Error in the Measurement of Mortality: An Application to the Analysis of Racial Mortality Disparity

Yu-Chieh Hsu, Carnegie Mellon University
Lowell Taylor, Carnegie Mellon University

A large empirical literature studies the forces that shape racial disparity in mortality. Given that factors early in one's life can be important for subsequent mortality outcomes, such research often relies on panel data. An important example is the National Longitudinal Survey of Older Men (NLS-OM), which collected data for men aged 45--59 in 1966 and several subsequent years, and then also reported deaths as indicated by death certificate data collected in 1990. An important methodological issue arises in studies that use such data: deaths are likely to be under-reported, most likely in systematic ways. In the NLS-OM, for example, the matching procedure appears to have missed a substantial number of deaths. We work out a simple model that illustrates the effect of this measurement error, and then show that inappropriate handling of the measurement error in survival analysis causes serious problems for inference.