B-SPLINE ESTIMATION FOR PARTIALLY LINEAR REGRESSION MODELS WITH HETEROSCEDASTICITY
尤进红,陈歌迈
DOI: https://doi.org/10.3321/j.issn:1000-8134.2004.05.014
2004-01-01
Abstract:For partially linear regression model with heteroscedastic error variances yij = x'ijβ+ g(tij) + eij, i = 1,2,…,k, j = 1,2,…,ni, and sum from i=1 to k=n where yij's are responses, β= (β,…,βP)' is an unknown parameter vector, g(.) is an unknown function over R, Xij = (xij,…,XijP)' and tij ∈[0,1] are known and nonrandom design points, and eij's are independent errors with mean 0 and variance σi2 which may be different. Based on the nonparametric component g(.) approximated by a B-spline series, a semiparametric generalized least squares estimator (SGLSE) of the parametric component βis constructed. The asymptotic distribution is established under some moment conditions on the error distributions. Most of the error distributions encountered in practice satisfy these moment conditions. A consistent estimator of the asymptotic covariance matrix is also given. Moreover, the B-spline estimation of the nonparametric component is also considered. The large sample properties of these estimators are derived for increasing k, assuming the numbers ni in the groups from a fixed sequence. Based on these asymptotic results asymptotically valid the test statistics and confidence intervals for parametric component and nonparametric component can be constructed.