Hence, this model does not only address agreement between methods but additionally provides some information about bias (Agresti, 1992). Information about bias can be obtained by the comparison of . the one-variable effects that represent the marginal distributions. If these effects differ between observers, the observers have different classification probabilities for a given object, which simply means that they do not use every category in the same manner. This model is called the quasi-symmetry model because the expected cell frequency to receive a particular response by the first rater (e.g., hyperactive) and a particular response by the second rater (e.g., normal) differs by the same ratio [exp (AAB)] from the expected cell frequency given only the one-variable effects as the contrary combination (normal by A and hyperactive by B). In other words, associations between both raters are "mirrored" around the main diagonal (see Table 17.7a). For example, expected frequencies of the
Was this article helpful?