Book Review: Empirical Model Building

As one who has been involved during the past decade in the application of management science/operations research (MS/OR) techniques for the purpose of marketing decision making, I have long known that high quality texts covering both the art and science of model building are rare. Though many good publications address the techniques of MS/OR, virtually no texts examine the issues associated with the application of those techniques to real-world problems. Fortunately, Empirical Model Building is an exception to that rule. Though not associated with marketing or its related disciplines, the author of this text is eminently qualified to discuss issues related to the meaningful application of statistical and mathematical techniques to real problems. James R. Thompson is Professor and Chairman of the Department of Statistics at Rice University. He also serves as an Adjunct Professor of Biomathematics at the M. D. Anderson Cancer Center at the University of Texas at Austin. His background includes a BE degree in chemical engineering from Vanderbilt University and a PhD degree in mathematics from Princeton University. As the book demonstrates, the author has developed an interest in an eclectic set of problem areas to which he applies his model-building skills. The underlying theme of Thompson's text is that the process of model building to solve real-world problems requires a perspective not usually found in model-building texts. As a result, his book offers a refreshing departure from texts and articles that present one or more techniques and advocate that those techniques be applied to a problem. Instead of emphasizing a "methodology in search of a problem," Thompson views modeling from the perspective of "a problem in search of a solution." With that theme, the book is organized so as to build readers' skills in developing and using models that cope with the complexities of a real-world environment. Nonstandard modeling situations are presented throughout the book to demonstrate the true usefulness of the modeling process and to encourage the use of creative solutions in model building. To accomplish his objective, the author has organized the text into five major chapters and two technical appendices. The first chapter presents several models of growth and decay. Varied application areas are used to demonstrate how assumptions about growth and decay can be translated to statistical models. Applications include a pension and annuity plan, federal income tax "bracket creep," the retirement of a home mortgage, the population growth and food supply models proposed by the nineteenth-century British economist Malthus (who is considered a pioneer in modem population study), and the metastatic progression of a cancerous tumor in the human body. Though these applications are clearly associated with financial analysis, economics, and oncology instead of marketing, they do serve as interesting problems for which growth and decay models can be readily applied. Chapter 2 provides an interesting discussion of more complicated systems of growth and decay models. The author introduces examples from the competition among species, military combat situations, and epidemiology. These examples demonstrate the usefulness of statistical techniques in modeling such all-too-real problems as manning NATO defense positions in Europe and the spread of AIDS in the U.S. In many cases, historical information taken from military, population, and medical records is used to form the databases from which the author works. Relatively simple models are applied to the data, and the results provide some interesting implications for understanding the problem at hand. The subsequent discussions, especially about the employment of military strategies and the use of social rather than medical solutions for decreasing the spread of AIDS, are very interesting. The material in Chapter 3 is quite different from that in the two preceding chapters. The chapter is devoted to the topic of simulation-based techniques, which the author strongly advocates as a preferred alternative to the use of differential and integral-difference equation calculus in the practice of model building. The obvious reason for his preference is the difficulty of finding closedform expressions for many of the complex equations that are applied to today's complicated problems. In other words, better solutions to today's problems may be found