Modelling causal error structures in longitudinal data /

The purpose of the present research was to investigate how

Bibliographic Details
Main Author: Sivo, Stephen Anthony
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=739891601&sid=1&Fmt=2&clientId=2945&RQT=309&VName=PQD
Description
Summary:The purpose of the present research was to investigate how
well moving average and autoregressive-moving average models
that specify correlated measurement errors fit longitudinal
data compared to autoregressive, quasi-simplex and one-factor
models. Sufficient evidence is available from theoretical
and applied studies to encourage testing for stochastic
models other than the predominantly used quasi-simplex
models. Three criteria for successful detection of
underlying structures were supported: fit, propriety, and
parsimony. Sometimes fit indices alone were sufficient in
ruling out models as representative of dynamic occurring in a
given data set. Yet, often more than one model fit a given
data well. In such cases the propriety of the solution
estimated for the models first required examination.
Improper solutions were recognized when fit indices or
estimated parameters were out of bounds. When neither fit
nor propriety were in question, competing models were
eliminated as contenders on the basis of parsimony.
Conclusions concerning model fit included the following: (1)
when testing for one stochastic process hypothesized to be
present in a longitudinal data set, testing for other
processes as well prevents erroneous conclusions, because
incorrect models often fit other processes well enough to be
deceiving; (2) when measurement error correlations are
relatively high, the fit of other models, particularly the
quasi-simplex and the autoregressive model, should first be
compared to the fit of a moving average model; (3) analysis
of simulation data disclosed that the moving average model
exhibits measurement error correlations when fit to some
autoregressive data sets, although the moving average model
fit indices are expected to be lower than those for an
autoregressive model; and (4) the one-factor model may be a
legitimate alternative to stochastic models and the
assumption longitudinal data sets always contain measurement
error correlations is false. In summary, when evaluating
longitudinal data, all five models should be considered and
compared.
Item Description:Vita.
"Major Subject: Educational Psychology".
Physical Description:xiv, 150 leaves ; 28 cm.
Issued also on microfiche from University Microfilms Inc.
Bibliography:Includes bibliographical references: pages 121-126.