In this study, we examined consistency of infant race/ethnicity across two data sources (N=2,63) using measures of sensitivity and positive predictive value. First we created and compared conventional measures of infant race/ethnicity from 2007 Oklahoma birth certificates and SEED for Oklahoma Kids baseline survey data, classifying infants as White, African American, American Indian, Asian, or Hispanic. Then we created and tested alternative measures with a biracial classification, based on biological parentage from birth certificates or parent report of infant biracial identity in the survey. We find that, for conventional measures, sensitivity is highest for Whites and African Americans and lowest for Hispanics. Positive predictive value, meanwhile, is highest for Hispanics and African Americans, and lowest for American Indians. Alternative measures improve values for Whites, but yield mostly low values for minority and bi-racial groups.
Our main conclusions are that health disparities research should consider the source and validity of infant race/ethnicity data when creating sampling frames or designing studies that target infants by race/ethnicity. The common practice of assigning the maternal race/ethnicity as infant race/ethnicity should continue to be challenged.
Project: SEED for Oklahoma Kids