Fit Analyses |
For linear models, the partial leverage plot for a selected explanatory variable can be obtained by plotting the residuals for the response variable against the residuals for the selected explanatory variable. The residuals for the response variable are calculated from a model having the selected explanatory variable omitted, and the residuals for the selected explanatory variable are calculated from a model where the selected explanatory variable is regressed on the remaining explanatory variables.
Let X[j] be the n×(p-1) matrix formed from the design matrix X by removing the jth column, Xj. Let ry[j] be the partial leverage Y variable containing the residuals that result from regressing y on X[j] and let rx[j] be the partial leverage X variable containing the residuals that result from regressing Xj on X[j]. Then a partial leverage plot is a scatter plot of ry[j] against rx[j]. Partial leverage plots for two explanatory variables are illustrated by Figure 39.26.
Figure 39.26: Partial Leverage Plots
In a partial leverage plot, the partial leverage Y variable ry[j] can also be computed as
For generalized linear models, the partial leverage Y is also computed as
Two reference lines are also displayed in the plots. One is the horizontal line of Y = 0, and the other is the fitted regression of ry[j] against rx[j]. The latter has an intercept of 0 and a slope equal to the parameter estimate associated with the explanatory variable in the model. The leverage plot shows the changes in the residuals for the model with and without the explanatory variable. For a given data point in the plot, its residual without the explanatory variable is the vertical distance between the point and the horizontal line; its residual with the explanatory variable is the vertical distance between the point and the fitted line.
Copyright © 2007 by SAS Institute Inc., Cary, NC, USA. All rights reserved.