Investigating consumer heterogeneity in multiblock L-shape data by combining SO-PLS regression and two-step procedure

Abstract

This manuscript proposes, for the first time, a new way to analyse L-shape data by combining the Sequential and Orthogonalized – Partial Least Square (SO-PLS) regression, and the Two-step PLS based Procedure (TSP) approach in the so called Sequential and Orthogonalized – Two Step Procedure (SO-TSP) approach. In a sensory and consumer study conducted in Norway, two blocks of sensory properties data (i.e. beginning, and middle-end of eating process), one block of consumer preferences liking data, and two blocks of consumer attributes data (i.e. health, and taste) were collected. Data were analysed in two steps. In step 1, consumer liking data are regressed onto sensory properties at the beginning and middle-end of eating process using the SO-PLS regression. Then, consumers were then segmented according to the Principal Components of Predictions (PCP) correlation. In step 2 consumer dummy variables, which represent two segments from step 1, are regressed onto the two consumer attributes blocks data (i.e. attitudes to taste and healthiness) using SO-PLS regression. These findings demonstrate that the SO-TSP approach is a useful and flexible tool to investigate and understand the complex relations among consumer segments, and the different blocks of dynamic sensory properties, and different types of consumer attributes data. Methodological implications and recommendations for academia, and future research avenues are discussed and outlined.