Speaking Stata: Graphing Model Diagnostics

Abstract

Plotting diagnostic information calculated from residuals and fitted values is a long-standard method for assessing models and seeking ways of improving them. This column focuses on the statistical mainstream defined by regression models for continuous responses, treated in a broad sense to include (for example) generalized linear models. After some comments on the history of such ideas (and even their anthropology and psychology), the commands available in official Stata are reviewed, and a modeldiag package is introduced. A detailed example on fuelwood yield from fallow areas in Nigeria illustrates a variety of general points and specific tips.