ABSTRACT
The estimation of and inference from the regression model depend on several assumptions. These assumptions should be checked using regression diagnostics before using the model in earnest. Diagnostic techniques can be graphical, which are more flexible but harder to definitively interpret, or numerical, which are narrower in scope, but require less intuition. Regression diagnostics often suggest specific improvements, which means model building is an iterative and interactive process. It is quite common to repeat the diagnostics on a succession of models. The artificial generation of plots is a good way to “calibrate” diagnostic plots. It is often hard to judge whether an apparent feature is real or just random variation. Repeated generation of plots under a known model, where there is or is not a violation of the assumption that the diagnostic plot is designed to check, is helpful in making this judgment.
