Structural Equation Modeling in Information Systems Research Using Partial Least Squares

Abstract: Empirical studies that use structural equation modeling (SEM) are widespread in information systems research. During the last few years, the component-based approach partial least squares (PLS) for testing structural models has become increasingly popular. At the same time, this approach's limitations have become a greater concern. Some researchers even suggest using alternative approaches that are considered superior to PLS. However, we believe that PLS is an adequate choice if the research problem meets certain characteristics and the technique is properly used. Thus, the intention of this article is to resolve potential uncertainties that researchers intending to use PLS might have. Consequently, we provide a nontechnical overview of PLS and outline the ongoing discourses on SEM in general and the PLS approach in particular. Furthermore, we present a basic framework for empirical research applying PLS as well as a detailed explanation of the different process steps. Finally, examples of information systems research using PLS are summarized to demonstrate its beneficial application and the appropriateness of the proposed framework. This article can serve as a helpful guide for inexperienced researchers applying PLS for the first time, but also as a reference guide for researchers with a better understanding of the field. Keywords: structural equation modeling, partial least squares, information systems research I. INTRODUCTION The information systems (IS) discipline examines socioeconomic systems that are characterized by the interplay between hardware and software on the one hand, as well as individuals, groups, and organizations on the other. For example, technology adoption, acceptance, and success, as well as the conditions under which these can be achieved are typical issues that are addressed by IS research. These research fields are similar in that their investigation requires the researcher to cope with constructs such as the beliefs, perceptions, motivation, attitude, or judgments of the individuals involved. These constructs are usually modeled as latent variables (LVs) that can be measured only through a set of indicators. Structural equation models describe the relationships between several of these LVs. A number of algorithms and software programs are available to estimate their relationships based on a dataset. Among these algorithms, the partial least squares (PLS) algorithm has become increasingly popular both in IS research and in other disciplines such as marketing (Albers 2010; Henseler et al. 2009) or strategic management (Hulland 1999). However, reminders of this approach's limitations have recently become more prominent. Consequently, researchers opt for a more careful application of PLS. Especially its statistical power at small sample sizes, the overall model fit, as well as the misspecification of measurement models have been the focus of recent discussions. To resolve the uncertainties that researchers intending to use PLS might have, we investigate current discourses on the PLS approach. In the following sections, we: * demonstrate the increasing popularity of PLS in the IS research community * outline the PLS approach by providing a nontechnical overview and reflecting on the ongoing discussion on structural equation modeling (SEM) in general and on PLS in particular * discuss differences between PLS and covariance-based approaches * present a basic framework for empirical research applying the PLS approach * provide examples of its beneficial application in IS research As a result, this paper helps SEM beginners and advanced researchers to make an informed decision about whether to use SEM or other alternative approaches to SEM. Even more, we present up-to-date recommendations on how to apply PLS appropriately. Section II demonstrates the increasing popularity of PLS for SEM in IS research by conducting a systematic review of literature that appeared in two prestigious IS journals during the last fifteen years. …