分位数回归在竞争失效加速寿命试验统计分析中的应用

Based on accelerated life tests (ALT) data, inferences on quantiles of the lifetime distribution at the use condition are obtained via an assumption of a specific working model. But in engineering practice, model misspecification can result in significant estimation bias. In order to solve this problem, a statistical analysis approach based on quantile regression is proposed to estimate quantiles of the lifetime distribution with competing causes of failure. Quantile regression is distribution-free and more flexible in modeling life-stress relations. Full consideration is given to the incompleteness of test data, which is due to Type-II progressive censoring as well as competing risks. The martingales based on cause specific hazards are used to construct unbiased estimating equations for the quantile regression model, and the solution of equations is equivalent to search for minimizations of convex functions. By using perturbation resampling approach, the interval estimations are presented. Finally, the Monte Carlo method is used to evaluate the performance of the proposed method.