A note on the influence of pairwise misclassification in two-way tables

The influence of subject misclassification on log-linear analysis of two-way contingency tables is investigated. We define pairwise misclassification and focus on both deviance and Pearson statistics as goodness-of-fit criteria for testing independence between the variables. Attention is paid to the extremal values of the test statistics that can be reached on the basis of the data set. We derive theoretical properties and use them to detect potentially troubling cells in a simple computing routine. Examples illustrate our purpose.