Accurate proportional hazards estimation in iii-conditioned cases

Abstract

Under certain conditions, such as the presence of near colinearity in the data matrix, the computation of the Newton direction in proportional hazards estimation is subject to numerical difficulties that may contaminate the final estimates. This paper demonstrates an algorithm that yields incrased accurancy when computing the Newton direction in such cases and presents the results of a Monte Carlo trial comparing the algorithm with standard computations.

Keywords:

• Proportional hazards estimation
• Cox model
• collinearity
• ill conditioning
• iteratively reweighted least squares
• Newton' method

^∗This work supported in part by grant CAI1430 from the National Cancer Institute.

^∗This work supported in part by grant CAI1430 from the National Cancer Institute.

Notes

^∗This work supported in part by grant CAI1430 from the National Cancer Institute.