Academic journal article Journal of Travel and Tourism Research (Online)

Partial Least Squares Path Modelling (PLSPM): A New Direction for Research in Tourism and Hospitality

Academic journal article Journal of Travel and Tourism Research (Online)

Partial Least Squares Path Modelling (PLSPM): A New Direction for Research in Tourism and Hospitality

Article excerpt

Abstract

The use of partial least squares path modelling (PLSPM) has escalated in the areas of marketing, management, information systems, and organizational behaviour. Researchers in tourism and hospitality have to date been reluctant to use this approach, instead, focusing on covariance-based structural equation modelling (CBSEM) techniques conducted in Lisrel or AMOS. This article highlights the main differences between CBSEM and PLSPM and describes the advantages of PLSPM with regard to (1) testing theories and analyzing structural relationships among latent constructs; (2) dealing with sample size limitations and non-normal data; (3) analyzing complex models that have 'formative' and 'reflective' latent constructs; and (4) analyzing models with higher-order molar and molecular constructs. These advantages are put into practice using examples from a tourism context. The paper demonstrates the application of PLSPM in the case of destination competitiveness, and illustrates how this approach could enhance the theoretical and practical usefulness of tourism modelling. This paper also presents a step-by -step guide to PLSPM analysis, providing directions for future research designs in tourism. This presents valuable knowledge for researchers, editors, and reviewers with recommendations, rules of thumb, and corresponding references for appropriately applying and assessing structural models.

Keywords: Quantitative methods, structural equation modelling, partial least squares, tourism, rormative indicators.

Introduction

Structural equation modelling (SEM) is now widely used in business and tourism research (e.g., Babin et al., 2008; Assaker et al., 2010; Hallak et al., 2012). SEM allows for the analysis of latent variable(s) at the observation level (measurement/outer model), and to also test simultaneous relationships between latent variables at the theoretical level (structural/inner model) (Bollen, 1989). It can be used to examine research questions related to causal relationships among a set of latent factors each measured by one or more manifest [observed] variables within a single comprehensive method. There are two main approaches to SEM analysis 1) covariance-based SEM analysis (CBSEM)(Jöreskog, 1978, 1993), and 2) component-based, or partial least squares SEM (also referred as partial least squares path modellingPLSPM) (Wold, 1982, Esposito Vinzi et al., 2010).

The two approaches serve different research purposes. CBSEM typically employs a fall information maximum likelihood estimation process that yields parameter estimates that minimize the discrepancy between the implied and the observed covariance matrices. This approach examines the 'goodness-of-fit' of the computed covariance matrix from the model compared to the observed matrix from the data sample (Nunkoo and Ramkissoon, 2012). Partial Least Square Path Modelling (PLSPM) is an alternative SEM method that examines a network of relationships among latent variables (Wold, 1979). It is a partial information method that maximizes the explained variance of all dependent variables based on how they relate to their neighbouring constructs with a predictive purpose (Tenenhaus et al., 2005).

Tourism studies have mostly favoured CBSEM, using programs such as AMOS or LISREL, when dealing with structural models (e.g., Hallak et al., 2012). A review of papers published in the Journal of Tourism Management, Tourism Analysis, and Journal of Travel Research over the past five years yielded 196 studies that used SEM. Of those studies, only 29 used PLSPM, and this had only been in recent years (2011-2013); while the remaining 167 (85%) reviewed studies relied on CB-SEM. However, the CBSEM approach requires that the following assumptions to be met before the results can be validated: 1) multivariate normality of the data, 2) a large sample size, 3) the latent constructs are reflective (i.e., directional arrows progress from the constructs to the indicators), 4) the model is relatively simple with a limited number of latent variables, and 5) there is a strong theoretical basis for the model. …

Search by... Author
Show... All Results Primary Sources Peer-reviewed

Oops!

An unknown error has occurred. Please click the button below to reload the page. If the problem persists, please try again in a little while.