Estimation of small-area life expectancy: A comparison of two methods

Life expectancy (LE) has become increasingly popular as a tool for demonstrating health disparities across countries, across U.S. states, and across small geographic areas where area-level characteristics may, in part, account for variability in health outcomes. Numerous methodological challenges arise when life table methods, which are well suited to larger populations, are applied to smaller populations. In a study funded by the California Endowment, we compared two approaches to estimating LE at the census tract level: the traditional life table method and Poisson modeling. We tabulated all mortality records from California for the period 1999-2001 and combined those with 2000 US Census data to estimate age-specific mortality rates for all populated California census tracts. The LEs produced by these two methods were compared across several characteristics: consistency across point estimates, standard deviation and confidence bounds, and number of census tract estimates marked as unreliable based on extreme population sizes or tabulated deaths. The results, and the respective abilities of the models to predict LE, suggest that the life table approach is more reliable than Poisson modeling. We will discuss our rationale and the strengths and drawbacks of each LE calculation method. Despite these limitations, our study shows that reliable LE estimation is possible when properly applied to small geographic areas.