To the Editor.
—The failure of Hahn et al1 to use appropriate statistics for gauging the consistency of categorical data leads to conclusions that are at best overstated and at worst misleading. Reanalysis of the birth/death agreement data presented by Hahn et al using co efficient κ instead of the uncorrected "percentage agreement" statistics2 (Tables 1 and 2) indicates that coding for most categories of race and Hispanic origin meet or exceed the standards for fair to good levels of agreement and are not "remarkably inconsistent." Point estimates for coefficient κ suggest that the only category of race coded with poor consistency is "Other"—a category that is vaguely defined in the first place. Hahn et al were correct in noting that whites and blacks are coded with a high level of consistency; inspection of κ coefficients suggests that coding of two other race categories, American Indians and Hawaiians,
Fendrich M. Inconsistent Coding of Race and Ethnicity in Infants. JAMA. 1992;267(23):3151–3152. doi:10.1001/jama.1992.03480230043015