The first hypothesis was that income inequality increases health inequalities. In all models and with both versions of the health inequality indicators this could be confirmed. The Gini index appeared as the only independent variable showing a stable significant relation with health inequalities throughout all model specifications.
The second hypothesis regarding social policies is not confirmed. Social protection expenditure is not significantly related to health inequalities even though the coefficients are in the expected direction: Higher social protection expenditures are related with lower health inequalities. Since social protection expenditures are correlated with average population health (.49, p < .001, own analysis) it appears that social policies have a health-promoting impact for all of society—though not specifically for certain groups in need, e.g. the lower income groups. Social policies contribute to better population health but do not show a reducing effect on health inequalities.
As a third hypothesis, I assumed that income inequality and social policies have additive effects on health inequalities. This assumption can neither be declined nor confirmed, since both income inequality and social protection expenditures do influence each other’s impact on health inequalities only slightly (Model 4 compared to Model 1 and 2, respectively). On the one hand, this speaks against the psychosocial mechanism of the relation between social policies and health inequalities. Harmful effects of income inequality on health inequalities are only slightly balanced by social policies (Model 4). On the other hand, the neo-material mechanism, i.e. that the availability of public services directly reduces health inequalities because lower income groups benefit the most, seems to play a part, as economic performance reduces the impact of social policies on health inequalities (Model 6).
Regarding the control variable ‘economic performance’, the findings show a negative link between GDP p.c. (logged) and health inequalities, which means that higher economic performance is related to lower health inequalities. This is contrary to previous studies that found only weak or no associations between GDP p.c. (logged) and health inequalities [
11,
21]. However, a specific of this study is the EVS data which comprises a wide range of countries with various levels of national income (see
Appendix B). Some countries are indeed at a lower stage of economic development, where additional GDP matters for the reduction of health inequalities—contrary to the country selections of the above mentioned studies.
In the introduction, I described two processes of distribution of national income. The analyses show that only the distribution of personal earnings, measured by the Gini index, seems to play a role regarding health inequalities. Redistribution via social policies, measured by social protection expenditures, does not reduce health inequalities. Consequently, when thinking about reducing income inequality in order to reduce health inequalities, social policies do not seem to be the best fit to balance out unequal incomes. However, the reason is the mechanism of how social policies affect health inequalities rather than the mechanism of redistribution by itself. According to Dallinger [
56], government income redistribution works effectively in the way that indeed the lowest income group benefits from public redistribution while the highest income group experiences income losses. The middle class holds its position. Even though social policies are targeted towards lower income groups, they might be too diverse in their impacts to show a distinct health-promoting benefit for disadvantaged income groups. However, to solve this question, further research on specifically health-promoting effects of various social policies is necessary.
Strengths and limitations
With respect to future studies, the limitations of this study should be discussed. In 2008, the European Values Study covered the whole geographical area of Europe. Although the EVS represents a unique dataset that integrates various European societies, it may include field work that varies in quality across different countries.
For macro-comparative analyses, low numbers of units of analysis are typical [
3]. In this case, the number of countries analysed (42) was an inevitable constraint that should be kept in mind when interpreting the results. Usually, to study people nested in countries, the typical approach is to use simultaneous multilevel analysis; instead, to gain more detailed information on single countries, I used a two-step approach—I extracted country-specific effects of household income on subjective health from the micro level at the first step, and subsequently introduced them as dependent variable at the macro level in the second step. This led to the finding of the outlying case of Germany: high health inequalities are combined with a medium level of Gini index, social expenditures, and GDP, as well as medium subjective health at the mean (see
Appendix A and
Appendix B). Future research could show if this is a specific finding and hence an artefact of the EVS data, or whether income-related health inequalities did indeed increase compared to findings based on earlier data.
Since little research has used a comparative approach to focus on
inequalities in health [
51], an agreement on the best indicator for socio-economic health inequalities does not yet exist. Subjective health includes both the physical and mental aspects of health. Even though it is often criticised because it is based on individual perceptions, subjective health is widely used in research on population health as well as health inequalities [
57]. Since this present study is based on within-country income-related health inequalities, cross-national differences in response styles of self-assessed health [
58] are negligible. The question whether socio-economic factors such as income influence respondents’ self-assessment of health, which would bias the estimation of health inequalities, is not solved yet. Jürges [
59] finds that response behaviour varies according to socio-economic groups. On the other hand, Van Doorslaer and Gerdtham [
60] conclude that income-related health inequalities are ‘unlikely to be biased by such reporting tendencies’ (p. 14).
A strong point of this present study is that it tests two different dependent health variables in country-specific models at the first step. When using the effect of income on health, both health variables have certain advantages and disadvantages as indicators of health inequalities. The interpretation of marginal effects at the mean is more straightforward when running regressions on the health dummy variable. However, after combining the categories, less information was obtained compared to using the original 5-point response scale; also, the way the categories were combined is perhaps controversial. Therefore, health inequalities were also calculated on the basis of a dummy variable of (very) poor health versus fair and (very) good health as recommended by Etilé and Milcent [
61]. Probably due to the rather low share of respondents with (very) poor health, only a few countries displayed significant income-related health inequalities. Since it was questionable as to whether this health dummy was an appropriate indicator for health inequalities if it targeted such a small number of respondents, I decided against presenting those results.
Regarding the index of dissimilarity as an indicator of health inequalities, I discovered that using the original 5-point response scale as a metric rather than an ordinal variable led to approximately equivalent results at both the first and second step.
Studying income-related health inequalities across countries imposes the challenge to generate one variable for income across a variety of countries. In this case, the variable had to ensure that respondents’ income in Luxembourg was comparable to respondents’ income in Moldova—to name two extreme cases. Additionally, some countries had a high rate of missing values. Both factors were taken into account when computing the income variable but nevertheless could be interpreted as a limitation of this study. For future studies, education instead of income might be an interesting measure for socio-economic health inequalities. However, given that half of the EVS dataset consists of post-communist countries, where a good part of the adult population was educated during Communism and equal access to education was emphasised [
62], educational health inequalities would need to be interpreted carefully, for they might not adequately describe socio-economic inequalities.
While the Gini index is a widely used and recognised indicator for income inequality, one single predominant measure for the impact of social policies in comparative health inequality research is missing. Dahl and van der Wel ([
63], p. 60) even claimed that ‘a social expenditure approach is new in this field of research.’ Using social protection expenditures in the percentage of GDP as a quantitative measure for social policies should be understood as just a starting point for further analyses. The number of various countries in the EVS made it impossible to find one single data source for social protection expenditures. However, with Eurostat, I found a database encompassing 30 countries (see
Appendix B). Furthermore, I took reasonable care in data investigation for the other countries and tried to double-check with other sources, e.g., national statistics. Although social protection expenditures already are a specification of the comprehensive understanding of social policies, it would be interesting for future research to look at the effects of schemes of social protection, i.e., minimum income protection, on health inequalities.