Using matched areas to explore international differences in population health

In this paper, we develop and test a method for examining the influence of national level contextual influences on population health. Acknowledging calls for the use of experimental study designs to explore contextual influences on health, we develop a study design in which sets of local areas from Britain and Belgium became akin to two ‘treatment’ groups; one exposed to British society and culture and the other exposed to Belgian society and culture. The areas are matched on the basis of showing very strong similarities in economic, demographic and historical characteristics. Data describing these characteristics are obtained from national census data. A principal component analysis of these variables permits areas in Britain and Belgium with similar scores on the resulting components to be matched into pairs. A sequence of logistic regression models identifies between-country difference in the risk of reporting poor health. Our final model compares the risk of reporting poor health among Belgians and people from Britain living in similar local contexts, adjusting for any residual differences in individual level characteristics. We compare results from this new method with those from more conventional approaches. All approaches show that residence in Britain is associated with a substantial and significantly higher risk of reporting poor health for both men and women, after adjustment for both individual and local contextual influences. We then critically reflect on our method and on the context-composition framework for research into area variation in health. We conclude that whilst our approach succeeded in applying the idea of comparable groups with different exposures to an observational, international comparison, it also brought associated questions about external validity and the extent to which a sample of matched areas captures a ‘national’ context.