The Contribution of First-name Information to the Accuracy of Racial-and-Ethnic Imputations Varies by Sex and Race-and-Ethnicity Among Medicare Beneficiaries

Background: Data on race-and-ethnicity that are needed to measure health equity are often limited or missing. The importance of first name and sex in predicting race-and-ethnicity is not well understood. Objective: The objective of this study was to compare the contribution of first-name information to the accuracy of basic and more complex racial-and-ethnic imputations that incorporate surname information. Research Design: We imputed race-and-ethnicity in a sample of Medicare beneficiaries under 2 scenarios: (1) with only sparse predictors (name, address, sex) and (2) with a rich set (adding limited administrative race-and-ethnicity, demographics, and insurance). Subjects: A total of 284,627 Medicare beneficiaries who completed the 2014 Medicare Consumer Assessment of Healthcare Providers and Systems survey and reported race-and-ethnicity were included. Results: Hispanic, non-Hispanic Asian/Pacific Islander, and non-Hispanic White racial-and-ethnic imputations are more accurate for males than females under both sparse-predictor and rich-predictor scenarios; adding first-name information increases accuracy more for females than males. In contrast, imputations of non-Hispanic Black race-and-ethnicity are similarly accurate for females and males, and first names increase accuracy equally for each sex in both sparse-predictor and rich-predictor scenarios. For all 4 racial-and-ethnic groups, incorporating first-name information improves prediction accuracy more under the sparse-predictor scenario than under the rich-predictor scenario. Conclusion: First-name information contributes more to the accuracy of racial-and-ethnic imputations in a sparse-predictor scenario than in a rich-predictor scenario and generally narrows sex gaps in accuracy of imputations.