Improving prediction with POS and PLS consistent estimations: An illustration

Abstract

Recent advances (Dijkstra and Henseler, 2015a, 2015b) have introduced methods that provide consistent PLSc estimates. In parallel, Becker et al. (2013) propose a novel prediction oriented segmentation (POS) approach which by taking into account unobserved heterogeneity increases the predictive power with regard to the dependent variables. Hence, the main objective of this paper is to show how the complementary use of PLSc and POS can increase the overall predictive ability of the PLS approach. A concrete example, carefully following the presentation guidelines provided by Henseler et al. (2016), in a Moroccan context demonstrates the plausibility of such a proposal and concretely shows the existence of three different groups of people with different reactions toward counterfeiting. The stability of this segmentation is verified as well as the causal asymmetry of data. Managerial implications with respect to these three groups are highlighted, thanks also to a complementary importance–performance matrix analysis.