Fit Analyses

Output Variables

Output variables based on the model you fit can be saved in the data window. From the data window, you can store these variables in a SAS data set. This enables you, for example, to perform additional analyses using SAS/STAT software.

Axis variables in residual plots are automatically saved in the data window used to create the analysis. For example, when you create a residual-by-predicted plot, residual and predicted variables are always generated. These variables are deleted when you close the analysis window.

You can save variables permanently by using the fit output options dialog or the Vars menu shown in Figure 39.48. Such variables remain stored in the data window after you close the analysis window.

[menu]
Figure 39.48: Vars Menu

SAS/INSIGHT software provides predicted and residual variables, a linear predictor, a residual normal quantile variable, partial leverage X and Y variables, and influence diagnostic variables.

Influence diagnostics are measures of the influence of each observation on the parameter estimates. These diagnostics include the hat diagonal values, standardized residuals, and studentized residuals. Cook's D, Dffits, Covratio, and Dfbetas also measure the effect of deleting observations.

Some influence diagnostics require a refit of the model after excluding each observation. For generalized linear models, numerical iterations are used for the fits, and the process can be expensive. One-step methods are used to approximate these diagnostics after each fit. The process involves doing one iteration of the fit without the excluded observation, starting with the final parameter estimates and weights from the complete fit.

You can also create generalized residuals such as Pearson, deviance, and Anscombe residuals with generalized linear models. These residuals are applicable to the nonnormal response distributions.

Generated variables use the naming conventions described later in this section. If a resulting variable name has more than 32 characters, only the first 32 characters are used. Generated variables also follow the same numbering convention as the analysis window when you create more than one fit analysis from the same data window. If the generated variable name is longer than 32 characters, the original variable name is truncated to the necessary length.

Output Variables

Hat Matrix Diagonal

Predicted Values

Linear Predictor

Residuals

Residual Normal Quantiles

Predicted Surfaces

Predicted Curves

Standardized and Studentized Residuals

Deviance Residuals

Pearson Residuals

Anscombe Residuals

Partial Leverage Variables

Cook's D

Dffits

Covratio

Dfbetas