Models for Small Area Estimation for Census Tracts

Estimating uncertainty in a socioeconomic index derived from the American community survey

Socioeconomic indexes are widely used in public health to facilitate neighborhood-scale analyses. Although they are calculated with high levels of precision, they are rarely reported with accompanying measures of uncertainty (e.g., 90% confidence intervals). Here we use the variance replicate tables that accompany the United States Census Bureau's American Community Survey to report confidence intervals around the Yost Index, a socioeconomic index comprising seven variables that is frequently used in cancer surveillance. The Yost Index is reported as a percentile score from 1 (most affluent) to 100 (most deprived). We find that the average uncertainty for a census tract in the United States is plus or minus 8 percentiles, with the uncertainty a function of the value of the index itself. Scores at the extremes of the distribution are more precise and scores near the center are less precise. Less-affluent tracts have greater uncertainty than corresponding more-affluent tracts. Fewer than 50 census tracts of 72,793 nationally have unusual distributions of socioeconomic conditions that render the index uninformative. We demonstrate that the uncertainty in a census-based socioeconomic index is calculable and can be incorporated into any analysis using such an index.