删失数据下回归函数的最近邻估计
Nearest Neighbor Regression Function Estimation with Censored Data
-
摘要: 设(X, Y)是一随机向量且变量Y的均值存在。假定Y被另一分布G的随机变量t删失,仅能观察到不完全数据(Xi,Yi Λti,δi),i=1,2,…n,其中Yi Λti=min(Yi,ti),δi=I(Yi ≤ ti)。为了给出回归函数m(x)=E(Y|X)的估计。文中使用了Stute提出的最近邻型回归估计,并给出了该估计的强相合性结果。Abstract: Let (X, Y) be an R2 valued vector with E|Y| <∞. Yi are censored by random variables ti with d.f.G and the available observations are of the form (zi, δi, Xi), 1≤i≤n, where Zi = Yi Λti andδi = I(Yi ≤ ti). To estimate the regression function m(x0) = E(Y|X =x0),we propose a Stute’s(1984) type estimator, i.e. the nearest neighbor regression estimator, when the data subject to cellsoring. It is established the strong consistency of the nearest neighbor regression estimator.