Abstract: Graphical methods are proposed for studying the contributions of selected predictors to regression problems. By developing low dimensional distributional index functions based on sliced inverse regression, problems with many predictors can be addressed. It is shown that added variable plots can fill this role under certain conditions, but that they may generally overestimate predictor contributions. Scatterplot brushing plays a basic role in the methodology.
Key words and phrases: Added variable plots, adjusted variable plots, conditional response plots, scatterplot brushing, scatterplot matrices, sliced inverse regression.