Attribute frequency and misclassification bias

Allowances for potential misclassification bias were performed for three epidemiologic studies. Although the error rates were assumed to be identical for the three studies, there were differences in the degree and even in the direction of the bias. It is shown that they are the result of differences in the true frequencies of the misclassified attributes in these studies. In this regard, the relation of misclassification bias to predictive value is described. Misclassification bias is highly dependent upon the interactions between error rates and the true frequencies of attributes. In general, relative risk estimates are more subject to substantial bias from disease misclassification when cohort studies have very low disease frequencies. Odds ratio estimates are more likely to be substantially biased from exposure misclassification when case-control studies have either very high or very low exposure frequencies. Attribute frequency is an important determinant of bias from misclassification.