resid() or resid(Model) |

resid(), with no argument, computes a REAL matrix of various quantities useful in the analysis of residuals. It uses side effect variables RESIDUALS, HII, etc. produced by the most recent GLM (generalized linear or linear model) command such as regress(), anova(), or poisson(). It is an error if any of the needed side effect variables do not exist. resid(Model) first executes manova(Model, silent:T) to compute the required side effect variables before computing the residual-related quantities. Model should be a CHARACTER variable or string specifying a linear ANOVA or MANOVA model. Any factors in the model will be treated as factors. If you want them treated as variates, use resid(Model,T). Each row of the result corresponds to a case. When the dependent variable Y is univariate, there are 5 columns, as follows: Col. 1 Y = observed response Col. 2 Studentized residuals = RESIDUALS/SE(RESIDUALS) Col. 3 HII = leverage Col. 4 Cook's distance Col. 5 t-statistics = externally studentized residuals = RESIDUALS/SE*(RESIDUALS), where SE* for each case is a standard error based on the model fit excluding that case. When Y is multivariate of dimension p, there are 4*p + 1 columns -- the p values for Y, the p standardized residuals, HII, the p Cook's distances, and the p externally studentized residuals. If a case has missing values, most entries for that case will be MISSING and there are no useful numbers. After non-linear GLM commands such as poisson() and logistic(), the results are based on the last stage of the iteratively reweighted least squares algorithm used to fit the model. Residuals are standardized by the error mean square in the linear scale. They should still be valid for diagnosing departures from the model. The output of resid() is modeled on what is printed by the resid() command in program Multreg. resid() is implemented as a pre-defined macro. See also topics 'glm', yhat(), resvsindex(), resvsrankits(), resvsyhat().

Gary Oehlert 2003-01-15