The component-based approach to Structural Equation Modeling (SEM) was initiated by Herman Wold under the name "PLS" (Partial Least Squares). Hwang and Takane (Psychometrika, 2004) have recently proposed a new component-based SEM method named Generalized Structured Component Analysis. Generally speaking, component-based SEM can be considered as a generalized data analysis approach to multiple tables connected by a network of "causal" relationships. As such, this approach is mainly used for scores computation and privileges a prediction oriented discovery process to the statistical testing of causal hypotheses. More specifically, PLS is a limited information two-step method essentially based on a set of interdependent simple and multiple OLS regressions both for the measurement and the structural model. The simplicity of the PLS algorithm makes it feasible also for (very) small samples. The recently proposed GSCA is, instead, a full information method that optimizes a global criterion. A few comparisons between covariance-based and component-based SEM have shown reasons in favour of one approach or the other depending on different factors such as, for instance, the nature of the model, the research objective, the sample size, the definition of latent variables by means of reflective or formative manifest variables, the estimation and practical meaning of factor scores. With reference to component-based SEM and, specifically, to PLS Path Modeling, we focus on current important issues such as the optimization of a criterion, the measurement model misspecification, the treatment of formative relationships between manifest and latent variables, the estimation and the intepretation of scores in presence of strongly correlated latent variables, the possibility of constraining parameter estimates as well as on some other open issues representing interesting themes for current and future researches.
ESPOSITO VINZI, V. (2008). Current Issues and Future Challenges in Component-based Structure. In: 7th International Conference on Social Science Methodology, RC33 - Logic and Methodology in Sociology.