The study of health determinants that operate at the national level is probably most developed in the research on income inequality and its effect on health. There have been many studies assessing correlations of income inequality across countries (or states within countries) with national (or state) levels of health. Doubts have remained about the mechanisms behind this association,1^–4 and the strength of it. Lynch and co-authors 4 recently conducted a thorough review of the topic and concluded that there is only little support for the view that income inequality is a major determinant of population health in rich countries. However, they stated that income inequality may have a direct influence on specific health outcomes such as homicide.5 On the other hand, Wilkinson and Pickett6 concluded that the large majority of a total of 155 reviewed papers were supportive of the hypothesis that higher income inequality is associated with poorer health. Clearly, the last word on the subject has not yet been written. Studies should now try to contribute to this issue by testing hypotheses about which specific health outcomes might be sensitive to income inequality and for what reasons.

An explanation that has been forwarded for the association is that societies with higher income inequality are characterised by a breakdown of social capital.7 8 One influential definition of social capital is that of Robert Putnam9: “features of social organisation such as networks, norms, and social trust that facilitate coordination and cooperation for mutual benefit”. There are several elements that constitute social capital, such as community networks, civic engagement, local civic identity, reciprocity and trust in the community.10 It has been suggested that these elements mediate the association of income inequality with health.7 8

In theory we would expect that if the effect of income inequality is mediated by (a breakdown of) social capital some of the most pronounced effects would be observed on suicides and indicators of intrapersonal violence. A strong lack of social capital will foster hostility towards others and will reduce the ability of people to benefit from a social safety net that may prevent them from committing suicide. There is already evidence that income inequality is correlated with violence,11 including homicide.6 12 However, if we would expect that income inequality has an especially pronounced effect on these specific causes of death (i.e., violence-related and self-inflicted injury mortality) we would not only expect countries with a larger income inequality to have higher absolute rates of these causes; we would expect that the share of these two causes in overall mortality within these countries would be larger.

The aim of this study was twofold. First, the aim was to investigate whether developed societies characterised by greater income inequality showed larger self-inflicted (suicide) and violence-related mortality. We expected that (1) income inequality would be associated with mortality rates from both causes of death, and (2) that these two causes of death would explain a larger percentage of overall mortality rates in countries with larger income inequality. Second, our aim was to assess the relationship between social capital and these two causes of death. We expected that indicators of social capital would be related to both causes of death and to income inequality, suggesting that a mediating role of social capital may exist.

We conducted our study on data from 35 developed societies, including 16 former communist countries (all European), and 19 non-former communist countries (for the sake of pragmatism, these will be termed “other developed countries” from here on). The former communist countries share a long history of social hierarchy shaped by communism, which sets them apart from other developed countries. Because these countries are characterised by complex changes in the social fabric following the political transition during the 1990s,13 it may be important to stratify analyses on income inequality, as these complex changes may modify the effect of income inequality on mortality. Our expectation was that associations of income inequality and social capital with the outcomes would be exacerbated by the political and social instability in former communist Europe, and that these would therefore be stronger in the former communist countries than in the other developed countries.

METHODS

Data

We included all countries for which there was information available on (1) standardised mortality rates for 2002 through the World Health Information database,14 15 (2) income inequality from the United Nations Development Report16 and (3) social values from printed summary measures of the European or World Values Survey for 2000.17 This amounted to a total of 35 countries (see online table). Of these 35 countries, 30 were European.

Data on income inequalities were obtained through the United Nations Development Report. Income inequality was expressed as the ratio of the share of total income of the richest 20% in the population to the share of the poorest 20%. All information on income inequality covered years at the end of the 1990s or the beginning of the 2000s. Data on gross domestic product (GDP) per capita by purchasing power parities were obtained from the International Monetary Fund World Economic Outlook Database.18

We used data from the European/World Values Survey from around the year 2000.17 The aim of the European and World Values Surveys is to map the basic values and attitudes of people across Europe and the globe. World Values Survey data have been used before by Lynch et al19 in a study on income inequality and health in a different set of countries. The survey data include several variables that cover elements of social capital. First, respondents were asked if they believed that, generally speaking, most people can be trusted or that you need to be very careful when dealing with people. For each country, the percentage of respondents indicating that people can be generally trusted was included in our study as a measure of civic trust. Second, respondents were asked if they belonged to, or performed voluntary work for, organisations such as social welfare services for older, handicapped or deprived people, labour unions, youth work and peace movements. Based on this information we constructed two variables. The degree of “belonging” was constructed by computing the mean of the percentage of respondents indicating that they belonged to each specific organisation. “Voluntary work” was computed by taking the mean percentage of respondents indicating that they performed voluntary work for each of the organisations. Third, respondents were asked to indicate the confidence they had in several institutions, including labour unions, the police, parliament and civil services, on a four-point scale. The variable for the degree of confidence was constructed by taking the mean percentage of respondents who indicated that they had quite a lot or a great deal of confidence in each of the specific institutions. Finally, respondents were asked to what extent they felt concern for the living conditions of specific groups in society, such as fellow countrymen, older people and the unemployed, and to what extent they were prepared to help those groups, on a five-point scale. “Concern” was computed by taking the mean percentage of respondents that stated that they felt (very) much concern for each of the specific groups. “Helping” was computed by taking the mean percentage of respondents indicating (absolutely) yes for each of the groups. Information on the degree of concern that people feel and their preparedness to help others out was available for 26 of the 35 countries (not for Canada, Japan, Singapore, South Korea, USA, Hungary, Bosnia/Herzegovina, Albania and Belarus). Analyses including these variables were conducted on the sample of these 26 countries. Information on the degree of confidence in specific institutions was not available for Singapore.

Together, the above-mentioned variables tap into several of the elements that are considered to be part of the concept of social capital20 and provide a unique opportunity to investigate the role of social capital in population health. Trust in other people and confidence in specific institutions can both be linked to Putnam’s trust in the community. Belonging to organisations can be linked to community networks, whereas doing voluntary work for organisations and feeling concerned about the living conditions of specific groups in society can be linked to civic engagement. Finally, being prepared to help others can be linked to reciprocity.

Standardised mortality rates for self-inflicted and violence-related injury mortality were obtained from the World Health Organization, Department of Measurement and Health Information database for 2002.14 In addition to estimating the association of income inequality with self-inflicted and violence-related mortality rates, our aim was to investigate whether income inequality had a particularly strong effect on these mortality outcomes, relative to other causes of death. Therefore, we computed the percentage of total mortality rates that could be contributed to self-inflicted and violence-related mortality. In other words, we investigated whether countries with a larger income inequality also had a larger percentage of their deaths due to self-inflicted and violence-related injuries, and not just higher mortality rates for these causes. These percentages are listed in the online table.

Analyses

Because the distribution of the data departed from the standard normal distribution the non-parametrical Spearman’s rank correlation was calculated to measure the associations of income inequality and social capital measures with mortality rates and the percentage of overall mortality due to self-inflicted and due to violence-related mortality. We also checked the association of GDP per capita with the outcome measures. In cases where there was a moderate to strong association of GDP per capita with an outcome (Spearman’s ρ⩾0.40) we performed additional regression analyses to assess the association of income inequality with the outcome adjusted for GDP per capita. These analyses demonstrate whether or not a strong association of income inequality with the outcome remains, after taking into account the absolute national level of income.

Correlations were calculated for all countries, as well as for other developed and former communist countries separately. Subsequently, we conducted regression analyses with mortality rates and the percentages of total mortality due to self-inflicted and violence-related deaths as outcomes, when income inequality was at least moderately associated with the outcome (Spearman’s ρ⩾0.40). In these analyses, we included all social capital variables that were also at least moderately correlated with income inequality and the outcomes as covariates.

The procedure of relying on the estimation of the strength of the association was chosen rather than relying on significance (p values <0.05) because of potential power problems due to small numbers of observations.

RESULTS

Spearman rank correlations of income inequality with self-inflicted and violence mortality rates are shown in fig 1A–F, and in table 1. For all countries combined, the correlation coefficients indicated that income inequality was less than moderately associated with mortality rates. However, the lack of associations overall appeared to be due to contrasting associations of income inequality with self-inflicted and violence-related mortality in former communist Europe, and other developed countries. In other developed countries income inequality was strongly related to lower self-inflicted mortality (−0.70; p value <0.05), and was not related to violence mortality (0.09; p value 0.72). In former communist countries income inequality was related to higher self-inflicted mortality (0.46; p value 0.07), and higher violence mortality (0.62; p value <0.05). The pattern was also observed for the percentage of total mortality that was due to the two causes of death: that is, no association for all countries combined, but contrasting associations between former communist and non-communist countries. There was a moderate association of income inequality with total mortality in the former communist countries only.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 (A) Income inequality by self-inflicted injury mortality in all countries combined. Spearman correlation of self-inflicted injury mortality rate per 100 000 person-years and the ratio of the share of total income of the richest 20% in the population to the share of the poorest 20% (income inequality). (B) Income inequality by violence-related injury mortality in all countries combined. Spearman correlation of violence-related mortality rate per 100 000 person-years and the ratio of the share of total income of the richest 20% in the population to the share of the poorest 20% (income inequality). (C) Income inequality by self-inflicted injury mortality in other developed countries. Spearman correlation of self-inflicted injury mortality rate per 100 000 person-years and the ratio of the share of total income of the richest 20% in the population to the share of the poorest 20% (income inequality). (D) Income inequality by violence-related injury mortality in other developed countries. Spearman correlation of violence-related mortality rate per 100 000 person-years and the ratio of the share of total income of the richest 20% in the population to the share of the poorest 20% (income inequality). (E) Income inequality by self-inflicted injury mortality in former communist Europe. Spearman correlation of self-inflicted injury mortality rate per 100 000 person-years and the ratio of the share of total income of the richest 20% in the population to the share of the poorest 20% (income inequality). (F) Income inequality by violence-related injury mortality in former communist Europe. Spearman correlation of violence-related mortality rate per 100 000 person-years and the ratio of the share of total income of the richest 20% in the population to the share of the poorest 20% (income inequality).

Several moderate-sized correlation coefficients were observed for social capital measures with the outcomes, but these correlations were not in all cases in the expected direction. In other developed countries, increased concern was associated with lower self-inflicted injury mortality and percentage of mortality due to self-inflicted injuries, whereas increased belonging and voluntary work was associated with higher violence-related mortality and percentage of total mortality due to violence. In former communist Europe, trust appeared to be related to higher self-inflicted injuries, violence-related mortality and total mortality. Helping and voluntary work on the other hand were related to lower mortality in these countries.

One indicator of social capital was significantly correlated with income inequality in all countries combined: the level of preparedness to help others (table 2). The higher the income inequality, the lower the level of preparedness to help others (−0.45; p value <0.05). In other developed countries income inequality appeared to be correlated with levels of trust (negatively) and belonging (positively). In former communist countries, moderate correlations were observed for belonging, voluntary work and helping with income inequality.

We performed regression analyses only in cases where there was an indication that associations of income inequality with an outcome should be adjusted for GDP per capita and in cases where there was an indication that both income inequality and one or more indicators of social capital were at least moderately associated with an outcome. This led to the formulation of the following regression models: for other developed countries, (1) income inequality and concern in relation to self-inflicted mortality rates, and (2) income inequality and concern in relation to the percentage of total mortality due to self-inflicted injuries; for former communist Europe, (1) income inequality, voluntary work and helping in relation to the self-inflicted mortality rate, (2) income inequality, GDP per capita, trust, voluntary work and helping in relation to violence mortality rates and (3) income inequality, voluntary work, and helping in relation to the percentage of total mortality due to violence. The results of the regression analyses are presented in table 3.

Increased concern did not substantially reduce the association of higher income inequality with lower self-inflicted mortality and the percentage of mortality due to self-inflicted injuries in other developed countries. Thus, the results of the regression analysis were not in line with a role of social capital as a mediator of the association in this group of countries. But the social capital factor of helping may be a mediator of the association of income inequality with self-inflicted injury mortality and the percentage of total mortality that is due to violence-related mortality in former communist countries. The association of income inequality with these outcomes was reduced after including voluntary work and helping, but only helping remained a significant predictor of the outcomes, meaning that voluntary work did not have an independent association with the outcomes (and hence cannot be a mediator).

DISCUSSION

The results of this study demonstrated striking differences in the associations of income inequality with (the percentage of) mortality due to violence-related and self-inflicted injury mortality between former communist and other developed countries. Across other developed countries higher income inequality was associated with lower self-inflicted injury mortality and a smaller percentage of deaths related to self-inflicted injuries, whereas in former communist Europe a higher income inequality was associated with a higher self-inflicted injury mortality rate, a higher violence-related mortality rate and a higher percentage of mortality due to violence. This means that only in former communist countries the relationship was in the direction that we had initially expected. Some indicators of social capital appeared to be moderately to strongly associated with either income inequality or with one of the outcomes. However, in a few cases they appeared to be related to both income inequality and one of the outcomes. Our results were in line with a potential mediating role of social capital on the association of income inequality with self-inflicted injury mortality and with the percentage of total mortality due to violence-related mortality in former communist Europe only.

Some limitations of the study should be addressed. First, our analyses were based on data for one point in time. For our particular study it means that social capital and mortality due to self-inflicted injuries and violence might both be outcomes of underlying processes of social and economic change, rather than social capital being a mediator of the association of income inequality with the mortality outcomes. However, it should be kept in mind that the results of our analyses correspond with the results of previous studies on the relation with violence and violent deaths,21 22 including some with more solid bases for causal inference.23 24 It remains unsure whether the time-lag in observations of income inequality and mortality corresponds with the time-lag of effect of the supposed exposure (income inequality) and outcome. This remains a problem in cross-sectional ecological analyses such as these, and it calls for trend studies that have access to data from multiple points in time on both the exposure and the outcomes.

Second, analyses of associations of income inequality with health are always based on a selection of countries, which can be a primary weakness of such analyses.19 The magnitude of associations is highly dependent on the final selection of societies. In fact, this principle is demonstrated by our own results in the variations of associations between the two broad groups of countries. The countries included in this study were mostly European, with only five countries from other parts of the world. When richer data on social capital and/or mortality from more countries become available it would be good to study associations of income inequality with these specific causes of death, and the potential mediating role of social capital again, in a broader range of countries.

Third, the number of observations in analyses such as these is small, which may lead to power problems. Rather than relying on p values for statistical significance of the results, we have paid more attention to the estimated size of the effect, so as not to miss potentially important associations that are of moderate size and do not reach statistical significance just because of the small sample size.

Finally, self-inflicted injury and violence-related mortality may be subject to bias due to miscoding of cause of death. Several studies have investigated this issue, with the aim of assessing the reliability of these cause of death classifications. For instance, there has been special interest in the reliability of suicide and violent death mortality in former communist countries following perestroika, and quantitative and qualitative analyses of classification processes have demonstrated that in the Baltic and Slavic former USSR these data were reliable.25^–27 But not all reports are positive, and especially suicide mortality coding and classification has been identified as a potential cause of artefactual cross-country variations.28 Studies often rely on investigating trends in suicide mortality, together with undetermined deaths, to take account of misclassification. Such analyses recently performed on data from countries in the European Union (including 15 countries that were included in our study) demonstrated that geographical and temporal variations in suicide mortality were unlikely to be explained by misclassification, although misclassification may explain some of the observed patterns in four of the 15 countries.29 Thus, while mortality statistics for these causes remain important tools for public health research, it should always be kept in mind that misclassification may lead to some biases in cross-country variation estimates.

A strong point of our study was that our dataset allowed for a distinction between the two sets of countries that are greatly heterogeneous with regard to social and political historical background. These two broad groups stand in stark contrast with regard to political stability, which theoretically legitimises performing analyses on these specific groups separately. In fact, it has been clearly demonstrated previously that adopting a geographically insensitive analysis of mortality patterns may hide divergent geographic patterns of correlations between suicide and homicide rates, and may lead to false conclusions about their relationship.30 Our results confirmed the necessity to stratify, as we found clearly contrasting patterns for former and non-former communist countries with regard to the relationship of income inequality with self-inflicted and violence-related mortality.

Our results for other developed countries departed from the general pattern that is reported in the literature with regard to the association of income inequality with homicide. While we observed a strong association in former communist Europe, in agreement with most of the literature on the issue, we observed no such association in the selection of other developed countries. This may be explained by the large selection of Western European countries and Japan in this group, with relatively low income inequality and violence-related mortality. Similar effects of such a selection of data from predominantly Western European countries on estimates of the association of income inequality and mental health have been observed previously.31^–33

The association of income inequality with the percentage of total mortality due to self-infliction was negative for Western Europe. This is because only a relatively small percentage of mortality in countries such as Spain, Portugal, Italy and Greece can be attributed to self-infliction, even though these countries range among those with high income inequality. Besides having relatively large income inequality, these countries have a traditionally strong religious orientation, which might prevent a lot of self-inflicted mortality by providing more tight-knit communities. This explanation is certainly not new, as Durkheim34 already commented on a protective effect of Catholicism on suicide rates in his landmark study on suicide as a sociological phenomenon. If this were the case we would expect relatively high social capital in these countries. However, inspection of the indicators of social capital in our data does not demonstrate relatively high social capital in these countries. On the other hand, there is the possibility that deaths due to self-inflicted injuries are more often coded as mortality from other causes in these countries, because of prevailing negative attitudes toward coding “suicide”.35 Even though this possibility cannot be excluded it should be noted that countries such as Spain and Italy also have relatively low prevalences of mental illness compared with other European countries according to the international comparison of the World Mental Health Surveys (Greece and Portugal were not included in that study),36 suggesting that these societies indeed exhibit a protective effect on mental ill health within the population. Moreover, of these countries only Spain has been identified as a country where misclassification of suicide mortality may explain part of the observed suicide mortality trends.29

In former communist Europe, there was a very strong association of income inequality with the percentage of mortality due to violence, in contrast to other developed countries, corresponding with what we had expected. We propose three potential explanations for this finding. Many of the former communist countries in Europe underwent enormous political and economic change during the last decades, with repercussions for population health.13 37^–41 The negative consequences of the social upheaval on levels of violence in these countries have been stressed already. It might be that income inequality, a lack of social capital and high violence-related mortality all result from political instability. From this point of view, one possible explanation for the association we found between income inequality and mortality due to violence is that this association is confounded by political instability. A second potential explanation is that income inequality itself lies on the pathway linking political instability to an excessive proportion of mortality accounted for by intrapersonal violence. A third explanation might be that political instability modifies the effect of income inequality, that is, more instability exacerbates the negative consequences of income inequality on the outcome. Our results are inconclusive with regard to these three possibilities largely because of the study’s limitations that have been discussed. Only through continuous monitoring of not only income inequality and health, but also other indicators of national economic, political and social well-being will we be able to point out whether or not the association of income inequality with health in these countries endures beyond periods of political upheaval.

To conclude, our expectation that income inequality would have a relatively strong association with the two causes of death indicating mortality due to self-infliction and violence was only partly confirmed, that is only for former communist countries and only for violence-related mortality. Also, our expectation that indicators of social capital might play a role in the association was only partly confirmed. Only in former communist countries did adjusting for the level of preparedness to help others reduce the association of income inequality with the percentage of overall mortality due to violence.