M-estimators of the uniform association model in RxC contingency tables

Abstract

If all cell counts $n_ij$ of a given R x C contingency table are positive, estimates of the expected frequencies mjj can be found by applying any regression estimator to the logarithm of the observed counts. If an R x C table contains outlier(s), ordinary least squares estimates will be affected by the outlier(s). Various authors have proposed several robust estimators sensitive to outliers. In this study, robust estimators were applied to an RxC contingency table with an outlier to obtain the robust parameter estimates instead of the maximum likelihood (ML) estimates, and the results were discussed.