Social capital, place and health: creating, validating and applying small-area indicators in the modelling of health outcomes

The aim of this project was to assess whether social capital, measured at the community level, had identifiable effects on health outcomes. A review and assessment of previous work on this subject showed that many studies, in the UK and elsewhere, had: • Been conducted for large spatial units, which bore little relation to the contexts in which people live their daily lives • Relied on aggregate statistical methods, which could not distinguish context from composition. To get round these problems we devised small-area measures or estimates of social capital. We first reviewed direct methods: this entailed an evaluation of the potential of indicators such as blood donation and voter turnout. We then turned to a method we term ‘synthetic estimation’. This was a multilevel modelling approach to the determinants of behaviours thought likely to contribute to social capital formation. It generated coefficients, which could be applied to census data for electoral wards, thereby producing estimates of the proportions of the population expressing attitudes, or engaging in behaviours, which were constitutive of social capital. We validated these measures against other directly-observed data supplied by other researchers. The bulk of the research effort on the project went into generating these estimates but we also applied the indicators in a modelling exercise. The response variable was the probability that an individual respondent to the original Health and Lifestyle Survey (HALS) of 1985 was still alive on re-survey in 1999. Explanatory variables, entered into a multi-level model, included individual characteristics, health-related behaviours, area measures of material circumstances, and area measures of social capital. We found that social capital added little or nothing to models which incorporated area measures of material circumstances. We therefore suggest that, at the scale for which we produced our estimates of social capital, and for the aspects of it deployed in our models, area-level estimates of social capital contribute little to the explanation of variations in health outcomes.