Original scientific paper

Estimation of nonlinear latent structural equation models using the extended unconstrained approach

Augustin Kelava orcid.org/0000-0001-6053-0415 ; Department of Psychology, Technical University Darmstadt
Holger Brandt ; Department of Psychology, Goethe University Frankfurt

Abstract

In the past two decades, several approaches for the estimation of single interaction effects in latent structural equation models (SEM) have been published (for an overview: Marsh, Wen, & Hau, 2006; Schumacker & Marcoulides, 1998). Recent literature has begun to consider nonlinear models involving simultaneous interaction and quadratic effects (e.g. Klein & Muthén, 2007; Lee, Song, & Poon, 2004; MacCallum & Mar, 1995). Marsh, Wen, and Hau (2004) developed an unconstrained product indicator approach for the estimation of single interaction effects with robust properties when distributional assumptions are violated. In this article, first, we give a brief overview on the approaches for the estimation of nonlinear SEM. Second, we describe the extended unconstrained approach for the simultaneous estimation of latent interaction and quadratic effects (Kelava, 2009; Moosbrugger, Schermelleh-Engel, Kelava, & Klein, 2009). Third, we apply the extended unconstrained approach to data from work and stress research using the freely accessible sem package (Fox, 2006) in R (R Development Core Team, 2008) and compare it with LMS (Klein & Moosbrugger, 2000) which is implemented in the commercial Mplus (Muthén & Muthén, 2007) software.

Keywords

nonlinear structural equation models; estimators; quadratic; interaction; unconstrained approach

Hrčak ID:

70644

URI

https://hrcak.srce.hr/70644

Publication date:

19.12.2009.

Visits: 3.050 *