Model Fitting: Generalized Linear Models

The Output Variables Tab

You can use the Output Variables tab (Figure 24.24) to add analysis variables to the data table. If you request a plot that uses one of the output variables, then that variable is automatically created even if you did not explicitly select the variable on the Output Variables tab.

For a multinomial response, residuals and influence diagnostics are not available.

The following list describes each output variable and indicates how the output variable is named. y represents the name of the response variable. If you use events/trials syntax, then y represents the name of the events variable.

Proportions for events/trials
adds a variable named Proportion_et, where e is the name of the events variable and t is the name of the trials variable. The value of the variable is the ratio et. This variable is added only when you use events/trials syntax.
Predicted values
adds predicted values. The variable is named GenP_y.
Confidence limits for predicted values
adds 95% confidence limits for the predicted values. The variables are named GenLclm_y and GenUclm_y.
Linear predictor
adds the linear predictor values. The variable is named GenXBeta_y.
Raw residuals
adds residuals, calculated as observed minus predicted values. The variable is named GenR_y.
Pearson chi-square residuals
adds the Pearson chi-square residuals. The variable is named GenChiSqR_y.
Deviance residuals
adds the deviance residuals. The variable is named GenDevR_y.
Likelihood residuals
adds the likelihood residuals. The variable is named GenLikR_y.
Cook's D
adds Cook's d influence statistic. The variable is named GenCooksD_y.
Leverage (H)
adds the leverage statistic. The variable is named GenH_y.
DFBETAS (influence on coefficients)
adds p variables, where p is the number of parameters in the model. A classification variable with k levels counts as k parameters. The variables are scaled measures of the change in each parameter estimate and are calculated by deleting the ith observation. Large values of DFBETAS indicate observations that are influential in estimating a given parameter. Belsley, Kuh, and Welsch (1980) recommend 2/\sqrt{n} as a size-adjusted cutoff. The variables are named DFBeta_j, for j=1 ... p.


ugfitgenmodoutputtab.png (9973 bytes)

Figure 24.24: The Output Variables Tab


Previous Page | Next Page | Top of Page