Exploring regression structure with graphics

Abstract

Added variable plots, conditional independence, Dimension reduction subspaces, Graphical regression, Three-dimensional scatterplots,

Identifying Regression Outliers and Mixtures Graphically

“…The numerical methods include standard fitting methods like ordinary least squares (OLS), SAVE (Cook and Lee 1999;Cook and Weisberg 199 l), SIR (Li 199 l), and principal Hessian directions (Cook 1998a;Li 1992). Graphical regression (Cook 1992(Cook , 1994a(Cook , 1996Cook and Wetzel 1993) is primarily a graphical method. It is based on decomposing the central subspace into a direct sum of lower-dimensional subspaces that can be estimated visually from appropriately chosen two-or three-dimensional plots.…”

An Introduction to Regression Graphics

“…If (6) is false for all values of then the structural dimension is at least 2, and we could select two other predictors to combine. This methodology was developed by Cook (1992aCook ( , 1994aCook ( , 1996Cook ( , 1998b), Cook and Weisberg (1994a), and Cook and Wetzel (1993), where it is shown that various 3D plots, including 3D added-variable plots, can be used to study the possibilities associated with (6) and to estimate visually. Details are available from Cook (1998b).…”

Discussion of the Paper by Ripley

“…For example, when do nl nz nodes in a single layer perform better or worse than n l nodes in each of nz layers? (b) Cook (1993a,b) and Cook and Wetzel (1993) are actively exploring the idea of graphical regression.…”