The influence of model size on the estimation accuracy of estimation methods in structural equation models with ordinal variables
(A2019-8102)

AUTHORS

Andreas Falke, Regensburg University

KEYWORDS

Structural equation modeling; Estimation method; Ordinal variables

ABSTRACT

Structural equation modeling has become a popular tool in marketing but a problem with the its application is that most researchers use ordinal answer scales in their surveys, whereas most of the popular estimation methods assume continuous variables. Estimation methods that can deal with ordinal scales have been published; however, the impact of model size on estimation accuracy of these methods has not been investigated. This study uses a Monte Carlo simulation to test, how well five different estimation methods (three that assume continuous variables, two that can deal with ordinal variables) perform under several model size constellations. Apart from estimation method and model size, sample size and two factors on construct validity are also considered. Results show that diagonally weighted least squares with a polychoric correlation matrix is among the best estimation methods most of the time, but, in several constellations, other estimation methods often perform equally well.