Self-Critical and Robust, Procedures for the Analysis of Binary Data

摘要

A new and general method for the critical analysis of binary data is given. Since the method invokes a memory of the distributions and structural model assumed in the processing of the data it is termed self-critical. A single parameter c generates a family of critical estimator - and tests of hypotheses, if warranted. When c=0, the procedure reduces to that of maximum likelihood. The method is derived from primitive considerations exactly parallel to those which result in critical methods for complete data. Both symmetric and asymmetric quantal distributions are considered for regression-type models. A critical procedure for the proportional hazards model, an alternative to the logistic regression model, is introduced and sucessfully applied. Asymptotic results are given. Experience with live data is reported in the examples. Keywords: Robust procedures; Gaussian distribution; Weibull distribution; Asymptotic Covariance matrix; Empirical efficiency ratio.