAMA, 276(4), 293299. WHO International Consortium in Psychiatric Epidemiology. (2000). Cross-national comparisons of the prevalences and correlates of mental disorders. Bulletin of the World Health Organization, 78(4), 413426. WHO. (2001). The World Health reportMental health: New understanding, new hope. World Health Organization. WHO World Mental Health Survey Consortium. (2004). Prevalence, severity, and unmet need for treatment of mental disorders in the World Health Organization World Mental Health Surveys. JAMA, 291(21), 25812590. Yatchew, A. (1998). Nonparametric regression techniques in economics. Journal of Economic Literature, 36(2), 669721. countries: New evidence from South Asia. British Medical Journal, 328(7443), 820823. Radloff, L. S. (1977). The CES-D scale: A self-report depression scale for research in the general population. Applied Psychological Measurement, 1, 385401. Ravallion, M. (1994). Poverty comparisons. Fundamentals in pure and applied economics, Vol. 56. Chur: Harwood Academic Press. Scott, K.M., Massagli, A., Kapetanovic, R., Mollica Lavelle, J. (2005). Measuring mental health in post-conict societies: An assessment of the Bosnia and Herzegovina experiment, mimeo, World Bank. Stillman, S., McKenzie, D., & Gibson, J. (2006). Migration and mental health: Evidence from a natural experiment. BREAD Working Paper No. 123. United Nations High Commissioner for Refugees (UNHCR). (2005). Refugees by the Numbers. Retrieved January 17, 2006, from /http://www.unhcr.ch/cgi-bin/texis/vtx/home/ opendoc.htm?tbl=BASICS&id=3b028097c&page=basics# NumbersS.