ASSESSING REFLECTIVE MODELS IN MARKETING RESEARCH: A COMPARISON BETWEEN PLS AND PLSc ESTIMATES

Headnote

ABSTRACT

The present study aims at contributing to the growing discourse on analytical methods in marketing research by highlighting the use of Consistent Partial Least Squares (PLSc) estimation to assess reflective models used in marketing literature. Specifically, it demonstrates the significance of using PLSc and compares it with the traditional PLS. The results show that PLSc is more robust than traditional PLS in estimating convergent validity and path coefficients, and yields better power - coefficient of determination (R²) and effect size (f²). It is also found that PLSc generates better holdout results than traditional PLS. This study complements and extends prior research on PLSc, and subsequently serves as a resource for marketing researchers who use variance-based approach in their research. Implications, guidelines and future research directions are discussed.

Keywords: Consistent Partial Least Squares; Traditional PLS; Path Modeling; SEM; Marketing

1.INTRODUCTION

Structural Equation Modeling (SEM) technique has become one of the most powerful statistical techniques across various disciplines in recent years. An increasing number of researchers have begun to recognize its ability to model latent variables, taking into account the various forms of measurement errors, and test the underlying theories in a structural manner (Pakpahan et al., 2017). Thus, they stand to benefit from this technique by acquiring more reliable and valid findings to answer respective research questions with accurate estimation.

There are two types of SEM techniques, namely covariance-based SEM (CB-SEM) (Jöreskog and Wold, 1982) and variance-based SEM (VB-SEM) (Lohmöller 1989; Wold, 1982; Hair et al., 2016; Memon, Ting, Ramayah, Chuah, & Cheah, 2017). Despite complementing each other, it is necessary to know that they differ greatly in their statistical methods, and have distinct goals and requirements (Hair et al. 2011; Henseler et al. 2009). In general, CB-SEM estimates the model parameters by means of the empirical covariance matrix. It is more often the preferred method if the hypothesized model consists of one or more common factors. VB-SEM, however, creates proxies as linear combinations of observed variables, and uses them to estimate the parameters. It is usually the method of choice if the hypothesized model contains composites.

Among the VB-SEM methods available to date, partial least squares (PLS) path modelling is regarded as the "most fully developed and general system"...