measure of influence
ranges from 0 (no influence) and 1 (complete influence over prediction)
average leverage value is (k+1)/n
k is the number of predictors in the model and n is the number of participants
Hoaglin & Welsch (1978) cases greater than twice the average are suspect
Stevens (2002) 3 times the average should be cut-off value
Was this helpful?
0 / 0