On the role of partial least squares in path analysis for the social sciences
R. Dennis Cook , Liliana Forzani b,
- School of Statistics, University of Minnesota, Minneapolis, MN 55455, United States of America
- Facultad de Ingeniería Química, UNL. Researcher of CONICET, Santa Fe, Argentina
A B S T R A C T
We describe the current and potential future roles for partial least squares (PLS) algorithms in path analyses, guided by recent advances in envelope theory. After reviewing the present debate and establishing a context, we conclude that, depending on specific objectives, PLS methods have considerable promise, but that their full potential, while reachable, is not now being realized. The future developments necessary for achieving their full potential in the social sciences are clear and doable, albeit demanding. A critique of covariance-based structural equation modeling (CB-SEM), as it relates to PLS, is given as well. Technical details are available in the appendix.
1. Introduction
Path modeling is a standard way of representing social science theories. The validation of a path model often involves inference about concepts like ‘‘customer satisfaction’’ or ‘‘competitiveness’’ for which there are no objective measurement scales. Since such concepts cannot be measured directly, multiple surrogates, which may be called indicators, observed or manifest variables, are used to gain information about them indirectly. One role of a path diagram is to provide a visual representation of the relationships between the concepts represented by latent variables and the indicators. A fully executed path diagram is in effect a model that can be used to guide subsequent analysis, rather like an algebraic model in statistics. A path model is commonly translated to a structural equation model (SEM) for analysis