Best linear unbiased estimation and prediction in spatial models with measurement error /

Natural resources data are often both longitudinal and

Bibliographic Details
Main Author: Lin, Chii-Dean
Format: Thesis Book
Language:English
Published: [Place of publication not identified] : [publisher not identified] ; 1997.
Subjects:
Online Access:http://proxy.library.tamu.edu/login?url=http://proquest.umi.com/pqdweb?did=736824281&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD
Description
Summary:Natural resources data are often both longitudinal and
spatial in nature. In addition, the predictors of the
outcome of interest may be measured with error. For example,
weather station data provide useful predictors in
dendrochronology models, but the weather stations are not at
the tree plot locations. Thus, the weather variables can
only be predicted from the nearby weather stations. In this
dissertation, we consider measurement error models where the
mismeasured predictors, X, are related to the observed
measurements, W, through W = []0 + []1 X + U or X = []0 + []1 W
+ U. Within these settings, we discuss best linear unbiased
estimation and prediction and demonstrate that using E (X[]W)
in place of X produces optimal results. Asymptotic inference
will also be examined when the amount of information on the
observed proxy of X, W, goes to infinity. Small sample
comparisons between these estimators and competitors are also
made. The limiting behavior of a spatial predictor as the
amount of data collected increases to infinity within a fixed
domain is known as infill asymptotics (Cressie (1991)).
Whiting (1995) investigated the infill-asymptotic behavior
of best linear unbiased predictor (BLUP) and estimated best
linear unbiased predictor (EBLUP) by examining mean squared
prediction error (MSPE). We first consider an extension to
temporal-spatial models. Under this setting, we examine the
behavior of (E)BLUPs when the amount of temporal and spatial
data tends to infinity. Various configurations are studied
under this setting. Additionally, we consider the use of
(E)BLUPs from one spatial model as predictors in a second
model. We examine the asymptotic behavior of the best linear
unbiased estimator in the second model when the MSPE of
(E)BLUP in the first model tends to 0. Simulation studies
are presented to illustrate the proposed results. Finally,
we apply these techniques to a problem in dendrecology. Data
from plots of trees throughout Mississippi and Louisiana were
collected through a grant from the USDA Southern Forest
Experiment Station, Pineville, LA. Weather data were also
provided by USDA Forest Service from 41 weather stations in
Mississippi and Louisiana. A tree growth model is developed
and significant variables are suggested and discussed.
Item Description:Vita.
"Major Subject: Statistics".
Physical Description:xiv, 111 leaves : illustrations ; 28 cm.
Issued also on microfiche from University Microfilms Inc.
Bibliography:Includes bibliographical references: pages 92-95.