Summary. Racial disparities in risks of mortality adjusted for socio-economic status are not well understood. To add to the understanding of racial disparities, we construct and analyse a data set that links, at individual and zip code levels, three government databases: Medicare, the Medicare Current Beneficiary Survey and US census. Our study population includes more than 4 million Medicare enrollees residing in 2095 zip codes in the north-east region of the USA. We develop hierarchical models to estimate the black–white disparities in risk of death, adjusted for both individual level and zip code level income. We define the population level attributable risk AR, relative attributable risk RAR and odds ratio OR of death comparing blacks versus whites, and we estimate these parameters by using a Bayesian approach via Markov chain Monte Carlo sampling. By applying the multiple-imputation method to fill in missing data, our estimates account for the uncertainty from the missing individual level income data. Results show that, for the Medicare population being studied, there is a statistically and substantively significantly higher risk of death for blacks compared with whites, in terms of all three measures AR, RAR and OR, both adjusted and not adjusted for income. In addition, after adjusting for income we find a statistically significant reduction in AR but not in RAR and OR.