Includes bibliographical references (p. 89-92).

Linear Least-Squares Regression -- The Regression Model -- Least-Squares Estimation -- Statistical Inference for Regression Coefficients -- The General Linear Model -- Collinearity -- Collinearity and Variance Inflation -- Coping with Collinearity: No Quick Fix -- Outlying and Influential Data -- Measuring Leverage: Hat-Values -- Detecting Outliers: Studentized Residuals -- Measuring Influence: Cook's Distance and Other Diagnostics -- Numerical Cutoffs for Diagnostic Statistics -- Jointly Influential Subsets of Observations: Partial-Regression Plots -- Should Unusual Data Be Discarded? -- Non-Normally Distributed Errors -- Normal Quantile-Comparison Plot of Residuals -- Histograms of Residuals -- Correcting Asymmetry by Transformation -- Nonconstant Error Variance -- Detecting Nonconstant Error Variance -- Correcting Nonconstant Error Variance -- Nonlinearity -- Residual and Partial-Residual Plots -- Transformations for Linearity -- Discrete Data -- Testing for Nonlinearity -- Testing for Nonconstant Error Variance -- Maximum-Likelihood Methods, Score Tests, and Constructed Variables -- Box-Cox Transformation of y -- Box-Tidwell Transformation of the xs -- Nonconstant Error Variance Revisited -- Recommendations -- Computing Diagnostics -- Least-Squares Fit, Joint Confidence Regions, and Tests -- Ridge Regression -- Hat-Values and the Hat Matrix -- The Distribution of the Least-Squares Residuals -- Deletion Diagnostics -- The Partial-Regression Plot -- Smoothing Scatterplots by Lowess -- Weighted-Least-Squares Estimation -- Correcting Least-Squares Standard Errors for Heteroscedasticity -- The Efficiency and Validity of Least-Squares Estimation When Error Variances Are Not Constant.

With Regression Diagnostics, researchers now have an accessible explanation of the techniques needed for exploring problems that compromise a regression analysis and for determining whether certain assumptions appear reasonable. The book covers such topics as the problem of collinearity in multiple regression, dealing with outlying and influential data, non-normality of errors, non-constant error variance and the problems and opportunities presented by discrete data. In addition, sophisticated diagnostics based on maximum-likelihood methods, scores tests, and constructed variables are introduced.