Performance Comparison between CERES and PR Plots in Detecting the Heteroscedasticity Problem using Liver Cancer and Simulated Data

Binomial regression model is a very important type of generalized linear model (GLM) and has been applied in wide area such as Liver cancer data. In diagnosing the model, we need to check for the heteroscedasticity problem because it happens when the standard errors of a variable are non-constant. heteroscedasticity will impact the validity and abuse the assumptions of binomial regression. Commonly, conditional expectations and residuals (CERES) and partial residual (PR) plots have been implemented for the identification of heteroscedasticity in the data set. In this paper, we present the comparison analysis between CERES and PR plots by using two different kind of data which are liver cancer data and simulated data. In simulation, we have selected four different sample sizes which are 25, 50, 100, and 200, and variance of error term = 0.00, 0.15, 0.38, 0.75. After that, a 10,000 simulated data is conducted using the R software. The results show the PR plots can detect heteroscedasticity better than CERES plots as due to the larger disperity and the PR plots gives better visual diagnostic for heteroscedasticity as compare to CERES plots. As a summary, this research found that PR plots is the appropriate solution for checking the heteroscedasticity.

Keywords: CERES plots, Partial residual plots, Heteroscedasticity