Trustful societies, trustful individuals, and health

Health & Place 16 (2010) 1022–1029 Contents lists available at ScienceDirect Health & Place journal homepage: www.elsevier.com/locate/healthplace Trustful societies, trustful individuals, and health: An analysis of self-rated health and social trust using the World Value Survey Min Hua Jen n,a, Erik R. Sund b, Ron Johnston c, Kelvyn Jones c a Department of Primary Care and Public Health, Imperial College London, London EC1a 9LA, UK Department of Geography, Faculty of Social Sciences and Technology Management, Norwegian University of Science and Technology, Trondheim, Norway c School of Geographical Sciences, University of Bristol, BS8 1SS, UK b a r t i c l e in f o a b s t r a c t Article history: Received 11 November 2009 Received in revised form 8 June 2010 Accepted 12 June 2010 This study analyses the relationships between self-rated health and both individual and mean national social trust, focusing on a variant of Wilkinson’s hypothesis that individuals will be less healthy the greater the lack of social cohesion in a country. It employs multilevel modelling on World Values Survey data across 69 countries with a total sample of 160,436 individuals. The results show that self-rated health are positively linked to social trust at both country and individual levels after controlling for individual socio-demographic and income variables plus individual social trust; increased trust is associated with better health. Moreover, this analysis of social trust gives some insight into distinctive results for the former Soviet Bloc countries, which have high reported levels of poor health, alongside the Scandinavian countries which have high levels of trust and better health situations. Our results support and extend the Wilkinson hypothesis that the level of trust, an indicator of social cohesion, is predictive of individuals’ health. & 2010 Elsevier Ltd. All rights reserved. Keywords: Social trust Self-rated health World value survey Multilevel modelling 1. Introduction 1.1. Population level health determinants There has been considerable recent interest in the links between characteristics of nation states and their populations’ health, partly spurred by Wilkinson’s (1996) Unhealthy societies which identified links between income inequalities, social cohesion, psychosocial stress and poor health outcomes. He claimed that western societies with increasing income inequalities experience poorer population health because of both the greater social distance between top and bottom of the society and the dissolving effect this has on the social fabric (see also Wilkinson and Pickett, 2009). Although disputed on many grounds, the most controversial issue of his work has perhaps been the relative weight given to the psychosocial mechanisms linking these contextual determinants to individual health. Partly with the support from the Whitehall studies (Marmot et al., 1991; Marmot and Shipley, 1996), Wilkinson claims that in advanced societies material disadvantage is no longer the most important determinant of health variations, but rather the psychosocial feelings associated with relative deprivation. The idea of such group social effects is not new: Durkheim (1895–1982) argued in The Rules of Sociological Methods that the effect of the group is more than a sum of its parts. In epidemiology, Rose (1985) similarly argued that the key to understanding disease incidence and prevalence lies in the characteristics of populations— not individuals. Wilkinson, Durkheim and Rose all apply a systems perspective in which societies are conceptualised as more than the mere aggregate of individuals. They suggest that some populationlevel, or similar macro-level, factors influence individuals independently of the individual’s own characteristics—an idea in stark contrast to the methodological individualism present in epidemiology where disease prevalence is explained solely in terms of facts about individuals (Diez-Roux, 1998a, 1998b, 2004). The two approaches generate very distinct policy recommendations—as illustrated by Marmot (2001, p. 158): ‘‘If a water supply becomes polluted, one could recommend that individuals buy bottled water, or one could take steps to clean the water supply’’. The former is an individual-level response, the latter a societal response. The population level becomes policy relevant in the aggregate approach although exactly what macro-level factors are responsible for varying disease rates in the developed world is heavily debated. A competing hypothesis to Wilkinson’s income inequality hypothesis states that countries can be distinguished by the degree of income redistribution policies (Lynch et al., 2000). This so-called neo-material hypothesis claims that material inequality affects health rather than any associated feelings of relative deprivation. n Corresponding author. Tel.: +44 0 2073328959. E-mail addresses: m.h.jen@imperial.ac.uk (M.H. Jen), erik.r.sund@svt.ntnu.no (E.R. Sund), R.Johnston@bristol.ac.uk (R. Johnston), kelvyn.jones@bristol.ac.uk (K. Jones). 1353-8292/$ - see front matter & 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.healthplace.2010.06.008 M.H. Jen et al. / Health & Place 16 (2010) 1022–1029 Its advocates do not deny the operation of psychosocial mechanisms but argue that the interpretation of the links between income inequalities and health must address the structural causes before focusing on the perceptions of these inequalities. Yet another line of inquiry states that political traditions, or the political ideologies of the governing parties, matter for the health of populations (Navarro et al., 2006; Navarro and Shi, 2001), a view in accordance with the classification into welfare regime types often used in the social sciences (Esping-Andersen, 1990, 1999). 1.2. Two views on social capital In this paper, we restrict our discussion of macro-level factors that might determine health outcomes to the concepts of social cohesion and social capital, which many researchers use interchangeably. Although they share some common features in terms of their components, however, they diverge considerably. Putnam et al. (1993) gave the social capital literature a major boost in Making Democracy Work in Italy. Regions in northern Italy had long traditions of cooperation and participation in organizations and performed better politically as well as economically than those in the south. In his later Bowling Alone, Putnam (1995) argued that the USA has suffered a major collapse in civic-, social-, associational-, and political life (social capital) since the 1960s, with serious negative consequences. He also makes an important distinction between two kinds of social capital. Bonding capital occurs when you are socializing with people like yourself: same age, same race, same religion, same socioeconomic status and so on. But in order to create peaceful and trustful societies in heterogeneous multi-ethnic countries, the key form is bridging social capital, which refers to the connections between dissimilar people and the extent of mutual trust. Both types are partly a byproduct of socialization in voluntary organizations, hence we refer to this as the ‘civil society perspective’. According to an ‘institutional perspective’, social capital and especially generalized social trust (i.e. trust towards strangers) is considered as a consequence of various institutional arrangements related to the society’s degree of economic equality and equality of opportunity (Kumlin and Rothstein, 2005; Rothstein and Stolle, 2003; Rothstein and Uslaner, 2005). The kernel of the argument is that universal social policies are more effective than selective ones in creating both types of equality and thereby social trust. By establishing universal social programmes (i.e. a welfare state) a government sends important signals to its citizens that are important for the creation of both solidarity and trust, an argument that clearly resembles Wilkinson’s social cohesion perspective, and to some extent the neo-material perspective. The common denominator is the emphasis given to the state as an important actor influencing the distribution of welfare and resources and ultimately health and human well-being. What becomes apparent when we contrast the civil society perspective (a bottom-up perspective) and the institutional perspective (a top-down perspective) is the policy implications. According to Putnam the main problem in societies with weak social capital is a lack of civic spirit and participation in associations; the policy prescriptions that follow are to get individuals more involved in voluntary associations and organizational life. The institutional perspective, in contrast, suggests more profound state-led social change to enhance social capital and ultimately population health. 1.3. Empirical evidence of social trust: social trust and health Much of the early work relating social capital and health used aggregate data and so demonstrated an ecological association 1023 only, with the potential for committing an ecological fallacy—i.e. assuming an individual relationship when none has been directly demonstrated. Further, those studies disagree: Kawachi et al. (1997), for example, found associations among social mistrust, income inequality and all-cause mortality across 39 US states, but Lynch et al. (2001), Lindstrom and Lindstrom (2006) and Kennelly et al. (2003) all found no link between social capital and health measures across 16, 19 and 20 countries, respectively. Ram (2006), on the other hand, found a weak relationship. Other studies have avoided the potential of an ecological fallacy by combining individual and aggregate (i.e. national contextual) data. Both Poortinga (2006) and Mansyur et al. (2008) identified no relationship between social capital and selfassessed health across a range of countries (as did Olsen and Dahl, 2007), but found that individuals with high levels of social capital in countries with high average levels reported better health than similar individuals in countries with lower aggregate levels of social capital; within Europe, Rostila (2007) found significant inter-country differences according to the type of welfare state regime. Within the United States, using the individual states as the contextual level, Subramanian et al. (2001) found that the probability of reporting poor health increased as aggregate social capital declined. One feature of several of the international studies is that the countries of the former Soviet bloc stand out as having particular patterns. Kennedy et al. (1998, p. 2029), for example, argued that ‘Citizens living in societies with a high degree of social cohesion – characterized by strong social networks and high levels of interpersonal trust – seem to be healthier than those living in socially disorganized societies’. In many parts of contemporary Russia, they argued, the absence of a strong civil society meant that people had to rely on informal sources of support (from family and friends, for example), and that those lacking such support – i.e. having low levels of social capital – may have been especially vulnerable to the hardships following the rapid and disruptive transition to a market economy. In our analyses, we explore whether this vulnerability remains a characteristic of the post-Soviet states. In this study, therefore, we apply multilevel modelling to World Values Survey data building on the papers discussed above (especially Subramanian et al., 2001, 2002; Kennedy et al., 1998) to investigate how a contextual characteristic, the national average level of social trust, is associated with self-rated health. We expect to find – if Wilkinson’s arguments are valid for a much wider selection of countries than those he studied – that selfrated health is related to both individual and national levels of social capital; individuals with high levels of social capital should be healthier, and those with high levels living in countries with high aggregate levels should be even healthier than similar individuals living in countries with lower levels. 2. Data 2.1. The World Values Survey data The analysis was based on the cross-national scientific samples of individuals undertaken in the World Values Surveys (Inglehart, 1997, 2004) and European Values Surveys; we amalgamate data from 60 surveys taken in four waves (1981, 1991, 1995–1997 and 1999–2001) across up to 69 countries which together represent almost 80 percent of the world’s population. In each country, the investigators provided a 4-digit weight variable to correct their sample to reflect national distributions on key variables. If no weighting was necessary, each case was simply weighted as 1. Given the relatively large size of the sample in each country for 1024 M.H. Jen et al. / Health & Place 16 (2010) 1022–1029 each wave (the minimum is 1000); there should not be a substantial problem of sampling reliability. With these data we cover a much wider range of countries (hence various levels of economic and social development as well as variations in political regime) than either Wilkinson or the many studies which have followed his lead. 2.2. Outcome measure Self-reported overall health status of individuals has been commonly used as a valid indicator of a population’s health status. This is widely used in a range of studies – many more than the 27 reviewed by Idler and Benyamini (1997), referenced by Barfordet al. (2010). Indeed in a recent web-reported interview, Ellen Idler argues that the evidence of a link between self-rated health and mortality at the individual level is overwhelming (Johnston et al., 2010). It was determined from people’s response to the following question: ‘‘How would you describe your state of health these days? Would you say it is excellent, very good, good, fair, poor?’’ We reclassified the fivefold category to form a dichotomous outcome of self-rated health in which 0 is for excellent, very good, and good and 1 for fair and poor. In other words, we analyzed the underlying probability of reporting fair/ poor health rather than excellent/good. 2.3. Independent variables To evaluate Wilkinson’s arguments, self-rated health will be related to predictors at both the individual-level and countrylevel. At the individual-level (Table 1), we consider key sociodemographic variables (age, gender, marital status) plus income. Age is the only continuous predictor, with a mean of 40 across all countries and waves. Marital status is consolidated from seven into four categories – couple (combining married couple and liveas-married couple), single, unknown and SWD (separated/ widowed/divorced as a group). Individual income is assessed on a separate quintile scale of household income distribution for each country, incorporating all wages, salaries, pensions and other incomes before taxes and other deductions, relative to the Table 1 Descriptive statistics for individual and country level variables. Response: ‘‘Fair and poor’’ health (n¼ 59,483, 37.1%) vs. ‘‘Excellent, Very Good and Good’’ health (n ¼100,953, 62.9%) Predictor variables Level 1, individuals, n ¼160,436: individual level Age Mean¼ 40 years (range¼ 15–97 years) Gender Base: Male (n¼ 77,974, 48.6%) Female (n¼82,462, 51.4%) Marital status Base: Married (n¼104,871, 65.4%) Single (n¼ 37,960, 23.7%) Separated/widowed/divorced (n¼ 17,321, 10.8%) Unknown (n¼284, 0.2%) Income (quintiles) Base: 3 (n¼ 34,065, 21.2%) 1(n¼ 31,168, 19.4%) 2 (n¼ 39,482, 24.6%) 4 (n¼ 23,372, 14.6%) 5 (n¼ 13,376, 8.4%) Unknown (n¼18,973, 11.8%) Individual social trust Base: Low trust (n¼110,895, 69.1%) High trust (n¼49,541, 30.9%) Level 2 Wave, n¼116 Base: Wave 1 (n¼ 19,460, 12.1%) countrynwave Wave 2 (n ¼45,889, 28.6%) Wave 3 (n ¼51,303, 32.0%) Wave 4 (n ¼43,784, 27.3%) Level 3, Countries, n ¼69: country-level Aggregated social trust Mean¼ 0.308 (range¼ 0.03–0.66) respondent’s national norm. The central group (i.e. the third quintile, including the median) is used as the comparator in the regression equations. Perceptions of individual trust were determined by individual responses to a general question on interpersonal trust (‘‘Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?’’) with the potential responses being ‘‘people can be trusted’’ as high trust and ‘‘you can’t be too careful’’ as low trust. At the country level, a contextual social trust variable (the percentage of respondents saying that ‘people can be trusted’), measured on a continuous scale, was derived by aggregating the individual responses. Values were calculated by taking the arithmetic average of the weighted individual-level measures for each country at each wave to approximate national population values. 3. Analysis Since the response is binary (fair and poor versus good and very good health) and taking into account the clustering of individual responses (level 1) nested within 4 waves (level 2) nested within 69 countries (level 3), we analyse these data as a three-level logistic binomial model based on a logit-link function (Goldstein and Rashbash, 1996). Multilevel logistic regression has been implemented using the MLwiN program (version 2.10; Rasbash et al., 2004) with MCMC estimation procedures (Markov Chain Monte Carlo; Gilks et al., 1996 gives more in-depth material). This corrects for any underestimation that could have occurred using maximum likelihood estimations (such as predictive/penalized quasi likelihood approximation) and allows the calculation of the deviance information criterion (DIC) which provides a comparative measure of how well the model has been fitted. The smaller the value, the better the model (Spiegelhalter et al., 2002). For ease of interpretation we have transformed the logits to proportions, odds ratios (OR) or both. A sequence of four models is fitted. Model 1 is the null model which does not include any predictors in its fixed part, with the exception of a set of dummy variables indicating wave (wave 1 as reference). This provides a baseline for the comparison of the degree to which the individual- and country-level variables account for the variation in self-rated health between countries or within country between waves in subsequent models. Model 2 builds on Model 1 by including all the individual predictors (except individual trust) in its fixed part, and so assesses the effect of individual predictors alone on self-rated poor health. In addition to the variables listed above and in Table 1 we also included an agensex interaction. This model thus looks at compositional variables alone. Model 3 extends Model 2, incorporating contextual variables by adding the fixed effect of country-aggregated social trust on individual self-rated poor health and the extent to which it explains the country-level differences. Model 4 extends Model 3, by incorporating the effect of interpersonal trust at the individual level to evaluate the relative importance of individual-level versus country-level social trust (Model 4A). In addition, we also considered how the effect of country social trust on selfrated poor health differed for low- and high-trust individuals (Model 4B). 4. Results Table 1 summarises the data used. Except for age, individual characteristics were specified as categorical variables, with a base and a set of contrast indicator dummies. The total number of 1025 M.H. Jen et al. / Health & Place 16 (2010) 1022–1029 Fig. 1. Between-country differences (based on Model 1); logit and 95% confidence intervals against rank of country. Each triangle symbol indicates a country; a solid triangle highlights a former Soviet Bloc country in which the 5 reporting the poorest self-related health are: Moldova, Hungary, Belarus, Ukraine and Russia. individual observations from 69 countries across four waves was 163,328. After excluding missing data on the predictor variables, we conducted a multilevel logistic regression analysis on 160,436 individuals nested within 4 waves (1981, 1991, 1995–97 and 1999–2000) at level 2 for 69 countries at level 3 (Table 1). Some 46.7% of the respondents in 1981 reported being in poor/fair health, but that percentage reduced to 19.1 in 2000. Gradually, fewer people said they are in poor health, but this apparent health improvement may at least in part be the result of different countries participating in different waves. Table 2 gives the results of sequential models as parameter coefficients and their standard errors plus the model-fitted diagnostic, DIC. A positive coefficient indicates a higher probability of reporting poor health compared to the reference group. A statistically significant reduction of DIC (an absolute reduction of 4 when one extra variable is included in the model) of a current model from the previous model indicates that the former is a better fit. Fig. 1 shows the between-country variation as point estimates on a logit scale, for poor health in comparison to good, using values derived across all the waves of the surveys. A ‘caterpillar plot’ shows the logit estimates and their 95% confidence intervals plotted against rank of variation from the global average at country level with the former Soviet States highlighted. The horizontal dashed line indicates the average of all countries set to Table 2 Parameter estimates of the logistic binomial analytical models (in logits) and deviance information criterion (MCMC). Fixed parameters Constant Wave 1 (base) Wave 2 Wave 3 Wave 4 Individual predictors Female Age (centered around 40) Femalenage Marital status Couple (base) Widowed/separate/divorced Single Unknown Income (quintiles) 3 (base) 1 2 4 5 Unknown Social capital High trust Country predictor Trust Individual/country interaction High trustntrust Random parameters Level 3 Between countries Covariance Individual low trust Level 2 Within country between waves Covariance Individual high trust Deviance information criterion Change in DIC from previous model n Model 1 (S.E.) Model 2 (S.E.) n Model 3 (S.E.) n  0.536 (0.106) Model 4a (S.E.) n  0.432 (0.106)n  0.331 (0.106)  0.530 (0.112)  0.045 (0.071)  0.186 (0.080)n  0.387 (0.087)n  0.193 (0.073)n  0.342 (0.081)n  0.543 (0.089)n  0.133 (0.075)n  0.373 (0.082)n  0.586 (0.090)n  0.131 (0.075)n  0.375 (0.083)n  0.587 (0.090)n  0.129 (0.075)n  0.374 (0.083)n  0.586 (0.090)n 0.268 (0.012)n 0.035 (0.001)n 0.001 (0.001) 0.268 (0.012)n 0.035 (0.001)n 0.001 (0.001) 0.269 (0.012)n 0.035 (0.001)n 0.001 (0.001) 0.269 (0.012)n 0.035 (0.001)n 0.001 (0.001) 0.176 (0.020)n 0.019 (0.016) 0.056 (0.140) 0.176 (0.020)n 0.019 (0.016) 0.057 (0.140) 0.171 (0.020)n 0.022 (0.016) 0.052 (0.141) 0.171 (0.020)n 0.022 (0.016) 0.050 (0.141) 0.536 (0.019)n 0.230 (0.017)n  0.193 (0.021)n  0.380 (0.027)n 0.148 (0.022)n 0.535 (0.019)n 0.230 (0.017)n  0.192 (0.021)n  0.379 (0.027)n 0.149 (0.022)n 0.526 (0.019)n 0.225 (0.032)n  0.187 (0.021)n  0.360 (0.027)n 0.140 (0.022)n 0.526 (0.019)n 0.225 (0.017)n  0.186 (0.021)n  0.359 (0.027)n 0.140 (0.022)n  0.357 (0.013)n  0.350 (0.014)n  1.338 (0.446)n  1.260 (0.448)n  1.624 (0.446)n  0.432 (0.106) Model 4b (S.E.) n  0.234 (0.104)n 0.473 (0.086)n 0.538 (0.097)n 0.442 (0.081)n 0.441 (0.081)n 0.440 (0.081)n 0.038 (0.008)n 0.039 (0.009)n 0.040 (0.009)n 0.040 (0.009)n 0.040 (0.009)n 195294.98 – 181867.99 13426.99n 181861.19 6.80n 181104.22 763.04n 181098.57 5.65n All estimates marked by * are significant at 0.05 probability level. 1026 M.H. Jen et al. / Health & Place 16 (2010) 1022–1029 zero. A number of countries are significantly either above or below the average (i.e. their mean value and its standard error band – shown in the diagram – are either above or below the global mean). Countries which reported low levels of poor health (on the left of the diagram) are mainly western wealthy economies (Ireland, Switzerland, Canada, etc.) including the Nordic countries (Denmark, Norway, Finland and Sweden). Nine out of ten countries which reported the highest levels of poor health are former Soviet States (shown by solid triangles in Fig. 1 in which the 5 reporting the poorest self-related health are: Moldova, Hungary, Belarus, Ukraine and Russia). In the random part of the results shown in Table 2, the level 2 variances within countries between waves are much smaller (0.038) than those between countries (0.473). Thus, there was some variation within countries across waves, but the effects are not as substantial as between-country differences. Those countries with poor health tended to remain so across time. 4.1. Compositional effects Model 2 adds the full range of individual variables for respondents, and the reference group is a 40-year-old, married male who belongs to the middle income quintile. The main effect for each of the covariates was estimated after controlling for the remaining ones (see Table 2, model 2). There is a substantial improvement as shown by the substantial lowering of the DIC from 195294.88 to 181867.99. As expected, age and gender were strongly and positively associated with poor self-rated health; older people were more likely to report being in poor health compared to the young, and women were more likely to report poor health than men. There is a different gender effect at different ages, with the gender gap larger for older people, but this interaction is not however statistically significant. A significant inverse gradient in the relationship between income quintile and self-rated poor health was found, in which the base category received an odds ratio (OR) of 1.0. People in the lower income quintiles had higher odds of poor health with the poorest having an OR of 1.71 while that for the second quintile was 1.26. In contrast, the higher household income groups (quintiles 4 and 5) had ORs of 0.82 and 0.68, respectively. Model 3 shows that country average level of trust was significantly and negatively associated with self-rated poor health (Table 2, Model 3). Fig. 2 shows a plot of the relation between country trust and the odds of self-reported poor health. In countries with higher average levels of trust, individuals were less likely to report poor health after controlling for their demographic and socioeconomic characteristics. The inclusion of this variable also accounts for some of the between-country variance, which falls from 0.538 to 0.442, but this remains statistically significant. Again, a smaller DIC than in the previous model (a relatively slight reduction of 6.80) shows that including country-level trust on a wave basis leads to an improved model with better fit. 4.2. Contextual effect In Model 4 the main contextual effect of country trust remains largely the same after taking account of individual differences in trust perception, suggesting that the aggregate community trust effect is not just an artefact of aggregating individual perceptions of social trust; there is a contextual as well as an individual effect. However, when explored further, a significant cross-level interaction effect between country trust and individual social trust was observed in model 4B. Fig. 3 plots the predicted relationship Fig. 2. The estimated relationship between countries’ aggregated social trust and relative odds of reporting poor health based on Model 3. Fig. 3. The estimated relationship between countries’ aggregated social trust and relative odds of reporting poor health with cross-level interactions between individual trust and place trust based on Model 4. between country trust and the odds of reporting poor health, for low (with responses being ‘‘you can’t be too careful’’) and hightrust (with responses being ‘‘people can be trusted’’) individuals, based on results from model 4B. It indicates a ‘consensual effect’: the individual (compositional effect) and contextual trust (contextual effect) variables are both important with signs in 1027 M.H. Jen et al. / Health & Place 16 (2010) 1022–1029 the same direction: if you have high individual trust and live in a country with a high average level of trust you tend to feel healthier. Further, compared to Model 3, Models 4A and 4B are much better fits according to the reductions in the value of DIC of 763.04 and 5.65, respectively. Cross-level interactions were also fitted, but were not found to be substantial. At the world scale low trust at both the individual and country level are mutually reinforcing in relation to poor health. The effects of societal trust are moderate, with a twofold difference in poor health between the extremes of trust. But they give some support to the ‘social capital hypothesis’ that individual and societal cohesion are related to individual health. 5. Discussion The main finding from this cross-national analysis concerns the association between the level of trust in a country and its citizens’ health. The country’s social fabric predicts an additional proportion of individuals reporting poor self-rated health over and beyond what is predicted by their personal characteristics alone. This supports one important aspect of Wilkinson’s hypothesis. Another crucial element in his argument, however, is that it applies to ‘‘western societies’’ only—he did not extend it to other parts of the world. Our sample includes a culturally diverse set of countries ranging from poor developing countries to advanced societies with a high degree of material welfare, so we have extended his work to a much wider range of socioeconomic, cultural and political milieus; nevertheless the findings are supportive of, and thus extend, his argument relating health status to social trust. This may indicate that the determinants for a country’s level of trust may at least partly lie elsewhere than in the distribution of income, as also suggested by Wilkinson. This study extends consideration of the Wilkinson relative income hypothesis into the realms of social capital and social cohesion and their effect on health. There is considerable recent interest in the links between growing income inequality, falling social cohesion, increasing psychosocial stress and worsening health. Health inequalities therefore are seen as a result of relative income inequalities undermining social cohesion and exacerbating psychosocial processes of stress (see the summary pathway in Fig. 4). Our analysis does not directly assess the relative contribution of psychosocial versus material mechanisms. Elsewhere, we found no support for the income inequality hypothesis (Jen et al., 2009a, 2009b); GDP had little effect on health status in either rich or poor countries and income inequality did not have Wilkinson’s postulated effect on morbidity once individual’s income and its differential impact were taken into account on self-rated health. Even when interactions were allowed between individual income and inequality, poor people in the most unequal counties do not appear to experience the worsened poor health predicted by Wilkinson. Given that absence of any relationship with regard to material inequalities (as measured by individual income and national GDP) it may be the case that psychosocial mechanisms are relatively more important in some advanced societies, but that material conditions still constitute the dominant explanation in others. However, material resources will in principle always have some psychosocial meaning attached to it (Kawachi et al., 2002) and the underlying mechanism may very well be the same; because humans are social animals their health is best protected when they cooperate (Dorling et al., 2007). The mechanisms generating the patterns identified between both individual and aggregate trust and self-rated health are complex. The association between individual trust and health is perhaps easiest to grasp and probably involves pathways whereby trust reduces stress which in turn is health conducive. Additionally, trust may promote involvement in social networks which themselves improve health. The association between aggregate trust and health is more compounded. Part of the association may stem from the simple fact that ‘living in a trustful society may make a difference to your happiness’. In addition, and related, living in trustful societies reduces both anxiety and fear about the behaviour of others and this mechanism is essentially similar to the reduced-stress-mechanism at the individual level where the harmful physiological effects of chronic stress is reduced or very limited. However, the disentanglement of aggregate trust from other important features of societies remains to be investigated. For example, poorly functioning democratic institutions (Bobak et al., 2007) and material deprivation accompanied with a lack of social cohesion may all be contributing to the adverse public health situation in the former Soviet republics. The favourable situation for the Scandinavian countries, on the other hand, suggests that welfare state characteristics may be crucial for both the creation of trust and their population health. We interpret this finding as supportive of the institutional social capital perspective. The interaction between individual and contextual trust in predicting poor self-rated health did not show the complex pattern reported by Subramanian et al. (2002), who found that low trusting individuals had a steadily increasing odds of reporting poor health by increasing level of aggregate trust but the opposite association for high trusting individuals. Our study, however, has shown that both high and low trusting individuals Mind the Gap (2001) and Unhealthy Societies (1996). Material Circumstances Income inequality Health Inequality Dominance Hierarchy Gaps in Social Status Social Cohesion Social Capital Societal Structures Anxiety Stress Violence Life Expectancy Psycho-social Outcomes Pathways Fig. 4. Underlying psychosocial model derived from a reading of Wilkinson’s Mind the Gap (2001) and Unhealthy Societies (1996). 1028 M.H. Jen et al. / Health & Place 16 (2010) 1022–1029 had decreasing odds of poor health with increasing country level of trust. These conflicting findings may be due to context-specific conditions – i.e. they are dependent on both time and place (culture) – although the findings by Mansyur et al. (2008) are in line with ours. Individual level associations were all consistent with much previous research with respect to age, sex, marital status, trust and income. The graded association (Wilkinson and Marmot, 2003) between health and income has once again been demonstrated, in a global context – incremental wealth gives better health. There is a strong, non-linear relationship between individual income and self-rated health. Inclusion of income variables in Model 2 gave a very substantial reduction in model fit (DIC), showing an important link between income and self-reported health. With repeated cross-sectional data, however, we cannot rule out the reciprocal path that poor health leads to poor income rather than vice versa. Also there are likely to be interaction effect between income and trust – people with poor health may have poorer social capital, with further consequences for their health. However, such complex relationships cannot be isolated with cross-sectional data. 6. Limitations Because the only data available to test Wilkinson’s hypothesis is in a cross-sectional design (no cross-national longitudinal data sets are available), the direction of causation cannot be firmly established. In addition, the aggregation of individual trust to make statements about higher level units might be subject to the atomistic fallacy, i.e. drawing inferences at the group level based on individual level data (Diez Roux, 1998; Tunstall et al., 2004). The possibility of an ecological fallacy is, however, avoided by our statistical framework. Our dependent variable – self-rated health – may be interpreted differently in a range of cultural contexts, but its predictive value in relation to subsequent mortality is well known, as recently confirmed in a systematic review (DeSalvo et al., 2006). We are therefore confident that the international variations demonstrated here are valid and robust, especially as by dichotomizing self-rated health into excellent-good/fair-poor we are eliminating more nuanced differences that might occur in the five-value scale. 7. Conclusions We have used a Durkheim perspective to analyse how, what he termed, social facts influence health at the individual level. The varying national average levels of social trust across a wide range of countries successfully predict health status, even when various individual characteristics – including social trust – are held constant. Our findings have potential policy relevance in terms of creating healthier societies. By establishing universal social programmes, governments lay the foundations for the creation of solidarity and trust and also a good public health. References Bobak, M., Murphy, M., Rose, R., Marmot, M., 2007. Societal characteristics and health in the former communist countries of Central and Eastern Europe and the former Soviet Union: a multilevel analysis. Journal of Epidemiology & Community Health 61, 990–996. Barford, A., Dorling, D., Pickett, K., 2010. Re-evaluation self-evaluation. A commentary on Jen, Jones, and Johnston. Social Science & Medicine 70, 496–497. DeSalvo, K.B., Bloser, N., Reynolds, K., He, J., Muntner, P., 2006. Mortality prediction with a single general self-rated health question. A meta-analysis. Journal of General Internal Medicine 21, 267–275. Diez Roux, A.V., 1998. Bringing context back into epidemiology: variables and fallacies in multilevel analysis. American Journal of Public Health 88, 216–222. Diez Roux, A.V., 2004. The study of group-level factors in epidemiology: rethinking variables, study designs, and analytical approaches. Epidemiology Review 26, 104–111. Diez-Roux, A.V., 1998a. Bringing context back into epidemiology: variables and fallacies in multilevel analysis. American Journal of Public Health 88, 216–222. Diez-Roux, A.V., 1998b. On genes, individuals, society, and epidemiology. American Journal of Epidemiology 148, 1027–1032. Dorling, D., Mitchell, R., Pearce, J., 2007. The global impact of income inequality on health by age: an observational study. British Medical Journal 335, 873. Durkheim, É, 1895–1982.The rules of sociological method. In: Lukes, S. (Ed.), (tr. Halls, W.D., 1938). Free Press, New York. Esping-Andersen, G., 1990. The Three Worlds of Welfare Capitalism. Polity Press, Cambridge. Esping-Andersen, G., 1999. Social Foundations of Postindustrial Economies. Oxford University Press, Oxford. Gilks, W.R., Richardson, S., Spiegehalter, D.J., 1996. Markov Chain Monte Carlo in Practice. Interdisciplinary Statistics. Chapman and Hall, London. Goldstein, H., Rashbash, J., 1996. Improved approxiamtions for multilevel models with binary responses. Journal of the Royal Statistical Society A 159, 505–513. Idler, E.L., Benyamini, Y., 1997. Self-rated health and mortality: a review of twentyseven community studies. Journal of Health Social Behaviours 38, 21–37. Inglehart, R., 1997. Modernization and Postmodernization: Cultural, Economic, and Political Change in 43 Societies. Princeton University Press, Princeton, N.J. Inglehart, R., 2004. Human Beliefs and Values: A Cross-Cultural Sourcbook Based on the 1999–2002 Values Surveys. Siglo Veintiuno Editores, México, D.F. Jen, M.-H., Jones, K., Johnston, R., 2009a. Global variations in health: evaluating Wilkinson’s income inequality hypothesis using the World Values Survey. Social Science and Medicine 68, 643–653. Jen, M.-H., Jones, K., Johnston, R., 2009b. Compositional and contextual approaches to the study of health behaviour and outcomes: using multi-level modelling to evaluate Wilkinson’s income inequality hypothesis. Health and Place 15, 198–203. Johnston, R., Jen, M.-H., Jones, K., 2010. On inequality and health, again: a response to Bernburg, and Barford, Dorling and Pickett. Social Science and Medicine 70, 498–500. Kawachi, I., Kennedy, B.P., Lochner, K., Prothrow-Stith, D., 1997. Social capital, income inequality, and mortality. American Journal of Public Health 87, 1491–1498. Kawachi, I., Subramanian, S.V., Almeida-Filho, N., 2002. A glossary for health inequalities. Journal of Epidemiology & Community Health 56, 647–652. Kennedy, B.P., Kawachi, I., Glass, R., Prothrow-Stith, D., 1998. Income distribution, socioeconomic status, and self rated health in the United States: multilevel analysis. British Medical Journal 317, 917–921. Kennelly, B., O’Shea, E., Garvey, E., 2003. Social capital, life expectancy and mortality: a cross-national examination. Social Science & Medicine 56, 2367–2377. Kumlin, S., Rothstein, B., 2005. Making and breaking social capital—the impact of welfare-state institutions. Comparative Political Studies 38, 339–365. Lindstrom, C., Lindstrom, M., 2006. ‘‘Social capital,’’ GNP per capita, relative income, and health: an ecological study of 23 countries. International Journal of Health Service 36, 679–696. Lynch, J.W., Davy-Smith, G., Hillemeier, M., Shaw, M., Raghunathan, T., Kaplan, G., 2001. Income inequality, the psychosocial environment, and health: comparisons of wealthy nations. The Lancet 358, 194–200. Lynch, J.W., Davy-Smith, G., Kaplan, G.A., House, J.S., 2000. Income inequality and mortality: importance to health of individual income, psychosocial environment, or material conditions. British Medical Journal 320, 1200–1204. Mansyur, C., Amick, B.C., Harrist, R.B., Franzini, L., 2008. Social capital, income inequality, and self-rated health in 45 countries. Social Science & Medicine 66, 43–56. Marmot, M., 2001. Income inequality, social environment, and inequalities in health. Journal of Policy Analysis & Management 20, 156–159. Marmot, M.G., Shipley, M.J., 1996. Do socioeconomic differences in mortality persist after retirement? 25 year follow up of civil servants from the first Whitehall study. British Medical Journal 313, 1177–1180. Marmot, M.G., Smith, G.D., Stansfeld, S., Patel, C., North, F., Head, J., et al., 1991. Health inequalities among British civil servants: the Whitehall II study. The Lancet 337, 1387–1393. Navarro, V., Shi, L., 2001. The political context of social inequalities and health. Social Science & Medicine 52, 481–491. Navarro, V., Muntaner, C., Borrell, C., Benach, J., Quiroga, A., Rodriguez-Sanz, M., Vergés, N., Pasarı́n, M.I., 2006. Politics and health outcomes. The Lancet 368, 1033–1037. Olsen, K.M., Dahl, S.A., 2007. Health differences between European countries. Social Science & Medicine 64, 1665–1678. Poortinga, W., 2006. Social capital: an individual or collective resource for health? Social Science & Medicine 62, 292–302. Putnam, R.D., 1995. Bowling alone: Americas declining social capital. Journal of Democracy 6, 65–78. Putnam, R.D., Leonardi, R., Nanetti, R.Y., 1993. Making Democracy Work: Civic Traditions in Modern Italy. Princeton University Press, Princeton, N.J. M.H. Jen et al. / Health & Place 16 (2010) 1022–1029 Ram, R., 2006. Further examination of the cross-country association between income inequality and population health. Social Science & Medicine 62, 779–791. Rasbash, J., Steele, F., Browne, W., Prosser, B., 2004. A user’s guide to MLwiN. Version 2.10. Centre for Multilevel Modeling, University of Bristol, UK. Rose, G., 1985. Sick individuals and sick populations. International Journal of Epidemiology 14, 32–38. Rostila, M., 2007. Social capital and health in European welfare regimes: a multilevel approach. Journal of European Social Policy 17, 223–239. Rothstein, B., Stolle, D., 2003. Introduction: social capital in Scandinavia. Scandinavian Political Studies 26, 1–26. Rothstein, B., Uslaner, E.M., 2005. All for all. Equality, corruption, and social trust. World Politics 58, 41–72. Spiegelhalter, D.J., Best, N.G., Garlin, B.P., van der Linde, A., 2002. Bayesian measures of model complexity and fit (with discussion and rejoinder). Journal of the Royal Statistical Society B 64, 583–639. 1029 Subramanian, S.V., Kawachi, I., Kennedy, B.P., 2001. Does the state you live in make a difference? Multilevel analysis of self-rated health in the US. Social Science & Medicine 53, 9–19. Subramanian, S.V., Kim, D.J., Kawachi, I., 2002. Social trust and self-rated health in US communities: a multilevel analysis. Journal of Urban Health 79, S21–S34. Tunstall, H.V., Shaw, M., Dorling, D., 2004. Places and health. Journal of Epidemiology and Community Health 58 (1), 6–10. Wilkinson, R.G., 1996. Unhealthy Societies: The Afflictions of Inequality. Routledge, London. Wilkinson, R.G., 2001. Mind the Gap: Hierarchies, Health and human evolution. Yale University Press. Wilkinson, R.G., Marmot, M.G., 2003. The Solid Facts. World Health Organization Regional Office for Europe, Copenhagen. Wilkinson, R.G., Pickett, K., 2009. The Spirit Level: Why More Equal Societies Almost Always do Better. Allen Lane, London.