Community conditions contributing to high life expectancy: An analysis of outliers

Life expectancy (LE) varies substantially across U.S. communities and even neighborhoods. Place-based characteristics play an important role, but few studies have used a fine geographic scale to identify local assets that might enhance health outcomes. We are studying census tracts in California in an attempt to identify local assets that might explain “outliers”: census tracts with unexpectedly high or low LE given their level of poverty. To plot census tract LE against poverty, we abstracted all 1999-2001 death records from California vital statistics and geocoded each record to the census tract of the decedent's last residence. Records were aggregated by age group and census tract and combined with 2000 US Census population data to create abridged life tables and calculate census-tract LE. LE values were regressed against census tract-level poverty, education, age, gender, marital status, race, and Hispanic origin. We used weighted linear regression, with reciprocal variance of the LE as regression weights. The results showed a monotonic negative association: LE decreased as the population with incomes below 100% (Federal poverty threshold) or 200% of the poverty threshold increased, and this association held even after adjusting for confounders. We will present scatter plots to show how we identified outliers (census tracts with large residuals that depart from the regression equation) and eliminated outliers suspected to be erroneous. Census tracts thought to be true outliers will now undergo quantitative spatial and non-spatial analysis and qualitative inquiry (key informant interviews) to identify positive assets that might explain their more favorable outcomes.