Premium
Beyond principal component analysis (PCA) of product means: Toward a psychometric view on sensory profiling data
Author(s) -
Dettmar Burkhard,
Peltier Caroline,
Schlich Pascal
Publication year - 2020
Publication title -
journal of sensory studies
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.61
H-Index - 53
eISSN - 1745-459X
pISSN - 0887-8250
DOI - 10.1111/joss.12555
Subject(s) - principal component analysis , profiling (computer programming) , sensory system , interpretability , artificial intelligence , pattern recognition (psychology) , curse of dimensionality , computer science , sensory analysis , perception , psychology , data mining , mathematics , statistics , cognitive psychology , operating system , neuroscience
Principal component analysis (PCA) has its origin in psychology, where it was developed as a psychometric tool to measure latent variables of human cognition, personality, or behavior. This psychometric approach is also suitable to measure human perception based on sensory profiling data. To do so, we apply the PCA to a matrix that maintains the individual panelist's judgments, the matrix structure is in line with the “Tucker‐1 common loadings model.” Our approach (“Tucker‐1 PCA”) differs from the routine method of analyzing sensory profiling data, where PCA is applied to the matrix of mean scores of the product‐by‐attribute table (“Means‐PCA”). This article discusses the specific properties of the Tucker‐1 PCA and compares it to the Means‐PCA via a meta‐analysis on 422 datasets from Sensobase, a collection of sensory profiling studies. Tucker‐1 PCA provides advantages over Means‐PCA in terms of dimensionality, interpretability, and replicability of the factor structures. Practical Applications Tucker‐1 PCA is an easily applicable variant of PCA for sensory profiling data. Like Means‐PCA, it can be used to create product maps. It provides, however, more stable and easier interpretable axes. As a psychometric tool, Tucker‐1 PCA provides a measurement of products on underlying sensory dimensions.