Estimating the variance in a pseudo-observation scheme with competing risks

摘要

The Huber-White sandwich variance estimator of the variance of estimates from a regression using jack-knife pseudo-observations from the Aalen-Johansen estimator in a competing risks situation is known to be biased. In this paper, an expression of the asymptotic bias of the Huber-White variance estimator is found, and the Huber-White variance estimator is seen to be biased upwards. An alternative variance estimator, obtained by plugging in empirical values in the true variance expression, is studied by simulation, and its performance is compared with the performance of the biased Huber-White variance estimator. On the basis of the simulation study, recommendations of its use are given. The alternative variance estimator works well in large samples but, because it estimates the asymptotic variance, may be less useful on small samples.