You are here

Model Selection and Model Averaging

Gerda Claeskens and Nils Lid Hjort
Publisher: 
Cambridge University Press
Publication Date: 
2008
Number of Pages: 
312
Format: 
Hardcover
Series: 
Cambridge Series in Statistical and Probabilistic Mathematics 27
Price: 
70.00
ISBN: 
9780521852258
Category: 
Textbook
[Reviewed by
Ita Cirovic Donev
, on
01/27/2009
]

Model Selection and Model Averaging is a book about making choices, technical model-based choices, that is. It teaches us how to develop a sound statistical model based on certain criteria.

For academics and professional statisticians this road of model selection does not necessarily need to be addressed in full, as they already acknowledge the importance of model selection and understand the consequences of taking shortcuts. However, for “practical” statistical model building this is more than a needed book. More and more we witness statistical model building with the push of a button, i.e. trusting our lovely statistical software application to do the work and, the worst of all, taking the results for granted. Needless to say, the software is never used in its full force. I can just hope that book like this will find its way into the practitioner’s world of statistics and model building.

The main focus of the book is on selection methods. The first part of the book presents theoretical derivations of AIC (Akaike information criterion) and BIC (Bayesian information criterion) methods. It also introduces the datasets that we will work with throughout the book. The examples are presented in the simplest form in the first chapter, then in later chapters they are developed. The working out of these examples holds the book together. In the following chapter the FIC (Focused information criterion) is discussed.

Chapter 7 is devoted to model averaging. The authors discuss both the frequentist and Bayesian point of view on this topic. The construction of lack-of-fit and goodness-of-fit tests is also discussed. The theoretical presentation is not in formal theorem-proof style, but rather a narrative presentation of results. The main focus, it seems, is on conveying the methods via the examples. The results are presented with very detailed applied presentation. I would say this serves it purpose well.

Chapter 9 takes the established examples and presents the final thought on the model selection. Also, additional smaller examples are provided to combine the theoretical expositions presented earlier in the book. If there were any “broken pieces” this chapter unifies them. The examples are worked out in great detail with tabular as well as illustrative presentation.

Each chapter has a little section on literature. This I appreciate a lot; it could be of great use to graduate students interested in this area. Exercises are provided at the end of each chapter. They are divided between theoretical (proofs) and applied. The applied exercises are all based on some form of model building and gathering thoughts on the results. These are quite essential for proper understanding. Some exercises are vague in the sense that they do not provide the reader with guidance as to what to look for (we need to keep in mind that this book is not for someone who excels in this area but rather for “students” in various disciplines).

Overall, given the inviting style of the presentation and the quality of the material, this book could be quite a catch for graduate students as well as for practitioners where models really do make difference.


Ita Cirovic Donev holds a Masters degree in statistics from Rice University. Her main research areas are in mathematical finance; more precisely, statistical methods for credit and market risk. Apart from the academic work she does statistical consulting work for financial institutions in the area of risk management.

 

Preface; A guide to notation; 1. Model selection: data examples and introduction; 2. Akaike’s information criterion; 3. The Bayesian information criterion; 4. A comparison of some selection methods; 5. Bigger is not always better; 6. The focussed information criterion; 7. Frequentist and Bayesian model averaging; 8. Lack-of-fit and goodness-of-fit tests; 9. Model selection and averaging schemes in action; 10. Further topics; Overview of data examples; Bibliography; Author index; Subject index.