Global measures of local influence for proportional hazards regression models.

Center for Health Studies, Group Health Cooperative, Seattle, Washington 98101-1448, USA.

Cox proportional hazards regression models have been extended to multiple failure times and other situations by using a robust covariance matrix instead of the usual inverse of the information matrix. We modify an existing measure of individual influence by substituting the robust covariance matrix for the inverse of the information matrix. This provides a scalar measure of influence with known mean and bounded range on the interval (0,1). The measure is applicable to marginal multiple failure time models. Two examples are presented.

PMID: 9290233 [PubMed - indexed for MEDLINE]