Addressing census data problems in race imputation via fully Bayesian Improved Surname Geocoding and name supplements

Abstract

Prediction of individuals’ race and ethnicity plays an important role in studies of racial disparity. Bayesian Improved Surname Geocoding (BISG), which relies on detailed census information, has emerged as a leading methodology for this prediction task. Unfortunately, BISG suffers from two data problems. First, the census often contains zero counts for minority groups in the locations where members of those groups reside. Second, many surnames—especially those of minorities—are missing from the census data. We introduce a fully Bayesian BISG (fBISG) methodology that accounts for census measurement error by extending the naïve Bayesian inference of the BISG methodology. We also use additional data on last, first, and middle names taken from the voter files of six Southern states where self-reported race is available. Our empirical validation shows that the fBISG methodology and name supplements substantially improve the accuracy of race imputation, especially for racial minorities.