Does every modern engineer or scientist faced with multivariate data need to study linear discriminant analysis, multivariate analysis of variance, correspondence analysis and the like to analyze his or her data? No, it is sufficient to have one set of tools to analyze 90 percent of one's data sets. That is the clear-cut philosophy of this book, with the authors stating that "there is no single best way to analyze data; many different approaches might work as long as they are used correctly. So the question for the researcher who does not want to spend too much time on data analysis, is to choose one versatile approach, and learn to use it well" (p. 5). The basis of this philosophy is formed by the opinion that "... the person who 'owns' the problem and has the contextual back-ground knowledge should try to do as much as possible of the data analysis, himself or herself. On the other hand. it is important to know when to call for help, like in all do-it-yourself activities" (p.12). The 'versatile approach' is soft bilinear modeling extended with graphics and resampling methods for validation. Bilinear modeling is based on the property that a large data matrix X can usually be summarized in two (much) smaller matrices containing the essential information. The same can be done with a matrix Y with responses if available. Hence. bilinear modeling can be used to study relationships within one data set X (e.g.. pattern recognition). and relationships between two data sets X and Y (e.g., calibration and property prediction). New add-ons for bilinear models are also presented such as jack-knifing regression coefficients ill partial least squares regression models.
展开▼
