Influence contours in linear regression (Q1424613)

A visual method of studying influence measures in linear regression is proposed. The authors suggest a contour plot approach based on adding new observations to the existing data sets (instead of omitting observations in usually used procedures) for investigating the behaviour of the well-established influence measures. Influence contours are defined via the equation: ``Influence of the new observation \((y,x_1,\dots,x_{p-1})=c\)''. A solution to this equation in \(y\) defines an influence contour. Examples of contour plots are given for four types of influence measures including, particularly, Cook's distance and the determination ratio. The influence contours are shown to be a useful tool in investigating and comparing the regions of influence for various influence measures.