Use of Spanish surnames to identify Latinos: comparison to self-identification

Difficulties in identification of Latinos from secondary datasets have often been addressed by the use of a Spanish surname list. To evaluate the sensitivity and specificity of the 1980 U.S. Census Spanish surname list in correctly identifying Latino ethnicity, we conducted a cross-sectional survey of adult members of a prepaid health plan in the San Francisco Bay Area. We compared a randomly selected list of 1345 Spanish surnames and 717 non-Spanish surnames from health plan membership lists with their self-identified race and ethnicity obtained by telephone interviews. The sample was stratified according to surname group (Spanish or non-Spanish), sex, age, and county. Compared with self-identified ethnicity, a Spanish surname was 88.4% sensitive in identifying Latino men and 70.4% sensitive in identifying Latina women. The non-Spanish surname list was 94.1% sensitive and 95.0% specific in correctly identifying non-Latinos. Although the overall negative predictive value was 97.8%, the positive predictive value of a Spanish surname was only 68.3% for men and 55.7% for women. The low positive predictive value was largely due to the 316 (61%) Filipinos among the 517 persons with Spanish surnames who were not Latinos. We conclude that the use of the Spanish surname list alone to ascertain race and ethnicity in the San Francisco Bay Area both falsely identifies a large number of non-Latino persons as Latino and fails to identify a small proportion of Latinos. Although the level of sensitivity and specificity for Spanish surnames will vary with the area under study, this source of inaccuracy must be considered in studies that plan to use data based on Spanish surnames.