Evaluating small area differential privacy life expectancy

Abstract

The decennial census offers high-quality geographic and demographic information to various stakeholders. Differential privacy (DP) refers to the process of introducing random error into public facing data products such as the decennial census. Studies suggest DP may statistically bias health indicators. This study evaluates neighborhood-level (census tract) life expectancy (LE), highlighting the potential trade-offs between privacy-preserving methods and detecting health disparities in vulnerable populations. LE was calculated using mortality records from Florida between 2009 and 2013 across 4175 census tracts. This study compared US Census Bureau demonstration products with/without DP for two different uncertainty levels (ϵ) = 4, (ϵ) = 29.2. LE estimates were defined as biased (greater than 3 years), and then temporarily replaced with a value to indicate a biased/missing LE value (e.g., ‘1′). Correlation coefficients compared LE calculated from the original (non-DP) age-specific population estimates and the DP age-specific population counts. Next, the study examined the demographic and socioeconomic characteristics of the biased DP LE estimates. Factors significantly correlated with increased log-odds of biased LE estimates included the % Black population (1.03, p < 0.001). Conversely, total population (0.21, p < 0.001), % female-headed households (0.86, p < 0.001), and % population 25+ years with no high school diploma (0.94, p < 0.001) exhibited significant negative associations with biased LE estimates. Future studies should consider the additional uncertainty created by DP when assessing public health interventions and tracking population health over time.