A New Method for Estimating Racial/Ethnic Disparities where Administrative Records Lack Self-Reported Race/Ethnicity

Marc Elliott, RAND
Allen Fremont, RAND
Nicole Lurie, RAND
Peter A. Morrison, RAND
Philip Pantoja, RAND
Allan Abrahamse, RAND

With increasing ethnic diversity, equity is a continuing focus of policy formulation and political debate. We consider the need by health plans to monitor racial/ethnic disparities in health care quality among their enrollees. Few plans acquire racial/ethnic data from their entire membership. Where classification variables are missing, individuals’ surnames and neighborhood contextual measures can provide useful surrogate data elements for comparing population subgroups. Building on the strengths of surname analysis and neighborhood contextual analysis, we present and evaluate a hybrid method which is broadly applicable where researchers must rely on administrative records lacking racial/ethnic detail. This Bayesian Algorithm integrates both sources of information and substantially outperforms other approaches. It performs well when race/ethnic classification is the only goal or when estimated race/ethnicity is to be a predictor in regression or other models. Thus its potential applications are not limited to estimation of disparities or to health applications.

  See paper

Presented in Session 119: Alternative Measurement of Race and Ethnicity