Abstract
Recently Chen and Chang (1999) proposed two approaches, the discrete and sieve methods, for regression-transformation model with heteroscedastic and right censored data. In their paper, they only compare with two methods under situation that the error term is standard extreme value distributed and there is two sample data. In our paper, we would like to compare these two approaches with more approaches, and simultaneously we also consider the situation that the error term is standard logistic distributed. By these more extensive simulative comparisons, we research the advantage of using these two methods when regression-transformation models are appropriate.2 Estimation methods 2.1 The estimation method of Cheng et al. (1995) 2.2 The maximum partial likelihood method 2.3 The estimation method of Dabrowska & Doksum (1988a) 2.4 The ROC method 2.5 The estimation methods of Chen and Chang (1999) 2.5.1 The discrete method 2.5.2 The sieve method3 Simulations on two-sample case 3.1 σ-proportional hazard model 3.1.1 γ is known and equal to zero 3.1.2 γ is unknown 3.2 σ-proportional odds model 3.2.1 γ is known and equal to zero 3.2.2 γ is unknown4 Simulations on one-sample case 4.1 σ-proportional hazard model 4.1.1 γ is known and equal to zero 4.1.2 γ is unknown 4.2 σ-proportional odds model 4.2.1 γ is known and equal to zero 4.2.2 γ is unknown5 DiscussionReferences