Pilot testing a proposed method to assign race and ethnicity to administrative data

Background and Objectives: The evidence of occupational health disparities while growing, is still limited, relying primarily on small scale studies or supplements to larger national studies, both of which are frequently missing detailed data on occupational injuries and outcomes. Workers' compensation data, while not complete, offers an alternative, and contains rich detail on the injury and subsequent medical sequelae. Workers' compensation data typically lacks information on race and ethnicity. The objective of this study is to evaluate a method for imputing race and ethnicity using data from 795 workers in Washington State with self-reported race and ethnicity.

Methods: We used the Bayesian Improved Surname Geocoding or BISG, which uses probabilities from Census surname lists and the Census measure for each subjects' neighborhood racial/ethnic makeup to develop a Bayesian probability for race and ethnicity either at the aggregated group level or for an individual injured worker.

Results: We expect findings similar to Elliott et al., (2009), who found correlations between imputed and self-reported data to be 0.82 for Latino, 0.77 for Asian, 0.70 for Black, and 0.76 for White for individual predicted race/ethnicity.

Conclusions: While no method of imputing race or ethnicity will ever be superior to self-reported data, given the current and likely future difficulties in obtaining race and ethnicity in many occupational injury datasets, using the BISG, is a relatively low-resource way to identify highly probable racial and ethnic disparities in both occupational injuries as well as health service outcomes.