Una Modificación a la Distancia de Cook

Abstract

A modification of the classical Cook´s distance is proposed in this paper, being based on the distance of widespread Mahalanobis in the context of multivariate normal linear regression model. Further more, the exact distribution of a pivotal type statistics based on this generalized Mahalanobis distance is established, providing critical points for the identification of outliers in data points. The procedure is illustrated with an example, in the case of multiple and multivariate linear regression.