Small group estimation for public health.

We used synthetic estimation and linear regression to estimate the prevalence of selected risk factors and health status indicators in small populations. The derivation was based on the sociodemographic characteristics of the populations and the relationships between these variables and the health variables, as measured by the Ontario Health Survey (OHS). The estimates were validated by a comparison with the direct results of the OHS (gold standards). Synthetic estimates were much less dispersed than the regression estimates or the direct OHS estimates. Regression estimates performed better than synthetic estimates on most validation indicators, and combined approaches performed marginally better yet, although there were few clear patterns. Although correlation coefficients with gold standards in excess of 0.8 were obtained for some variables, the estimates rarely met pre-determined criteria for accuracy. At present these techniques have limited value for public health workers, but further work is justified, especially on approaches combining synthetic and regression estimation.