Cook’s Statistics
Cook's statistics measure the influence of each run on the fit. Values are usually between zero and 1, but can be larger. A larger value indicates that a run has more effect on the final fit than a run with a smaller value. Values larger than most of the others, for example larger than .5 or especially larger than 1, suggest that the run is unusually influential. The run might be an outlier, or at least should be examined to make sure it is accurate.
Di = ri2*hii / p*(1 - hii)
Di - i
th Cook's statistic
ri - i
th Studentized residual (See
Studentized Residuals)
hii - i
th entry on hat matrix diagonal, hat matrix H = X(X
TX)
-1X
T p - Number of coefficients (terms) in regression polynomial
References
■DS - Applied Regression Analysis, Draper and Smith (pg 210)
■MM - Response Surface Methodology - Process and Product Optimization Using Designed Experiments, Myers and Montgomery (pg 49)