Discovering Partial Least Squares with JMP Reviews

"As nicely stated by the authors, Partial Least Squares (PLS) can deal effectively with "wide data, tall data, square data, collinear variables, and noisy data." These different characterizations of uncertainty often make standard analysis difficult, if not impossible. PLS can handle these situations and, with the combination of JMP applications, the book positions PLS within reach of practitioners and researchers in various domains of applications.

This book is not about the theory of PLS but about its applications in real-life problems. It does include, however, a historical perspective and the mathematical foundations of the PLS algorithms. Combining this theoretical foundation with practical implementations provides unique insights that make this an important contribution to the statistical literature."

Professor Ron S. Kenett
Research Professor, University of Turin, Italy
International Professor, NYU Center for Risk Engineering, USA
Past President, European Network for Business and Industrial Statistics (ENBIS)
Past President, Israel Statistical Association (ISA)
Chairman and CEO, The KPA Group, Israel

"The authors have written a text which is an excellent supplement to the manuals supplied with JMP. The techniques of multiple linear regression (MLR) and principal components analysis are reviewed in the context of application within JMP before the principles of PLS are described. Instructions for performing PLS within JMP are provided together with examples of model specification, fit, and diagnostic reports. Detailed case studies are provided from a range of disciplines, such as predicting octane value from NIR spectra; predictive models for consumer preference, and taste panel data for bread.

A number of JSL scripts are provided so that the reader can perform the operations described within the text with simulations used to illustrate key points; for example, the effect of multiple colinearity on parameter estimates in MLR. The scripts and simulations bring the text to life making it a valuable addition to the JMP multivariate modeller's bookshelf."

Alan Brown
Principal Technical Expert (Statistics)
Syngenta UK Ltd

"After recalling the essence of what partial least squares (PLS) is doing and how it relates to other widely used multivariate techniques such as PCA or MLR, Discovering Partial Least Squares with JMP provides a series of case-studies. Each case-study is self-contained in its own chapter and the reader can focus independently on areas of his own interest.

As a French reader, I obviously jumped to the last case-study, which addresses the problem of "baking bread that people like." The analysis is conducted in a multi-stage manner, starting with a visual inspection of histograms and bi-plots. A PLS model is then built to relate consumer-rating items with Overall Liking. Similarly, another PLS model is built to relate expert sensory ratings with consumer ratings. Both resulting models are finally combined to provide a useful model involving a limited number of sensory ratings that explain Overall Liking. All of this is done with detailed explanations, numerous figures, and even JSL scripts that appear on the left panel of the data table so the user can reproduce each single step of the analysis.

While the analysis is performed, the authors do not neglect potential pitfalls that the reader should be aware of, in particular with respect to variable selection. They also illustrate with JMP a unique interactive profiler that can be used to how selected predictors relate effectively to the response of interest. Technical details for Singular Value Decomposition (SVD) and PLS are given in the appendix, along with useful explanations of widely used indicators such as VIPs. The only slightly negative note is that little is said on the lack of interpretability of latent variables, even if this lack of interpretability does not hinder the predictive power of PLS. Readers might want to consider reading Shmueli's 2010 article, "To Explain or to Predict?", that appeared in Statistical Science. Written by Dr. Ian Cox and Marie Gaudard, this book will be extremely useful for statistics practitioners who want to apply effectively predictive models, such as PLS, and fully benefit from JMP graphical and interactive functionalities. "

Dr. Paul Fogel