A NOTE ON LARGE-SAMPLE PROPERTY OF COOK'S DISTANCE

Motivated by influence curve for the parameter vector, Cook's distance has several interesting features and thus plays an important role in linear regression diagnostics. Although Cook's distance is usually compared with various percentiles of an F-distribution, it sometimes puzzles statistics users since F-distribution is not the correct sampling distribution of Cook's distance. In this paper, we study the large-sample property of Cook's distance under the assumption of multivariate normality of predictor variables. Also we construct a table of critical values of Cook's distance as a diagnostics tool in linear regression.