Multicollinearity in Binomial Regression: A Comparison between Conditional Expectations and Residuals (CERES) and Partial Residual (PR) Plots for Detection

Abstract

Conditional expectations and residuals (CERES) and partial residual (PR) plots have been used in linear regression model for the identification of multicollinearity. But not much work has been done on how they perform in generalized linear models (GLM). Binomial regression model (BRM) is a very important type of GLM which has wide applications in dealing with heart disease and many other types of data. In this paper we have offered a comparison between CERES and PR plots in BRM to detect the multicollinearity problem. At first, we have developed a comparison tool and then apply them to real-world and simulated data. We observe the performance of these plots on the detection of a possible multicollinearity separately. We observe that both these plots perform well in order to diagnose this problem for a real data. However, the overall performance of the CERES plot is found better as compared to the PR plots