Customer Reviews

Bayesian Inference in Statistical Analysis

5 star:		(3)
4 star:		(0)
3 star:		(0)
2 star:		(0)
1 star:		(0)

 

 

This is a book written in 1973 but showing the brilliance and forethought of George Box. Wiley reprinted it in its popular paperback classic series in 1992. The first few chapters introduce Bayesian ideas and show how with noninformative priors the Bayesian results resemble the classical frequentist results. This essentially reviews the work pioneered by Harold Jeffreys.
In the latter chapters more complex problems are introduced including many that do not have nice classical solutions. Box and Tiao show how Bayesian methods contribute ideas that provide new insights into these problems. The discussion of hierarchical models anticipated the developments in Bayesian methods that occurred in the 1990 when the MCMC methods burst onto the scene.

This book is nice for a historical perspective but anyone seriously interested in doing modern Bayesian analysis needs a book that deals with the MCMC methods and there are many nice books available these days.

Have you ever wondered about the origins and meaning of statistical concepts? Most of the books on Statistics shy away from this topic, they just throw formulae at you! Not this book. In the first two chapters it goes to great extent to explain a very important concept of noninformative prior. It also states very clearly the differencies between more traditional Sampling Theory approach and Bayesian Analysis. While majority of Statisticians prefer the ideas and notions of Sampling Theory, majority of Scientists and Control System Engeneers are more inclined to use Bayesian Analysis because of its practicality. This book gives a plenty of material on Bayesian Inference and shows how to put theoretical knowledge into practice. It presents the material in transparent and orderly fashion but it requires certain degree of mathematical sophistication. A prerequisite would be any standard text book on Statistical Inference.

This book is a classic, particularly with biologists and medical research; less so with physicists. I wish I had read it when I was an undergraduate. I would have wasted less time with Classical methods.

E.T. Jaynes was not overly impressed. In his assessment of this book, He wrote:
"G.E.P. Box is, like L.J.Savage, a curious anomaly in this field; he was assistant to R.A. Fisher and married his daughter, but became a Bayesian in issues of inference while remaining Fisherian in matters of significance tests, which he held to be ouside the ambit of Bayesian methods. In Jaynes (1985), we argue that, on the contrary, any rational significance test requires the full Bayesian apparatus."

I think this is unfair. The book shows its age, but it gives fully Bayesian solutions to many classical problems and it is well written, although the material is dense for a beginner.