A data-driven smoothing parameter selection for robust nonparametric regression /

Nonparametric regression methodology provides flexible tools

Bibliographic Details
Main Author: Wang, Shaohung
Format: Thesis Book
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1995.
Subjects:
Online Access:http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=742164431&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD
Description
Summary:Nonparametric regression methodology provides flexible tools
for analyzing unknown regression relationships. In the
application of nonparametric methods, it is crucial to select
a proper smoothing or bandwidth parameter which controls the
smoothness of the resulting function estimate. This
dissertation presents a procedure to estimate the optimal
bandwidth for bivariate local linear regression when the data
are contaminated by a heavy-tailed error distribution. First,
a method for robust estimation of a bivariate regression
function is proposed, and the asymptotic properties of the
estimator are derived. Second, an asymptotically efficient
bandwidth selection rule, the "plug-in" method, is proposed
to estimate the optimal bandwidths from the data. Third, a
nonparametric estimator of residual variance is proposed. It
basically is an application of local least squares
regression. A minor modification to robustify the variance
estimate is also discussed. These methods are demonstrated
and evaluated through various simulation studies. It is
found that the "plug-in" rule improves upon the cross-
validation procedure in various ways: (1) better relative
rate of convergence; (2) computational saving from not
requiring numerical minimization; (3) reliably good
performance for smooth regression surfaces in simulations.
Item Description:Vita.
"Major Subject: Statistics".
Physical Description:viii, 76 leaves : illustrations ; 28 cm.
Issued also on microfiche from University Microfilms Inc.
Bibliography:Includes bibliographical references.