Variation in Hispanic Self-Identification, Spanish Surname, and Geocoding: Implications for Ethnicity Data Collection

This study examines the variation in surname analysis and geocoding, and their association with self-identified Hispanics in an HMO. We collected ethnicity data from three studies, and employed Spanish surname software and cen- sus tract level geocoding to create proxies for Hispanic ethnicity. We computed sensitivity, specificity, and estimated mul- tivariate logistic regression models to examine the variation in the likelihood of a match between self-identified Hispanics and surname. Sensitivity and specificity with respect to surname varied across the three studies, ranging from 57%-91% and 89%-96%, respectively. Relative to self-report, the sensitivity of the census tract measure of density of Hispanics, var- ied from 5%-15%. Multivariate models suggest that the likelihood of a match between self-identified Hispanics and sur- name was not associated with age or gender. Self-identified Hispanics living in neighborhoods with the highest density of Hispanics were less likely than those in more mixed neighborhoods to have a Spanish surname. Employing the Spanish surname software on only densely populated Hispanic census tracts may not always improve the likelihood of correctly identifying Hispanic subjects.