Influence in Linear Hazard Models
研究了线性风险模型中单个观测对Aalen累积风险估计量的影响，推导出类似Cook距离的病例删除诊断量，并与比例风险模型下的影响函数比较，发现两种模型对异常值的稳健性各有优劣。
Linear hazard models have been considered recently as alternatives to the well known Cox proportional hazards model for the regression analysis of censored survival data. In this paper we examine how individual observations influence Aalen's cumulative hazard estimator for such models. A diagnostic for the effect of case deletion, akin to the Cook distance, is obtained in closed form and its use in detecting outlying or unusual observations is illustrated. The influence function for the estimator is then obtained and examined. A comparison is made with the corresponding influence function for the usual maximum partial likelihood estimator under proportional hazards. Results indicate that neither model is uniformly more robust to unusual observations. The use of the influence function in estimating variance is examined.