Change in income and change in self-rated health: Systematic review of studies using repeated measures to control for confounding bias

Abstract

It is generally assumed that income is strongly and positively associated with health. However, much of the evidence supporting this assumption comes from cross-sectional data or analyses that have not fully accounted for biases from confounding and health selection (the reverse pathway from health to income). This paper reports results of a systematic review of panel and longitudinal studies investigating whether changes in income led to changes in self-rated health (SRH) in adults. A variety of electronic databases were searched, up until January 2010, and thirteen studies were included, using data from five different panel or longitudinal studies. The majority of studies found a small, positive and statistically significant association of income with SRH, which was much reduced after controlling for unmeasured confounders and/or health selection. Residual bias, particularly from measurement error, probably reduced this association to the null. Most studies investigated short-term associations between income and SRH or the effect of temporary (usually one year) income changes or shocks, so did not rule out possibly stronger associations between health and longer-term average income or income lagged over longer time periods. Nevertheless, the true causal short-term relationship between income and health, estimated by longitudinal studies of income change and SRH that control for confounding, may be much smaller than that suggested by previous, mostly cross-sectional, research.

Introduction

The relationship between income and health has been the subject of much research, but most of this research has used cross-sectional data or analysed income and health at only one point in time. Randomised controlled trials of income changes are scarce. A review of those that do exist produced inconclusive evidence of an effect of income supplementation on health outcomes, due to poor quality data (Connor, Rodgers, & Priest, 1999). Another review including all studies, even non-randomised, that evaluated interventions aimed at increasing incomes of those in poverty, also found mixed results and many studies of poor quality (Ludbrook & Porter, 2004). Evidence is suggestive, particularly for the importance of income for the health of those in poverty (Benzeval and Judge, 2001, Kawachi et al., 2010) but a definite causal relationship between income and health has – perhaps surprisingly – not yet been established.

The most thorough review of income and health using longitudinal data was in 2001, which identified sixteen English-language studies that investigated the relationship between income and health over time (Benzeval & Judge, 2001). However, this review was non-systematic, the last included study was published in 1997 and studies of income and health changes were not the focus, although these types of studies are best placed to control for confounding bias. The majority of studies that included a measure of income change found a significant relationship with health, but the health outcome was usually mortality, so was potentially affected by confounding bias, as unmeasured differences between individuals (associated with income) could account for at least some of the mortality variation.

Systematic reviews on the income–health relationship are otherwise lacking, and reviewing the entire literature on this topic would be a massive undertaking.

However, a review of only repeated measures studies that minimise confounding is feasible, and arguably the most relevant. The major source of confounding is from unmeasured time-invariant confounders, or those fixed characteristics of individuals that are unknown or unquantified, such as personality traits, ability, genetics and beliefs (also known as unobserved heterogeneity) that are associated with both income and health (Imlach Gunasekara, Carter, & Blakely, 2008). Methodologically, in observational research, the two best ways to minimise this confounding are: (1) fixed effects models using longitudinal data and repeated measures of both the exposure and the outcome, so that only within-individual changes over time can be analysed, eliminating between-individual differences (Wooldridge, 2002), and (2) instrumental variable (IV) analyses, which may arise in the context of natural experiments (Glymour, 2006), where people are unexpectedly ‘randomly assigned’ an additional source of money. The IV is an exogenous or external variable, unrelated to the outcome except through the exposure variable, that is used when the exposure and outcome relationship is affected by bias from health selection (the reverse causal pathway from health to income) or unobserved heterogeneity (Angrist & Krueger, 2001). However, strong and valid IVs are notoriously difficult to identify.

This paper presents a systematic literature review of longitudinal studies investigating whether a causal relationship between income and health exists, for the question: ‘Do changes in income lead to changes in self-rated health in adults?’ Only papers using longitudinal data and repeated measures data analysis are included. This review differs from Benzeval and Judge (2001) because both change in income and change in health are inclusion criteria, allowing more thorough control of confounding and other biases. Self-rated health (SRH) is the outcome of interest, as it is one of the commonest repeated measures of general health asked in longitudinal surveys. This paper does not attempt to review the effect of income on mental health, the income–health association in children or the effect of wealth on health, all of which are separate but important research questions.