1 Introduction

Social scientist have long been interested in determining the role that objective material conditions play in shaping better, healthier or more successful societies (Hall and Lamont 2009; Pritchett and Summers 1996; Wilkinson and Pickett 2009a; Torssander and Erikson 2010). Among the range of material conditions that could impact the wellbeing of societies and of their inhabitants, income inequality has captured the most attention. The so-called “income inequality thesis”, which claims a causal relationship between the degree of income inequality in countries and the presence of social maladies, is appealing to many policy-makers, because it promises a tangible solution to a large variety of social problems ranging from physical and mental health to crime, low social cohesion, teenage births, etc. As Wilkinson and Picket phrased it: “if the United States was to reduce its income inequality to something like the average of the four most equal of the rich countries (Japan, Norway, Sweden and Finland),… rates of mental illness and obesity might… each be cut by almost two-thirds, teenage birth rates could be more than halved, prison population might be reduced by 75 per cent, and people could live longer while working the equivalent of 2 months less per year.” (2009b: 268).

While the list of social maladies allegedly associated with income inequality is extensive, most empirical work has focused on the relationship between income inequality and health, and decreasing inequalities in the distribution of income has been proposed as a key measure to improve population health (Kondo et al. 2009; Layte 2011; Marmot 2005; Wilkinson and Pickett 2006). The interest in the potential corrosive effect of income inequality on population health reflects, on the one hand, concerns of policy-makers with the persistent health inequalities that defy the general rise in the standard of living of the population (Mackenbach et al. 2008). On the other hand, it reflects an ethical view shared by organizations such as the World Health Organisation according to whom “social injustice is killing people on a grand scale” (CSDH 2008: 26). It is in this global context that the work of Rodgers (1979), who found a negative correlation between the level of income inequality of a country and population health, marked the starting point of a rich body of research which, at times, stirred intense debates between the supporters and the critics of the “income inequality thesis”. Not surprisingly so, since the straightforward solution to address health inequalities by ‘simply’ decreasing the income inequality of societies is compelling for policy-makers, given that it is supported by empirical evidences. As Kondo and his colleagues calculated, if indeed the relationship between inequality and population health is causal, “1.5 million deaths (9.6 % of total adult mortality in the 15–60 age group) could be averted in 30 OECD countries by levelling the Gini coefficient below the threshold value of 0.3” (Kondo et al. 2009: 7).

What supporters of the “income inequality thesis” maintain is that reducing disparities in the income distribution is the key to creating healthier societies, especially in the economically high-developed countries. While, as they argue, economic growth is relevant for the improvement of population health in countries that are less economically developed, reducing income inequality would be the thing to do in more developed countries, because there is a “threshold of material living standards after which the benefits of further economic growth are less substantial” (Wilkinson and Pickett 2009b: 10). When this supposed threshold of wealth is reached, it is not the level, but the distribution of income that would further improve population health. In other words, the effects of income inequality and wealth on health are dependent on the level of economic development of the country. However, although a central component of the debate, this idea is rarely empirically analysed adequately, because empirical studies of the relationship between income inequality and population health are mostly limited to samples of wealthy countries, while countries that are less economically developed received relatively less attention. Therefore, the first aim of the current study is to examine how population health is affected by a country’s wealth and income inequality and by the interplay between the two for countries with various levels of economic development.

Importantly also, meta-analyses and reviews of the accumulated studies show that the current empirical evidence does not necessarily fully support the core idea that greater equality would benefit health. For instance, in a meta-analysis of 168 studies linking income inequality and population health, the outcomes of 87 studies (52 %) were found wholly supportive of the idea that higher income inequality relates to worse population health while the outcomes of the rest were partially supportive or not-supportive (Wilkinson and Pickett 2006). Another meta-analysis evaluating only studies applying a multi-level design found a modest adverse effect of income inequality on population health, although the authors advise that these results need to be interpreted with caution given the heterogeneity between studies (Kondo et al. 2009). In opposition to this evidence that is to some extent supportive for the “income inequality thesis”, another extensive review of the literature ends with a more critical tone: among wealthy countries, income inequality is not systematically related to population health (Lynch et al. 2004). Judge et al. (1998: 578), based on a review of the literature and their own analyses, caution again: “statistically significant associations between income inequality and population health in the developed world are anything but secure”.

Apart from the lack of consensus among scholars whether the empirical evidence provides support for the “income inequality thesis” or not, we note that many of the studies included in these reviews are cross-sectional in nature. Furthermore, despite the limitations of the cross-sectional studies, authors postulated conclusions in terms of the benefits of decreasing the income inequality, for instance: “The evidence shows that reducing [italics added by author] inequality is the best way of improving the quality of the social environment” (Wilkinson and Pickett 2009b: 29). However, as Ellison states: “none of the cross-sectional ecological studies […] can actually establish that income inequality precedes the social and material circumstances which undermine health at an individual level. Longitudinal studies, multi-level modelling and path analyses should provide better evidence of causality…” (2002: 563). Clearly, in order to conclude that a reduction in inequality would improve health, we need to perform dynamic analyses, using longitudinal types of data, in which changes in inequality and changes in health are examined. The second aim of our study is thus to improve upon the testing of the “income inequality thesis” in this way, i.e., by using also longitudinal data.

Given the above, we formulate research questions along two lines, by looking at the levels of and at the changes in income inequality and wealth, and we ask to what extent they relate to population health and whether the strengths of these relationships are different for countries with various levels of economic development. Furthermore, from a methodological point of view, we inquire whether the use of dynamic versus static models leads to different conclusions. Our strategy is to conduct our analyses separately for subsample of countries defined by their level of economic development, based on the analytical categories employed by World Bank (2009). We chose this method to classify countries for two reasons. Firstly, this classification method was used in previous research in order to select the sample of countries used to provide evidences for the “income inequality thesis” (Wilkinson and Pickett 2006). Secondly, this classification is widely used by prominent international economic institutions, such as World Bank or United Nations. We make use of high-quality comparable data on income inequality across countries and time that has become available only recently (Solt 2009). We enrich this initial dataset with information on societal wealth measured as GDP per capita and population health measured as life expectancy at birth. Our working sample covers a number of 140 countries and 2360 country-year observations ranging from 1987 to 2008. These data enable us to both replicate previous studies and improve upon them.

2 Theory and Hypothesis

In a nutshell the “income inequality thesis” states that increasing societal wealth leads to improving population health only to a certain level of economic development. When this threshold of wealth is reached reducing disparities in income distribution is the key to further improve the health of the population (Wilkinson and Pickett 2009b). However, scholars disagree with regard to the “authenticity” of the observed association between higher levels of income inequality and worse population health indicators. The critics talk about a spurious relationship due to some unmeasured characteristics (Lynch and Smith 2000). Other scholars contributed to the issue by focusing on the essential question of what the potential mechanisms would be that might explain why greater inequalities would relate to worse health in samples of rich countries (Wagstaff and Van Doorslaer 2000). One of the main suggested explanations is that the relationship between income inequality and population health has to do with the concave relation between income and health at individual level (Gravelle 1998). The argument is the following: if a monetary transfer occurs from a rich individual to a poor one, at societal level one would observe a decrease in income inequality while the average income of the society remains constant. At individual level, the impact of the transfer on the health of the poor individuals will be significant, because it allows the acquisition of goods and services that positively influence health, while it will affect the health of the rich individuals only marginally. At aggregate level, we would observe that between countries with similar average income, the ones with lower income inequality would display better average health outcomes. In addition, within a country, reducing income inequality would relate to an improvement in population health.

Another prominent idea is that income inequality may serve as a measure of how stratified a society is (Wilkinson and Pickett 2009a, b). In this view, a person’s income is important not in its absolute value, but in its relative value compared to the other members of society. Material hierarchies lead to status differentiation that, in turn, triggers social comparisons. Such social comparisons are at the basis of psychosocial effects of income inequality: stress and anxiety (Wilkinson 1999). In turn, the stress associated with the prolonged negative social comparisons within an unequal society is a precondition for increased vulnerability to a wide range of health problems. Subsequently, in countries with more income inequality, generalized worse individual health would aggregate into worse societal health compared to countries with less income inequality. The same would apply in periods characterized by an increase in income inequality.

Thirdly, income inequality might also affect health through the erosion of communities and through violent crime. Regarding the corrosive effect on community life, the argument is that a gap between the poor and the rich has led to declining levels of social cohesion and trust, which in turn resulted in lower levels of social support and via this mechanism the health of individuals was negatively affected (Kawachi and Kennedy 1997). Regarding crime the argument is twofold. On the one hand, it was argued that the exacerbated feelings of shame and humiliation resulting from social status differences trigger the involvement in violent acts. On the other hand, people living in environments characterised by crime, anti-social behaviour and violence are argued to experience more stress which on the long term, affects their health (Wilkinson and Pickett 2009b).

We argue that these mechanisms could very well apply to countries in all categories of economic development. For instance, in countries at the lower end of the wealth continuum, large income inequalities may cause political and social systems to be very unstable, with continuous conflicts and tensions. This is likely to decrease the level of social cohesion, to increase levels of violence and to enhance stress and anxiety in the population. In addition, there are no reasons to suspect that the relationship between income and health at the individual level is not a general one, applying to countries in various levels of economic development. Therefore, the expectation is that the negative relationship between levels (H1a) or changes (H1b) in income inequality and life expectancy is applicable to countries in all categories of economic development.

While plausible to think that the negative relationship between income inequality and health is a general one, the strength of the relationship might be different between groups of countries with different levels of economic development. In the economic literature, it is sometimes argued that in societies with higher average income the public provision of essential goods and services is higher and the population has more command over these goods and services (Anand and Ravallion 1993; Ranis et al. 2000). In turn, the provision of public services, the improved access to better living conditions and the capacity to better access and use the public goods through, for instance, increased levels of education, have their own effect on improving population health (Elo 2009; Torssander and Erikson 2010). As a result, higher level of economic development would provide more protection against the damaging effects of income inequality. This reasoning leads to modifying the previous expectations in the following direction: the expected negative relationship between levels (H2a) or changes (H2b) in income inequality and life expectancy is weaker among well-developed countries.

An important element of the “income inequality thesis” is the robust finding that the level of wealth of a country adds more to population health when the country is on a lower level of development, while in the countries with higher level of economic development this effect diminishes or even disappears (Deaton 2003; Preston 1975; Pritchett and Summers 1996). The observation that “wealthier is healthier” can be explained by the fact that in wealthier countries or in countries that experience an increase in wealth, population has more resources and better living conditions and the state also has more resources that can be invested in the health services and infrastructure, which in turn add to the health of the population. We therefore hypothesize that there is a positive relationship between the levels (H3a) or changes (H3b) in wealth and life expectancy. On the other hand, the relationship between wealth and health should be stronger for countries in the low-income group, because the gains in population health deriving from economic growth should be more substantial than in the case of their richer counterparts. As a result, we expect that the positive relationship between levels (H4a) or changes (H4b) in wealth and life expectancy to be weaker as the level of economic development increase.

We also need to take into account that wealth and income inequality change simultaneous within a country from 1 year to another. If an increase in income inequality is related to a decrease in life expectancy, an increase in wealth might be protective for population health, thus, diminishing the corrosive effect of income inequality. Our last hypothesis reads: in periods when an increase in wealth is observed, the negative effect of changes in income inequality on life expectancy is weaker (H5).

3 Data and Methods

3.1 Income Inequality

The most important measures for our study are the level and changes in income inequality of countries. Previous research already pointed at the importance and problematic of comparability of observations across time and countries (Judge et al. 1998; Moran 2003; Leigh et al. 2009): “Many of the studies use multiple sources of income distribution data and/or data from a wide range of years, which makes comparability between countries questionable… In fact, we believe it is the generally poor quality of the income data that poses the most serious weakness in most of the studies we have reviewed.” (Judge et al. 1998: 569). To overcome these difficulties, we make use of a new dataset that was recently developed (Solt 2009) with the goal of increasing the coverage across country and time while also improving the comparability across observations: the Standardized World Income Inequality Database (SWIID). The starting point of the SWIID dataset is the Gini Index measure from the World Income Inequality Dataset (UNU-WIDER 2008). In the next step, this database was enriched with two measures of Gini Index derived from the Luxembourg Income Study—in gross and net income. Next, a procedure is developed to account for the fact that the data in the two original datasets differ with respect to several key elements: the reference unit of the source data (e.g., household per capita, household adult equivalent, household without adjustment, employee, person) and the definition of income (e.g., net income, gross income, expenditures or unidentified). A custom missing-data algorithm was used to generate time series standardized on the Luxembourg Income Study household adult-equivalent gross and net-income data, which is considered nowadays to be the source with the highest quality and comparability. For our study, we used the net-inequality series, which covers 153 countries with 3331 country-year observations.

3.2 Wealth and Level of Economic Development

We derived the level of economic development of countries for a specific point in time using the level of GNI per capita in a certain year and the different benchmarks provided by the historical dataset compiled by World Bank (2009). Based on the World Bank methodology, countries are classified in four analytical income categories: low-income, low middle-income, upper middle-income and high-income. Preliminary analyses showed similar patterns in the effects obtained in the low middle and upper middle-income category, and for reasons of parsimony, we collapsed these two categories into one.

The World Bank historical dataset provides information starting from 1987. For some countries and years, we have incomplete classifications. In order to deal with missing values, we replaced them with the closest valid value. Since the level of missing values is not alarmingly high and since the level of economic development is not expected to fluctuate suddenly from 1 year to another, we believe that this procedure is suited in order to keep these country-years observation in the analysis.

Note that due to the long period under investigation, 37 of the countries in our analyses undergo periods of transitions between categories, while 103 have the same level of economic development between 1987 and 2008. All of the 37 countries only go though one change: either from low to middle-income countries or from middle to high-income countries. However, within the time span under investigation, the number of years in each economic development category is different for every country. We opted to recode the level of economic development of the 37 countries, such that this measure is time invariant, by looking at the numbers of years within each economic category and choosing the one with most years. Four of the countries had equal numbers of years in low and in middle-income categories and we opted for categorizing them as low-income countries. The reason for this choice is the fact that we assume structural differences between the categories of economic development. While on paper these countries are officially over the threshold of wealth that places them in the middle-income category, due to institutional inertia they could still resemble the profile of others in the low income category.

In order to quantify the wealth and change in wealth of a country we used a measure of GDP per capita (PPP international $) derived from World Development Indicators (World Bank 2011). While the coverage of the country-years available from the SWIID dataset was quite good, for some country-years we had missing observations on the GDP per capita measure. These missing country-year observations were eliminated from analyses. Since an increase of 1 $ PPP in the GPD per capita is expected to have a very small effect on the level of life expectancy, in our analyses we used a rescaled measure by dividing the level of GDP per capita by 100.

3.3 Population Health

A general accepted convention of the research testing the “income inequality thesis” is that life expectancy at birth is an appropriate and widely used measure of population health (Rodgers 1979; Wilkinson and Pickett 2009b). We follow this line in order to provide results that are comparable with those of previous studies and we use as dependent variable figures of life expectancy at birth, both sexes combined, derived from the United Nations’ World Prospects (2009). We matched these figures with the information on income inequality and wealth. We excluded the country-year points where information on the dependent variables was not available.

3.4 Final Samples

Previous studies have suggested that tax havens need to be excluded from analyses on income inequality and population health (Wilkinson and Pickett 2009b). An argument for excluding them from the analyses is the fact that their level of measured wealth is not the result of a sustainable developed economic system and thus, these figures do not correspond to the social reality. We used the OECD (2009) classification of tax havensFootnote 1 and identified among our initial sample a number of five countries that we excluded from further analyses.

In addition, we excluded countries that were observed in only one time point. Our final working sample covers a number of 140 countries and 2360 country-year observations: 50 low developed countries with 685 country-year observations, 61 middle developed countries with 1084 country-year observations and 29 high-developed countries with 591 country-year observations. Descriptive information of the measures used is found in Table 1.

Table 1 Descriptive information on the dependent and independent variables
Full size table

3.5 Analytical Strategy

3.5.1 Static Estimation

For each sample of countries, we estimated the partial correlations between the level of income inequality and of life expectancy for each time-point, controlled for the level of GDP per capita. In the next step, we estimated the partial correlations between GDP per capita and life expectancy, controlled for the level of income inequality, in each available year in our sample. This approach was also used in some prominent ecological studies (Wilkinson and Pickett 2006), in which support was found for a detrimental effect of income inequality on health. In our analyses, we used 2-year lagged measures of income inequality and wealth, to allow for a temporal ordering between the alleged cause and its alleged effect. The static estimation addresses only hypotheses that regard relationships between levels in the dependent and independent variables (i.e., H1a, H2a, H3a and H4a).

3.5.2 Dynamic Estimation

In order to estimate whether changes in our independent variables are related to changes in the independent variable we used a technique similar to fixed effects regression (Allison 2009). The main advantage of this technique is that it controls for unobserved time-invariant variables, which are allowed to have whatever correlations with the observed ones. In essence, using fixed effects regression we test whether changes in income inequality and wealth are related to changes in life expectancy within countries. The disadvantage of this method is that it does not allow estimating time-invariant effects (e.g., the level of economic development.)

In our case, we would like to simultaneously estimate the effects of changes in income inequality and wealth, but also the effects of levels of income inequality and wealth. Allison (2009) proposed a solution to this problem in the form of a hybrid fixed effects method. The basic idea of this method is to decompose the time-varying predictors into two parts, one representing between-country variation, and the other representing within-country variation. In practice, this is done by calculating (1) the country means of the time-varying covariates across the time span investigated and (2) the deviations within countries from these country means. Both these variables are then used as predictors. The coefficients for the within-country components (i.e., the deviations from the country means) will be identical to those of conventional fixed effects regression. The hybrid model is estimated using random effects methods in order to obtain correct standard errors. This method allows simultaneously testing hypotheses regarding the relationships between levels and changes in the dependent and independent variables (i.e., all our formulated hypotheses).

We used the hybrid fixed effects method to estimate models for samples of countries in the three analytical income categories. All the models included effects for the years of measurement as dummy variables (effects not presented).

In order to rule out multicollinearity between the wealth and the income inequality variables we checked the correlations between the GDP per capita and the SWIID Gini Index. In the country-period file of low-income countries the correlation was −.21 (p < .01), for the subsample of middle-income countries the correlation was −.21 (p < .01) and for the high-developed countries the correlation was −.002 (p = 0.96). For the cross-sectional dataset, we also computed the correlations between the GDP per capita and the SWIID Gini Index for each year in every subsample, and these were also not alarmingly high: in the low-income sample they ranged from −.60 to −.38, in the middle-income sample they ranged from −.62 to −.35, and in the high-income sample they ranged from −.56 to −.42.

4 Results

We start with a few descriptive analyses. Table 2 summarizes information on the trends in our dependent and independent variables. Life expectancy at birth increased across the pooled sample from an average of 66.3 years in 1987 to an average of 77.24 years in 2008. The rates of increase varied between the three income categories: the stronger increase was observed in middle-income countries, followed by low–income and high-income countries.

Table 2 Averages of the three dependent variables for the three categories of economic development (start and end point of the time series)
Full size table

For the whole sample, from 1987 to 2008 we observed a decrease of around 7 percent in income inequality. However, this trend was mainly caused by a decrease of income inequality within the group of low-income and middle-income countries while in the group of high-income countries we observed an increase in income inequality. We also note that the average levels of income inequality in low and middle-income countries were higher than the average income inequality observed in the sample of high-income countries, both in 1987 and 2008.

On average, we observed a strong increase of 472 percent of GDP per capita between 1987 and 2008. However, the rates of increase differed between countries in various levels of economic development. The middle-income countries registered the strongest relative increase (in 2008 on average their wealth was 3.9 times higher than in 1987), followed by the high-income countries (2.83 times higher average wealth in 2008) and by low-income countries (only 1.89 times richer in 2006 compared to 1987).

4.1 Static Estimation

Table 3 summarizes the results of the partial correlations between income inequality and life expectancy, controlled for GDP per capita. For each of the three samples of countries we present the number of years when we found a significant partial correlation between income inequality and life expectancy controlled for GDP per capita and the average of the significant partial correlations found.

Table 3 Cross sectional correlations income inequality and life expectancy controlled for the level of GDP per capita
Full size table

Looking at the number of years when the partial correlation was found significant, we note that this correlation was not robust in time and that there were differences between the three income categories in the prevalence of such significant relationships. We found a significant partial correlation in more years in the low-income countries (16 out of 18 years) than in middle-income countries (9 out of 20 years) while in the high-income countries the correlation was not significant in all the 20 years observed. All the significant correlations were negative. Based on the “income inequality thesis” argument we expected to find a negative relationship between income inequality and life expectancy in every sample of countries and in every year. Based on the above, hypothesis 1a did not receive general support by our cross-sectional analyses.

Looking at the average of the significant correlations we note that it was the highest in the low-income sample, somewhat lower in the middle-income sample and was statistically indistinguishable from zero in the high-income sample. We expected the negative effect of income inequality on population health to be weaker in countries with higher level of economic development. Our cross-sectional analyses provided support for hypothesis 2a, however, one needs to keep in mind that the relationship per see was not found to be robust in time and was not significant in the high-income countries.

In Table 4 we summarize the results of the partial correlations between GDP per capita and population health indicators controlled for the level of income inequality. The structure of Table 4 is the same as for Table 3.

Table 4 Cross sectional correlations GDP per capita and life expectancy controlled for the level of income inequality
Full size table

In the low-income countries, the relationship was found significant in 17 out of 18 years, in middle-income countries in 14 out of 20 years and in high-income countries in none of the 20 years observed. All these significant correlations were positive. Furthermore, looking at the strength of the significant partial correlations we note that the stronger relationship between GDP per capita and life expectancy was found in low-income countries, followed by middle and high-income countries. This said our cross-sectional analyses provide some support for our expectations stated in hypotheses 3a and 4a, although the relationship was not robust in time.

4.1.1 Additional Static Estimations

In additional models, we used the full sample of countries and for each year we estimated two separate regressions with life expectancy as dependent variable, and as predictors GDP per capita, Gini Index, dummies for the level of economic development (low-income category as reference) and the interactions between the economic development dummies with GDP per capita and the Gini Index, respectively. Based on these regressions, we calculated the effect of the level of wealth and of income inequality on life expectancy for each year where data was available. Across the 20 years in the analyses, the average un-standardised effect of GDP per capital on life expectancy was for low-income countries .86, for middle-income countries it dropped to .05 while for the high-income countries it dropped further to .004. Regarding the effect of the Gini Index on life expectancy we observed a similar pattern: the average un-standardised effect was −.32 in the low-income countries, −.13 in the middle-income countries and −.07 in the high-income countries. These results supported the same conclusion as the previously presented static estimation results: the level of economic development seems to moderate the effect of the levels of wealth and of income inequality on life expectancy.

4.2 Longitudinal Analyses: Dynamic Estimation

Table 5 summarizes the results of the hybrid fixed effects model for the subsamples of countries based on their level of economic development. In Model 1 we estimated the effects of levels and changes in income inequality (e.g., the coefficient of dGini reflects the effect of changes in income inequality within a country while the coefficient of Mean Gini reflects the effect of the average levels of income inequality of a country). In Model 2, we introduced the levels and changes in wealth and tested their relationship with life expectancy, while in Model 3 we present the controlled effects of the income inequality and wealth. In Model 4, we added the interaction between changes in income inequality and changes in wealth, thus testing whether within a country the effect of changes in income inequality depends on the changes in the levels of wealth.

Table 5 Results of the hybrid fixed effects model
Full size table

We first look at the effect of income inequality. According to the “income inequality thesis”, we expected that levels and changes in income inequality to be negatively related to life expectancy and this relationship to be weaker as the level of economic development is higher, but still significant in the high-developed economies. Looking at the results presented in Model 1, we observed a different picture. In the categories of low and middle-income, countries with higher average levels of income inequality in the period 1987–2008 had lower life expectancy, as the thesis predicted. However, in the high-income category, the average income inequality was not significantly related to life expectancy. This said our results do not provide full support for hypothesis 1a.

In Model 1, we also tested whether changes in income inequality are related to changes in life expectancy. Our results showed that only in the low-income countries changes in income inequality were significantly related to changes in life expectancy, but in opposition to our expectations derived from the “income inequality thesis”, an increase in income inequality related to an increase in life expectancy. In the middle and high-income countries, we found negative effects, in line with the thesis, but they did not reach significance. Based on these findings, hypothesis 1b was not confirmed.

We argued that higher levels of economic development would “temper” the negative relationship between levels of income inequality and health. In Model 1, the effects of average levels of income inequality clearly showed a decreasing trend in their strength, with the strongest relationship being observed in low-income countries, followed by a weaker one in middle-income countries and a very weak and statistically not different from zero in high-income countries. The confidence intervals of the effects of the level of income inequality on life expectancy for the three samples (not in table) were not overlapping only for the high-income countries and low-income countries. Based on these findings we partially accept hypothesis 2a which stated that the level of economic development moderates the effects of the level of income inequality on life expectancy. Clear differences in the effects of income inequality on life expectancy due to this moderation were found, however, only between the low and high-income countries.

Looking at the differences in the effects of changes in income inequality on life expectancy between economic development categories, we also expected that the negative relationship between changes in income inequality and changes in life expectancy to be weaker with higher economic development. However, as previously noted, we did not observe a consistent negative and significant relationship between changes in income inequality and life expectancy in the three income categories. Thus, our data do not provide support for hypothesis 2b.

We now turn to the effect of changes and levels of wealth on life expectancy. In Model 2 dGDP per capita estimates the effect of changes in wealth within countries while Mean GDP per capita estimates the effect of levels of wealth between countries. In the categories of low and middle-income, countries with higher average levels of wealth across period 1987–2008 also had higher average levels of life expectancy. In addition, for the high-income countries we did not find a significant relationship between average wealth and average life expectancy, although the effect was in the expected direction. These results provide support for hypothesis 3a. The strength of the effects was also variable between the three economic development categories; the level of wealth had a stronger relationship with life expectancy in the low-income category, a weaker one in the middle-income category and was non-existent in the high-income category. The confidence intervals of the effects were not overlapping. Based on these findings we accept hypothesis 4a according to whom the level of economic development moderates the effect of the level of wealth on life expectancy.

Model 2 also showed that in general changes in wealth were positively related to changes in life expectancy; however, only for the low income countries this effect was significant. This result is consistent with our expectations that for richer countries further increases in wealth cease to play a significant role in the improvement of the health of the population. When inspecting the confidence intervals of the effects we found that they were overlapping only in the high-income category and in the middle-income. Subsequently, our longitudinal analyses provide partial support for hypothesis 3b and full support for hypothesis 4b.

Model 3 presents the estimated effects of income inequality and wealth controlled for each other, and looking at the estimates, previous conclusions remained unchanged. Note that the negative significant effects of the average levels of income inequality became weaker as compared to Model 1 (a decrease of 15 % for the low-income countries and 19 % for the middle-income countries). Furthermore, Model 3 improved the overall explained variance more compared to Model 1 (where we had only the uncontrolled effect of income inequality) than compared to Model 2 (where we had only the uncontrolled effect of wealth). For the low-income countries, the improvement in explained variance was the highest, while for the high-income countries it was very modest. These observations suggests that wealth has more weight in explaining variability in life expectancy than income inequality, and this is particularly valid for the low and middle-income countries.

In Model 4, we introduced the interaction between changes in income inequality and in GDP per capita, as a mean to test whether changes in income inequality have a different relationship with changes in life expectancy based on changes in wealth. For all the three samples, the interaction was negative; however, it did not reach significance. This said we do not find evidence that the effect of changes in income inequality on life expectancy within a country is dependent on the changes in wealth, and thus we reject hypothesis 5.

The static and the dynamic estimation were similar on several points. On the one hand, based on both types of analyses we did not find support for the expectation that the level of income inequality is negatively and significantly related to life expectancy in all categories of economic development or in all time points, as stated in hypothesis 1a. Most surprising, based on both types of analyses, we concluded that in the high-income countries the relationship is statistically not distinguishable from zero. Second, the moderation hypotheses received some support from both types of analyses, and the differences in the effects of income inequality and wealth between countries with different levels of economic development were more clearly seen in the comparison of low-income versus high-income countries. All in all, both static and dynamic estimations point to the fact that income inequality and wealth are both most important for life expectancy in the low-income countries, suggesting a different pattern of relationships than that implied by the “income inequality thesis”.

5 Discussion

In the present study, we set out to answer the question whether income inequality is related to population health throughout a long period of time (1987–2008) and using a global sample of countries. We looked at the interplay between income inequality and wealth of the countries, and the way in which these two material conditions affect health, as measured by life expectancy. We used high quality comparative data for 140 countries and 2360 country-year observations, including low- and middle-income countries. This extends previous research, which usually focuses on the high-income countries only. Moreover, we provided both static and dynamic tests of the arguments in the literature.

We derived the following scenario from the literature. On the one hand, levels and changes in wealth should be positively related to life expectancy, but the strength of the relationship should be weak or non-existent among high-income countries. On the other hand, levels and changes in income inequality would be negatively related to life expectancy, the relationship would be weaker with higher levels of economic development but it would remain significant in the group of high-income countries. Subsequently, this scenario accommodates the claim that in the high-income countries further economic development would not significantly add to the health of the population, but the key to further improve the health of these societies is to diminish the inequality in income (Wilkinson and Pickett 2009b). Our first conclusion, based both on the results of our cross-sectional and longitudinal estimations, is that this scenario generally does not hold.

Regarding the relation between wealth and life expectancy, our expectations received support from the data both when employing static or dynamic types of analyses: higher levels and positive changes in countries’ wealth were related to higher life expectancy, and this positive relationship was weaker for countries with higher level of economic development. Furthermore, as expected, the relationship between wealth and life expectancy was moderated by the level of economic development: it became non-significant for the high-income countries. These findings are in line with previous results (Deaton 2003; Preston 1975; Pritchett and Summers 1996) and it emphasizes the importance of economic growth for the health situation in poorer countries.

Regarding the relationship between levels of income inequality and life expectancy, we found that the expected negative relationship is not robust in time. This lack of robustness of the association is a fact acknowledged even by supporters of the “income inequality thesis”. For instance Wilkinson and Pickett (2006) note that the negative association between income inequality and population health was not found in studies using data from between the later 1980s and mid 1990s. Their explanations for these findings have to do with structural processes that evolved in parallel with the increase in income inequality, and which worked toward making population health better, rendering thus the relationship between income inequality and health non significant. However, critics of the “income inequality thesis” have argued that the observed relationships between income inequality and health is spurious precisely because the developments in income inequality strongly correlate to other structural characteristics such as community infrastructure, educational and health policies, transportation, etc. (Coburn 2004; Lynch and Smith 2000). These two positions raise two questions. First, to what extent can we consider the (in some years) observed relationship between income inequality and life expectancy as being a “true” effect of income inequality? Second, for the years where the relationship was not observed, could this be due to some parallel processes at work? Unfortunately, due to data limitations, we are unable to address these questions, but future research might want to examine more closely these alternatives.

However, our results suggest more than just a temporary disappearance of the relationship between income inequality and population health, seeing that for countries with high level of economic development the observed relationships between levels and changes in income inequality and life expectancy were not significant. These findings are in line with the results of Beckfield (2004) who, using a similar fixed effects design, also did not find a significant relationship between changes in income inequality and population health. In his opinion, one important reason for the discrepancy in his (and our) findings and supporting evidences presented in the literature is the quality of data and the data modelling strategy. “Using a larger sample, better (though still imperfect) income inequality data, and more statistical controls reduces support for the inequality-health hypothesis, but accounting for unmeasured heterogeneity with a fixed-effects approach eliminates support. This suggests that heterogeneity bias may be the most serious limitation of the “classic” cross-national work in this area.” (Beckfield 2004: 240).

Note that this methodological explanation could only apply to the results of our dynamic estimation, while our static, cross-sectional analyses also support the same conclusion. This is most remarkable since a more recent study that replicates the original study of Rodgers (1979) with a larger sample of countries and with measures pertaining to the year 2000 did find a cross-sectional association between income inequality and life expectancy (Ram 2006). One explanation for the different findings is the fact that we perform our analyses on samples of countries based on the level of economic development, while in the above mentioned study the author uses a pooled sample with countries in different levels of development. Nevertheless, the fact that we failed to observe any significant correlation between income inequality and life expectancy in the high-income sample is still in contradiction with findings from studies that only look at samples of rich countries (Wilkinson and Pickett 2006). For this discrepancy we advance two potential explanations: a substantive and a methodological one.

From a substantive stand-point we argue that the high-developed countries have certain characteristics that counter the (potential) negative effect of income inequality on health. On the one hand, a floor/ceiling effect of the possible life expectancy might be at work. In the rich countries the starting average life expectancy at the beginning of the period under investigation is already high, and a further indefinite increase might not be physically possible. On the other hand, health services and infrastructures were already highly developed in 1987 and they were available for everyone, due to the social and health protection systems developed in their welfare states. In addition, an increase in income inequality in these rich countries does not mean a dramatic or immediate decrease in the quality or the access to health services. These two factors characteristic to high-income countries (i.e., a high life expectancy coupled with a natural limit of the life-span and high quality health services) could imply that the elasticity of life expectancy might be very low under conditions with different income inequality.

The second explanation for the fact that previous studies did find cross-sectional associations of income inequality with health measures is sample bias, also pointed out by other scholars (Babones 2008). In additional analyses we tested this explanation by conducting a simulation where random combinations of 21, 20, 19, 18, 17, 16 and 10 countries were draw from a sample of 23 high-income countries with data for the year 2000, and we repeated the static estimation procedure (results available on request). Our results showed that the composition of the sample of high-income countries in the analysis is one of the main factors in producing a significant or a non-significant association between income inequality and life expectancy. Our method to overcome this issue was to follow the guidelines of prominent institutions such as World Bank and OECD in defining the composition of the sample of countries in various levels of economic development. However, the question of why in some groups of high income countries the association between income inequality and life expectancy was found significant and in other groups no is a pertinent one and deserves further investigation.

Another finding of our study was that, for the low-income countries, an increase in income inequality goes hand in hand with an increase in life expectancy. This result is in line with Biggs et al. (2010) who also observed this in a sample of low and middle income countries. We might tentatively attribute this result to the increasing development of health services and infrastructure in the low-income countries under the pressure of two forces, an external and an internal one. The external force regards the extent of the foreign aid and international efforts targeted at developing countries, and recent research has shown that the foreign aid in the form of health assistance (e.g., vaccination, medicines, health technologies etc.) has positively contributed to improving population health in these countries (Mishra and Newhouse 2009). The internal force might be related to a form of collective action instigated by the interests of a growing middle class in order to address a common problem: the spread of diseases. This phenomenon also was documented in the nineteenth century Europe when epidemics affected major cities as Paris and London and when the solution for this problem was the construction of citywide sewage systems, a public service with wide implications for the health of the population, both rich and poor citizens (De Swaan 1998). If income inequality decreased while the health services and infrastructure increased, the effect of income inequality might pick up the effect of the two unmeasured time-varying variables. Unfortunately, we do not have available data to test this possibility, which leaves it as an open invitation for future research.

Finally, we also argued that the higher availability of resources and the general better quality of life associated with higher economic development would provide protection against the damaging effects of income inequality. Our findings provide support for this idea: average societal income inequality had the strongest negative relationship with life expectancy in low-income countries and the weakest in the high-income countries.

To sum up, the “income inequality thesis” did not receive general and full support from our data. More often than not we found non-significant results and also effects in the opposite direction as expected. Furthermore, we found large differences in the effects of wealth and inequality between countries at different levels of economic development. Referring back to the title of this study, this also has implications for the way in which the health of societies can be improved. Contrary to claims in the literature, our data did not support the idea that a reduction of income inequalities would increase the population health in the high-income countries. In less developed countries, this may be a good strategy though. The same goes for considering economic growth as a strategy to enhance health: our data indicated that this might work in less developed economies, but not in well-developed countries.

The present study shows that the choice of model and the composition of the sample can have far-reaching consequences for the conclusions drawn. Future research can provide better insight in the relationship between income inequality and population health by focusing on the explanation of the differences and similarities encountered between groups of countries even from the same categories of economical development and by investigating the role of the structural processes that are associated with the developments in the income inequality. Analyses that look at within-country and in time processes could provide detailed exploration of the complex relationships between the structural factors and their effects on individual health. In addition, other lines of research could focus on the underlying mechanisms at work, empirically testing the causal chain proposed.