We can understand the details of the model 3 distributions in Figure 10 as follows. In case (a), the bias values are c1 = 1, c2 = 6 and c3 = -1 using Eq (23) θ = 13. Using Eq (24), the ICC absolute compliance population is then ρ3A = 102/(102 + 13 + 52) ≈ 0.725. When using Eq (27), the icc consistency population ρ3C = 102/(102 + 52) = 0.8. Figure 10 shows that the maxima of the ICC (A,1) and ICC (C,1) distributions are positioned close to these values 3A and ρ3C in case (a). The width of the ICC distribution (A,1) obtained with model 3 is slightly less than the width of the ICC distribution (A,1) (“bias = 5”) obtained with model 2 of Figure 9, which reflects the fact that the width is due only to the variation in the values of subject rj and noise vij, since the cj are fixed. An F-test can be used to confirm if there are any distortions. Variances and confidence intervals can then be calculated from the resulting model (without or with biases). In case of distortion, both ICC absolute compliance and ICC consistency should be reported, as they provide different and additional information on the reliability of the method. A clinical example with data from the literature is given. We can dwell for a moment on the terminology. Model 2, which will be discussed later, is called a two-way random effect model, which means the introduction of a random sample of measurement distortions.

. . .