Exploring the generalisability of the association between income inequality and self-assessed health

Abstract

A growing between- and within-country literature suggests that the association between income inequality and health reflects individual- or area-level characteristics with which income inequality is associated, rather than the effects of income inequality per se. These studies also suggest that the association between income inequality and health is country-specific. Unresolved methodological issues include the geographical level at which to model the effects of income inequality, and the appropriate statistical methods to use. This study compares the results of single-level and multi-level logistic regression models estimating the association between income inequality and self-assessed health in local authorities in Scotland. The results suggest that there is a significant positive association between income inequality and health across local authorities in Scotland, even after adjusting for individual-level socio-economic status. They also suggest that there is significant local authority-level variation in self-assessed health, but this is small compared to the variation at the individual level. Income and other measures of individuals’ socio-economic status are more strongly associated with self-assessed health than income inequality. This study provides further evidence that the income inequality:health association is place-specific. It also suggests that methodological choices regarding the ways of estimating the association between self-assessed health, individual-level socio-economic status and area-level income inequality may not make a substantive difference to the results when contextual effects are small. Further work is required to test the sensitivity of these conclusions to alternative levels of geographical aggregation.

Introduction

A substantial literature demonstrating a negative association between income inequality and mortality, between and within countries, focussed attention on the potentially damaging effect of income inequality on population health. The robustness of this association in cross-country studies was subsequently challenged (Judge, 1995; Lynch, Davey Smith, Hillemeier, Shaw, Raghunathan, & Kaplan, 2001; Mellor & Milyo, 2001). Debate continues as to the causal mechanisms underpinning the relationship. Some authors suggest it reflects diminishing returns to absolute incomes at the individual level (Gravelle, 1998). Others implicate the physiological and psychological consequences of low social status in societies where incomes and other material resources are very unequally distributed (Wilkinson (1996), Wilkinson (1997)). Others emphasise the systematic variations between social and economic groups in the availability of private resources and health-related public infrastructure, which reflect the broader economic and social policies that have also created income inequality (Lynch, Davey Smith, Kaplan, & House, 2000; Coburn, 2004).

Explanations suggesting that the association between income inequality and health may not reflect a generalisable causal effect of income inequality per se are supported by several recent review articles (Wagstaff & Van Doorslaer, 2000; Mackenbach, 2002; Deaton, 2003; Lynch et al., 2004) and empirical analyses (Deaton, 2001; Mellor, & Milyo (2002), Mellor, & Milyo (2003); Deaton & Lubotsky, 2003; Blakely, Atkinson, & O’Dea, 2003), although the recent evidence is mixed (Subramanian & Kawachi, 2003a). Much of the evidence comes from the US, where the strongest and most consistent associations are found at the state level (Subramanian, Blakely, & Kawachi, 2003a; Subramanian & Kawachi, 2003b).

A growing literature from other affluent countries finds little or no association between income inequality and health, in particular after adjustment for individual and area-level demographic and socio-economic variables. The results of ecological studies are mixed. A study of English local authorities found that both income inequality and mean income were independently associated with all-cause, all-age SMRs (Stanistreet, Scott–Samuel, & Bellis, 1999). In contrast, although there was a significant relationship between income inequality and mortality in states and metropolitan areas in the US, no such relationship existed in provinces and metropolitan areas in Canada (Ross et al., 2000).

Individual-level studies are more consistent in finding limited or no associations between income inequality and health, after adjusting for individual-level covariates. One of the few UK studies found no overall relationship between income inequality and presence of mental disorders across the regions of the UK (Weich, Lewis, & Jenkins, 2001), although there was an interaction between income and inequality such that for the more affluent group, greater income inequality was associated with higher prevalence, whereas for the lowest income category the reverse was true. In a separate study, the same authors found a positive association between various measures of income inequality and the odds of poor self-rated health, but the relationship was only statistically significant for the Gini coefficient (Weich, Lewis, & Jenkins, 2002). For all income measures, it was stronger and more consistent for the lowest income quintile.

A Japanese study found a strong relationship between self-assessed health and income inequality at the prefecture level, which was no longer significant once income and other sociodemographic variables were included (Shibuya, Hashimoto, & Yano, 2002). A study of regions in New Zealand found an association between income inequality and mortality that was robust to the inclusion of regional and household income but which disappeared after the inclusion of ethnicity (Blakely et al., 2003).

Longitudinal studies in Sweden (Gerdtham & Johannesson, 2001), Denmark (Osler Prescott, Gronbak, Christensen, Due, & Engholm, 2002) and Israel (Shmueli, 2004) support the view that income inequality is not strongly associated with health after taking into account individual and other ecological variables, whilst a Canadian study found a positive relationship between income inequality and health after adjusting for health behaviours, education and individual- and area-level incomes (McLeod, Lavis, Mustard, & Stoddart, 2003).

Overall, the evidence provides little support for the hypothesis that there is an independent association between income inequality and health that is generalisable across countries and across geographical levels. Two lines of argument have developed in response to this growing body of evidence (Subramanian et al., 2003a). The first highlights reasons why results in non-American studies might be expected to differ from those in the United States. These include lower levels of income inequality within areas, which would reduce the impact of income inequality relative to the US if there is a threshold effect (Shibuya et al., 2002); smaller differences in income inequality between areas than in the US, which would reduce the power of studies to detect the impact of income inequality (Gerdtham & Johannesson, 2001; Blakely et al., 2003); and the impact of more comprehensive programmes of welfare provision than in the US, which might offset the impact of income inequality (Osler et al., 2002).

The second line of argument is that results regarding the effect of income inequality on health might differ between studies because of the different ways in which multi-level effects have been modelled (Subramanian et al., 2003a). Random effect or variance components models have been suggested as the most appropriate technique in this context.

Differences between studies may also reflect different levels of geographical aggregation at which analyses have been carried out (Weich et al., 2002). Ideally, the choice of levels should be determined by clear hypotheses about the mechanisms by which income inequality is thought to affect health and the levels at which these mechanisms might operate. However, this remains a contested area (Wagstaff & Van Doorslaer, 2000; Wilkinson, 1997). In addition, data availability limits the levels at which analyses can be carried out in practice.

This paper therefore has two objectives. Firstly, it uses data from the Scottish Household Survey (SHS) to test the hypothesis that income inequality at the level of local authorities in Scotland is associated with individuals’ health, after accounting for individual demographic and socio-economic status. Differences between the US and the non-US literature suggest that we need country-specific analyses of the income inequality:health association. It is unclear where the UK sits in relation to the mixed evidence of an association in the US and the absence of evidence of an association in most studies from other affluent countries. The UK has high levels of income inequality compared to many of the more affluent countries in which the non-US studies have been undertaken. If there is a threshold above which the effect of income inequality increases, then any effects of income inequality in the UK might be stronger than in other affluent nations. However, the UK also has a long-established and extensive welfare state.

The combined effect of these countervailing factors is not known. There are only three UK studies to date, which suggest that income inequality may have some effects on health (Lynch et al., 2004), and there are no studies of the income inequality:health association within Scotland. Scotland differs from the rest of the UK in a number of relevant respects. It has comparable levels of income inequality to many English regions and Wales, but lower income inequality overall, due in large part to a high level of income inequality in London. It also has a higher prevalence of low incomes, worse health and higher levels of some forms of social spending than the rest of the UK.

The second aim is to test whether estimates of the extent of any association are sensitive to the way in which the multi-level structure of the data is addressed.