Hierarchical Log-Linear Models Not Preserved by Classification Error

Abstract

Abstract A model of Bross (1954) for classification error in 2 × 2 contingency tables is extended to higher dimensional tables. It is known (Mote and Anderson 1965) that the usual hypothesis tests of independence for a sampled two-dimensional table have the nominal significance level in the presence of misclassification. A simple criterion is given to determine which hierarchical log-linear models in higher dimensional tables do not share this property, that is, are not preserved by classification error.