Visual Assessment of Residual Plots in Multiple Linear Regression: A Model-Based Simulator Perspective

Abstract

This article follows a recommendation from the regression literature to help regression learners become more experienced with residual plots for identifying assumption violations in linear regression. The article goes beyond the usual approach to residual displays in standard regression texts by taking a model-based simulation perspective: simulating the data from a generating model and using them to estimate an analytical model. The analytical model is a first order linear regression model; whereas the generating model violates the assumptions of the analytical model. The residuals from the analytical model are plotted to demonstrate assumption violations to provide experience for regression learners with characterized residual patterns. The article also briefly discusses remedial measures.