Abstract
The principal goal of Managerial Application of Multivariate Analysis in Marketing is to provide a nonmathematical overview of several multivariate statistical techniques that are frequently used in marketing research. The book is intended as a reference source to be used by marketing practitioners, including marketing researchers, product managers, and marketing managers. Although the book is written for practitioners who do not have formal training in statistical analysis, the authors suggest that doctoral students and other graduate students in marketing will find the book useful. The overall thrust is to provide descriptions of the techniques' purposes, the kinds of data needed to use the techniques, the assumptions required, and the interpretation of the results.
The authors do a good job of accomplishing their goal by explaining the purposes of the multivariate techniques and by demonstrating the methods with real-world examples. In some respects, the actual marketing research examples that are used liberally throughout the book are one of its greatest strengths.
Because the stated purpose of the book is to cover the topics in a nontechnical approach, there is some fairly superficial coverage of concepts. Readers who want a rigorous review of the topics will likely be somewhat disappointed. However, each chapter concludes with references that provide an opportunity to explore the subject matter in more depth. Despite the chapter references, the shallow treatment of statistics will probably limit the book's usefulness to graduate students.
The book is divided into four parts. Part 1 (Chapters 1 and 2) contains an overview of the book and a discussion that is intended to help readers understand why multivariate techniques are important. Chapter 1 describes dependence and interdependence techniques, discusses typical marketing applications of multivariate methods, and presents a review of the primary computer-software packages that are used in multivariate analysis. Chapter 2 explains deterministic and probabilistic models, types of measurement scales, reliability, and validity.
Part 2 (Chapters 3–10) discusses dependence methods. Chapter 3 explains basic linear-regression analysis and contains a straightforward review of linear regression and correlation, the types of measurement scales required, the interpretation of coefficients, and tests of statistical significance. As do most chapters in the book, Chapter 3 contains several detailed examples of actual marketing research projects. The real-world examples of regression and correlation analysis help readers understand how the techniques can be applied to actual marketing problems.
Chapter 4 discusses the topic of nonlinear regression with three methods: (1) the nonlinear correlation coefficient eta, (2) mathematical transformations to one or both variables, and (3) recoding of one or both variables. This is a weaker chapter in the book because the methods of the eta coefficient and recoding of the variables are problematic. Regression analysis normally assumes that data for all variables are at least on an interval scale. However, the eta coefficient requires a categorical independent variable (or an independent variable that is grouped into categories if the data are interval). Grouping interval data into arbitrary categories discards information and thus yields a less useful analysis. In some cases, an interval-scaled independent variable has multiple measurements at fixed values of the independent variable, in which case the eta coefficient is more informative and is equivalent to a measure based on within-group sums of squares computed in analyses of variance (Green 1978). The third method, recoding of the variables, uses what is called a “low-tech” approach. In addition to the assumption that the independent variable is categorical, recoding uses an arbitrary process. As the authors explain (p. 72, emphasis in original), “In this approach, all cases or respondents in each response category of a variable are given a new value that will ensure that the entire variable will exhibit a linear relationship with the other variable. One way to do this is to assign for each category the mean value of all cases in that category on the other (continuous) variable.” Chapter 4 contains a good discussion of the second method (mathematical transformations); however, much of the chapter does more to confuse the issue than to illuminate it. In addition, the chapter on multiple regression could have been used to discuss various ways for readers to deal with nonlinear relationships by means of polynomial or other models. On balance, Chapter 4 unnecessarily detracts attention from the overall quality of the book.
Chapter 5 discusses multiple regression and correlation analysis. This chapter contains a good discussion of the multiple regression model, multicollinearity, dummy variables, interpretation of the regression coefficients, and interpretation of the coefficient of determination. In addition, the authors explain the use of dummy variables as a way to model nonlinear relationships. The final section of the chapter presents a discussion of stepwise regression. However, Chapter 5 does not mention any of the problems with stepwise methods, such as the technique's tendency to capitalize on chance when entering independent variables.
Chapter 6 addresses logistic regression as an appropriate technique for categorical-dependent variables. An example in the chapter divides respondents into two groups: (1) respondents who used the sponsoring company's products and (2) respondents who used another company's products. The independent variables in logistic regression, just as in linear regression, should be at least interval-scaled data (Menard 2002). Chapter 7 covers discriminant analysis as an alternative to logistic regression and canonical analysis. The authors use several real-world examples to demonstrate how the techniques can be used to predict group membership and to create perceptual maps.
Chapters 8 and 9 contain a fairly complete discussion of conjoint analysis. Again, the authors use real-world examples to illustrate applications of the technique. Chapter 10 presents three interaction-detection methods: automatic interaction detector, chi-square automatic interaction detector, and classification and regression trees. According to the authors, these techniques search out “interactions among independent variables, which are nonlinear combinations of two or more independent variables that will predict or explain a dependent variable” (p. 178). The authors point out that these techniques require large sample sizes. In addition, because the techniques capitalize on chance variations in the data, they recommend that the analyses be cross-validated.
Part 3 (Chapters 11–16) presents interdependence methods. Chapter 11 covers factor analysis (using the principal components approach). The topics presented include data reduction, factor rotation (orthogonal and oblique), and factor interpretation. A common use of principal components analysis is to take many variables and reduce them to fewer underlying dimensions (Basilevsky 1994). The chapter provides a lucid discussion of factor analysis that clearly describes practical applications of the technique.
Chapters 12 and 13 present two approaches to cluster analysis: Chapter 12 covers hierarchical clustering, and Chapter 13 discusses partition clustering. These techniques can be used to arrange data into similar groups or clusters. Hierarchical methods have the advantage of not requiring that the number of cluster be defined in advance, but the interpretation of the dendrograms can be difficult. In contrast, partition clustering requires that the number of clusters be specified in advance, and the technique sorts the data into that number of clusters. The results of cluster analysis can be misleading. Therefore, it is important to pay careful attention to the interpretation and validation of the results (Everitt 1991).
Chapter 14 contains a discussion of correspondence analysis, which is used to analyze contingency tables and yields information somewhat similar to that produced by factor analysis. Correspondence analysis is described as a post hoc analysis tool that can be used after a researcher has rejected the null hypothesis of independence through a chi-square test.
Chapter 15 presents structural equation models. Topics covered include causal models, comparison with factor analysis, comparison with regression analysis, and path analysis. The authors describe (p. 337, emphasis in original) drawbacks to the use of structural equation models: “It is easy to denigrate structural equations models that are not based on a solid foundation of theory, objective measures (versus subjective ones), and carefully controlled experiments…. Critics might argue that without a theoretical foundation and carefully controlled experiments, structural equation modeling should not be attempted.” Nevertheless, structural equation models offer great promise if they are used properly. Among other uses, structural equation modeling can be used to confirm the findings from factor analysis (Kelloway 1998).
Chapter 16 discusses multidimensional scaling of similarities data. The chapter contains a useful explanation of the important topics of how to gather data for use in multidimensional scaling computer packages and the related question of how to develop a similarities matrix.
The book concludes with Part 4, which includes Chapter 17, titled, “Squeezing More Useful Information out of Expensive Consumer Surveys.” This chapter contains a useful framework “that would help readers understand which techniques are most appropriate for which tasks or objectives” (p. 365). A conclusion is that most real-world marketing research produces data that can benefit from multivariate analysis and that this type of analysis has the strong potential for improving the quality of the findings.
In summary, the book is a clearly written guide to applied multivariate analysis in marketing. It is filled with many actual commercial research examples that help readers understand the types of problems that the techniques can address. The book meets its goal of presenting complex topics in a nontechnical manner. The book's target market of marketing practitioners should be able to understand the content and recognize the benefits offered by multivariate analysis.