Page 357 - Contributed Paper Session (CPS) - Volume 7
P. 357
CPS2130 Abdul-Aziz A. Rahaman et al.
Comparative performance of estimation
maximisation and other known methods of
residual estimators in structural equation models
1
2
Abdul-Aziz Abdul-Rahaman , Albert Luguterah , Bashiru Imoro Ibn Saeed
1
1 Kumasi Technical University
2 University for Development Studies
Abstract
As the field of methodology has advanced, alternative estimation methods of
residuals have been developed including regression method, Bartlett’s
method and Anderson-Rubin method. Somehow, their performance have
experienced some level of challenges. Therefore, this study incorporated the
estimation maximization approach and compared it with the other methods
to identify the efficient method in estimating residuals under the structural
equation model framework. The results showed that the strength of the
existing methods are the weaknesses of EM method, and vice versa. It was
therefore found from the comparative model fits information that the
Bartlett’s based method gave better residual parameter estimates over the
regression-based method and the Anderson Rubin based method. However,
the EM method gave better residual parameter estimates than the other three
existing methods (i.e. the regression, Bartlett’s and the Anderson Rubin based
methods).
Keywords
Estimation maximization, Estimators, Structural equation modelling,
Maximum likelihood
1. Introduction
Structural equation models (SEM) have been successfully utilised in
different research areas, including educational studies (Miranda & Russell,
2011; Saçkes, 2014), clinical psychology (Little, 2013; Löfholm et al., 2014),
developmental psychology (Geiser et al., 2010), organizational studies
(Binnewies et al., 2010; Kiersch & Byrne, 2015; Mahlke et al., 2016), and multi-
trait multimethod (MTMM) analysis (CarreteroDios et al., 2011). Approaches
to SEM estimation may be described as covariance-based (e.g., ML) and
component-based (e.g., PLS, GSCA), or as frequentist (e.g., ML, PLS, GSCA) and
Bayesian (e.g., MCMC). Simply put, the primary distinction between
covariance- and component-based estimation is that the former is suited to
model testing and the latter is better suited to explaining variance and making
predictions (Hulland et al., 2010; Tenenhaus, 2008). Although it is difficult to
know whether or not theoretical models are specified correctly in applied
research, simulation-based research has illustrated the impact of
344 | I S I W S C 2 0 1 9