Bayesian Data Analysis
|
This is an excellent book. Philosophical ramblings are more or less avoided, and the authors get down to analysing data. Chapters 3 and 5 take the reader through basic Bayesian analysis including posterior simulation and hierarchial modelling. There's quite a bit of Greek stuff and assumptions of conjugacy, but the approach is still very practical. A worked example of a hierarchical model is given in sufficient detail for the reader to reproduce.
The chapter that introduces MCMC is extremely good and also very practical. From studying this chapter, I was able to go from knowing nothing about MCMC to programming my first Metropolis algorithm. Going back through the chapter a couple of times, I was then able to program a Metropolis algorithm for a novel application, and to build in tuning steps and assess convergence. All from one chapter.
The later part of the book has a vaguer feel to it. Many of the models are described in quite high level terms and details of computations are less forthcoming. It does, however, give a very strong impression of just how diverse the applications of MCMC are.
If you want to learn Bayesian data analysis, this book is the one you're looking for.
|
|
|
The Standard Work
|
In my view, this is the single best book on Bayesian statistics. It's set at about masters level for a statistics specialist, though it could be read by anyone with matrix algebra and calculus. It starts right from scratch, with basic ideas about probability and develops Bayesian ideas through simple one-parameter models right up to the most sophisticated types of heirarchical models extant. Because the subject matter was formerly the subject of heated debate at the philosophical level this book carefully avoids philosophical argument. The authors prefer to make their case by presenting the reader with
a wide range of powerful techniques and leaving the philosophy to others. Each chapter ends with a guide to the literature on the subject matter of the chapter. The tone of the book is practical and gives much guidance on computational issues. The second edition made a great book even better by beefing up the parts on computation to include more on how to implement state-of-the-art Markov Chain Monte Carlo methods using freely available software (R and WinBUGS) as well as how to write your own. Outstanding!
|
|
|
United States | 
United Kingdom | 
Canada