Stratifying Populations by Race

Epidemiologists attempt to understand variations within a population in a biomedical trait, e.g., risk of heart disease or rate of diabetes-related death, by stratifying the population using a demographic variable like sex, age, or race and studying the statistical relationship between the variable and the trait. In order to do so, they assign each member of the population or sample a race and assume that there is one correct way to make the assignment, e.g., by ancestry, other-reports or ancestry. I argue that there is no one correct way to assign an individual to a racial category and that what race an individual should be assigned depends of the trait whose variation the epidemiologist is attempting to understand; as a result, a member of the population might be assigned one race for the purpose of understanding a variation in one biomedical trait and a different race for the purpose of understanding a variation in a different one, white in relation to sickle-cell disease and black in relation to congestive heart failure. I propose an approach to race similar to one some economists have adopted towards an index like poverty, viz. that there is no best way to define ‘poverty' and which definition is best depends on what the term is to be used for. My proposal would improve the research in epidemiology on racial difference and oppose the common view that race is an intrinsic property of persons.