An introduction to Multi-block Component Analysis by means of a flavor language case study

Generalized Procrustes Analysis (GPA), SUM Principal Component Analysis (SUM-PCA) and Multi-block Component Analysis (MBCA) were applied to a cross-continental flavor language study from industrial practice. One aspect that these explorative methods have in common is that data from different tables (blocks) are jointly decomposed into a low dimensional sub-representation by pursuing a consensus between the blocks. For GPA and SUM-PCA the blocks are considered to be of equal rank. However, the data from this flavor language study consisted of three blocks (panels) of different rank due to different profiling methods and different descriptive performances of the panels. It is demonstrated that, when adequately scaled, SUM-PCA and GPA yield similar results but fail to describe consensus in higher flavor dimensions when the blocks are of different rank. MBCA allows blocks of different rank and provides flexible means to study the data structures within and between the blocks. (C) 2003 Elsevier Science Ltd. All rights reserved.