SELF-SELECTING ROBUST LOGISTIC REGRESSION MODEL

Abstract

Binary data is a common response data in many elds of research including nance, social sciences, psychology and medicine. The most common model used for the analysis of binary data is the logistic regression model. However, the problem of identi cation and corresponding treatment of in uential outliers still remains to be well studied to check the adequacy of the tted binary logistic models. Many researchers have developed robust statistical model to solve this problem related to the presence of atypical observations in the data. Gelman (2004) proposed a model that dealt with outliers problem by trimming the probability of success in logistic regression. The trimming values in this model are xed and the user is required to specify this value well in advance. We explore this work and other robust logistic regression models then extend this work to allow for the trimming value to be estimated from the data. In particular, this research work presents a self selecting robust logistic regression (SsRLR) model. We proved that the SsRLR model is more robust to the presence of leverage points in the data. Parameter estimations is done using a full Bayesian approach, implemented in WinBUGS 14 software.