Year: 2008
Type: dissertation
Abstract: New structural equation models (SEMs) for the analysis of multitrait- multimethod-multioccasion (MTMM-MO) data are presented. The definition and psychometric analysis of the models is based on stochastic measurement theory (Steyer, 1989; Suppes & Zinnes, 1963). The applicability of the new models is evaluated through a reanalysis of a real MTMM-MO data set and a Monte Carlo simulation study. In the introduction, an overview of existing SEMs for cross- sectional MTMM data is provided. The Correlated Trait-Correlated Uniqueness (CT-CU; Marsh, 1989), Correlated Trait-Correlated Method (CT-CM; Widaman, 1985), Correlated Trait-Uncorrelated Method- (CT-UM), and Correlated Trait- Correlated (Method Minus One)- [CT-C(M–1); Eid, 2000] models are briefly reviewed and compared. It is concluded that the CT-C(M–1) model for multiple indicators per trait-method unit (Eid et al., 2003) is one of the most useful models currently available for cross-sectional MTMM data. Subsequently, three different SEM approaches to the analysis of longitudinal MTMM data are discussed: Cole and Maxwell’s (2003) multi-occasion CU model; Burns, Walsh, and Gomez’ (2003) Correlated State-Correlated Method model (Burn & Haynes, 2006); and Courvoisier’s (2006) multi-method latent state trait model (Courvoisier, Nussbeck, Eid, Geiser, & Cole, 2007). It is shown that a general measurement model for analyzing MTMM data and for analyzing change in MTMM-MO studies has not yet been developed. Subsequently, basic principles of classical test theory (Steyer, 1989, Steyer & Eid, 2001) and latent state theory (Steyer, 1988; Steyer, Ferring, & Schmitt, 1992) are reviewed. These concepts are used in the formulation of the new MTMM-MO models. Afterwards, two versions of the Correlated State-Correlated (Method Minus One) [CS-C(M–1)] model are introduced. These models represent a combination of Eid et al.’s (2003) multiple indicator CT-C(M–1) model and the correlated state model for mono-method data (Steyer et al., 1992). A detailed psychometric analysis of the CS-C(M–1) models is provided. It is then shown how CS-C(M–1) models can be extended to latent difference models to study inter-individual differences in intra-individual change over time. The so-called CS-C(M–1) change model represents a multimethod extension of Steyer, Eid, and Schwenkmezger’s (1997) true change model (Steyer, Partchev, & Shanahan, 2000). The CS-C(M–1) change model can be used to study change in different methods simultaneously and to determine the degree of convergent validity and method-specificity of observed and latent change scores. In the empirical part, a 3-Step procedure for analyzing, testing and selecting an appropriate CS-C(M–1) model is presented and the applicability of the new models is investigated in a reanalysis of a MTMM-MO data set and a Monte Carlo simulation study. The results show that the new models are useful to analyze the complex structure of a MTMM-MO matrix obtained from multiple indicators per construct-method-occasion unit. In the final part, advantages and limitations of the models as well as detailed guidelines for potential users are discussed. Furthermore, the new models are compared with already established methods for analyzing MTMM-MO data and directions for future research are pointed out.
Author: Geiser, Christian
Language: English
Field: Decision Sciences
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W1575099107