AbstractThe effect of sampling variation on individual decisions and error rates in logistic discriminant analysis is discussed. The concept of the β‐confidence allocation rule is introduced, which allows testing of whether observations are (in)correctly assigned at a given significance level. The procedure applied to sample data adds valuable information on the sharpness and the stability of the estimated classification rule. The method also suggests that individual posterior probabilities should be associated with a credibility measure. The potential of the approach is illustrated by an example from patients with liver disease.