Using LISREL V to Perform a Covariance Structure Analysis of a Tripartite Model of Attitude

Abstract

Purpose of study: The purpose of this paper was to investigate a multivariate statistical analysis method used in a scientific article. The article was written by Breckler (1984) and published in the Journal of Personality and Social Psychology. Sources of evidence: According to Breckler, affect, behavior, and cognition are three hypothetical, unobservable classes of responses to attitude, a so-called tripartite model. Breckler tested the validity of this tripartite model by using the software program LISREL V. In his two studies, Breckler found that the results indicated that affect, behavior, and cognition were distinguishable components of attitude. Main arguments: However, there are some considerations that Breckler should have taken into account in his article. The first is the fact that high intercomponent correlations do not necessarily follow from the tripartite view. These three components may operate partially or even completely independently. Besides, observed measures may assess primarily the cognitive component, leading to an inflated estimate of intercomponent consistencies. The second consideration is that the NFI fit index is a widely used Type 2 fit index, but it is currently not recommended because it is affected by sample size and does poorly for small samples. The third consideration is that Breckler should have performed an analysis using power analysis of covariance structures. Conclusion: One conclusion in this paper is that a better focus on power and sample size when using structural equation analysis is something that Breckler should have taken into consideration.