Regression \(M\)-estimators with doubly censored data (Q1383098)

The paper deals with the estimation of the parameters \(\alpha,\beta\) in the regression model \(X_i= \alpha+ T_i\beta +e_i\), where \(X_i\) are the lifetime random variables and \(T_i\) are the (univariate) covariables which are assumed to be iid with d.f. \(F_T\). Only a doubly censored version of \(X_i\) is observed. Robust \(M\)-estimators \((\widetilde \alpha, \widetilde\beta) =\tau (\widetilde F_n)\) are solutions of \[ \iint \psi (x-\theta_1 -\theta_2t) d\widetilde F_n (x,t)=0 \quad \text{and} \quad \iint t\psi (x-\theta_1 -\theta_2t) d\widetilde F_n (x,t)=0, \tag{1} \] where \(\widetilde F_n\) is a suitable estimator of the bivariate d.f. of \((X_i,T_i)\) taking the censoring into account. A Campbell-type estimator \(\widetilde F_n\) is constructed and studied. The paper establishes strong consistency and asymptotic normality of \(\widetilde F_n\). The functional \(\tau (\cdot)\) defined by (1) is shown to be Hadamard differentiable and the asymptotic normality of \(\tau (\widetilde F_n)\) is obtained. The computation of the proposed \(M\)-estimator is discussed and an application is given.