Identifying racial and ethnic disparities in occupational health using administrative data

Background: Although occupational segregation by race and gender has declined since the 1960s, it is still widespread in the United States, with some of the most dangerous, repetitive and low-paying jobs reserved for people of color and white women. Unfortunately, the magnitude and specific impacts of work-related injuries and illness to individual workers, families and society are under-researched, in part because workers' compensation systems do not collect data on race and ethnicity. This study attempts to fill this gap by utilizing new statistical methods to impute race and Latino ethnicity in a population of injured workers. Methods: This study utilizes the Bayesian Improved Geocoding (BISG) method (Elliott et al., 2009) to impute race and ethnicity by calculating probabilities using Census surname lists and Census block groups to identify the probability that a subject belongs in a specific racial or ethnic group. Squared correlations and Receiver Operating Characteristic (ROC) curves are computed. Results: We found significant differences in squared correlations between men and women. Receiver Operating Characteristics (ROC) curves were plotted for each racial/ethnic group and stratified by gender. All imputed racial/ethnic categories showed exceptionally good ability to correctly classify individuals by race/ethnicity with areas under the curves between 0.89 (white women) to .98 (Latino men). Discussion: The BISG appears to be an acceptable method for imputing race and ethnicity in administrative datasets. More research with larger sample sizes is needed to test the utility of the BISG in identifying occupational health disparities in workers' compensation or other occupational racial inequalities.