Regression Diagnostic Plots

In Multiple linear regression models, problems arise when serious multicollinearity or influential outliers are present in the data. Failure to include significant quadratic or cross-product terms result in model specification error. Simple scatter plots are most of the time not effective in revealing the complex relationships of predictor variables or data problems in multiple linear regression. However, partial regression plots are considered useful in detecting influential observations and multiple outliers; partial residual plots or the added-variable or component-plus-residual plots are useful in detecting nonlinearity and model specification errors. The leverage plots available in SAS JMP software are considered effective in detecting multicollinearity and outliers. The VIF-plot, which is very effective in detecting multicollinearity, can be obtained by overlaying both partial regression and partial residual plots with a common centered X-axis. SAS macros for displaying these diagnostic regression plots are presented here