Predictions from Partial Least Squares Models

Abstract

This chapter seeks to introduce partial least squares (PLS) practitioners to the generation and evaluation of predictions from their path models, both as a means of validating the practical usefulness of their models and for forecasting future outcomes.Studies in tourism and hospitality currently offer strong research utility by explaining how personal, regional, or commercial factors generally relate to positive outcomes for the industry. However, the practical utility of our studies is limited to general policy-making suggestions based on the significance of antecedent factors. Could we use our models to predict, say, a new subject’s potential behavior in the future, or forecast how a particular region or commercial interest will fare in the coming year given new data? In addition, before we even have access to data on new subjects, can we use our existing samples to validate that our current models have the potential to make accurate predictions using new data? Being able to address these types of prediction problems would greatly magnify the practical utility of our models and better serve the immediate and quantifiable needs of governments, organizations, businesses, services, and persons involved in the tourism industry. Current approaches and metrics in PLS path modeling (PM) largely cannot answer these prediction-oriented questions about specific cases or scenarios, and are largely relegated to making highly generalized prognostications based on path significances.We believe that predictive PLS is one of the emerging and promising directions in PLS-PM. Although predictive techniques for PLS are still at a nascent stage of development, we hope this chapter brings you up to speed on the latest developments, informs you of predictive practices you can employ today, and gives you a foundation for following future developments in this exciting new direction for PLS-PM