Models for Discrete Data is a refreshing applied statistics book, refreshing for its clear presentation, style and contents. This is one of those books that one remembers fondly after the semester is over and done. A good applied statistics book can really open one’s eyes to the solution to the problem(s) at hand. The wrong or bad book can lead one in a wrong direction or, as in most cases, in no direction at all. It's a good thing that there are libraries ,so that you can always consult other texts. But if you start with Models for Discrete Data, you will not have to look elsewhere for additional references during your reading. This is an applied statistics book that every serious statistician, especially a student, should have on his desk.
Zelterman provides us with a detailed intermediate account on the modeling of discrete data. After a small introduction in chapter 1, all of chapter 2 is devoted to sampling distributions. It is very detailed. Logistic regression, log-linear models and coordinate free models are presented in detail in the following chapters. Both model development and diagnostic tools are discussed. Chapter 6 is reserved for additional topics such as longitudinal data analysis, case-control studies and advanced goodness of fit.
Applied statistics books, in general, do not vary all that much in style. There is a tendency for the authors to loose themselves in trying to explain both the applied and theoretical concepts without much regard as to the impact on the potential reader. For experienced statisticians, this is not a problem, but for students it results in a statistical cocktail with a very bad taste. Zelterman has the same idea but there is no “cocktail” in the end.
Browsing through the text one sees the inevitable mixture of theory and practical examples. Most books do not bridge this connection well. However, this book we can consider sort of an “outlier.” The theoretical concepts are explained separately from the practical examples. The presentation of the theory is very clear and extremely easy to follow, especially due to clear notation. Readers with no previous knowledge of modeling techniques for discrete data will find this book very forthcoming and generally useful. For more experienced readers the book can serve as a reminder and a reference in the course of practical statistical modeling. The chapters that cover logistic regression, log-linear models and coordinate-free models all have a separate practical section. It includes simple and more complex examples as well as the computer code. The examples are mostly oriented toward problems in medical studies. Examples are supported with SAS code and graphical outputs. All of the outputs are explained in detail.
As this is an intermediate text, readers will need good knowledge of linear models. Also, it would be advantageous to the reader to be acquainted with some statistical programming language such as SAS or S-Plus, but it is not necessary. A great follow-up to Models for Discrete Data would be selected sections of Regression Modeling Strategies by F. Harrell, a more advanced text but still my favorite book on advanced applied regression. After mastering Models for Discrete Data and Regression Modeling Strategies one would be ready for some serious discrete data modeling.
Ita Cirovic Donev is a PhD candidate at the University of Zagreb. She hold a Masters degree in statistics from Rice University. Her main research areas are in mathematical finance; more precisely, statistical mehods of credit and market risk. Apart from the academic work she does consulting work for financial institutions.