Multivariate hazard rates under random censorship (Q1368845)

The paper is devoted to the statistical estimation of unknown multivariate hazard functions on right randomly censored observations. Let \((X_1,\vec \delta_1),\dots, (X_n,\vec\delta_n)\) be random vectors, where \(X= (X_1,\dots, X_d)\), \(X_i= T_i\wedge C_i\), \(1\leq i\leq d\), is the random vector \(T=(T_1,\dots, T_d)\) having the distribution function \(F(x)= P(T>x)\), the censoring vector \(C= (C_1,\dots, C_d)\) is independent from \(T\) with the survival function \(G\), \(\vec \delta= (\delta_1,\dots, \delta_d)\) and \(\delta_i=\mathbf{1} (T_i\leq C_i)\), \(1\leq i\leq d\). In the paper for the unknown hazard function \(\lambda(x)= f(x)/F(x)\) the estimator \[ \widehat{\lambda}_I(x)= \sum_{i=1}^n K_n(x_I- X_{iI}) \frac{\mathbf{1}\{\vec\delta_{iI}= \text\textbf{1}_I, X_{iJ}> x_J\}} {\sum_{j=1}^n\mathbf{1}\{X_{jI}\geq X_{iI}, X_{jJ}> x_J\}} \] is proposed. The uniform consistency and asymptotic normality are obtained, and the asymptotic mean squared error is studied.