Variation in heart disease mortality across census tracts as a function of overdispersion and social class mixture

Variation in heart disease (HD) mortality rates across census tracts is greater than expected given binomial error and available explanatory variables. We extended an extra-binomial variation model for rates standardized by the direct method. The overdispersion parameter accounted for 36 per cent of the observed variation in standardized rates. Ignoring overdispersion resulted in a change in an estimate of the effect of social class on HD mortality and substantial underestimation of the error of the estimates of such effects. Ecologic regression on the proportional mixture of social classes within tracts provided an appealing approach to the problem of estimating fixed effects with aggregated data.