The
book is extensive in scope, covering so much background material that
quantitative genetics applications are not reached until page 560. The
overview of statistical theory starts with an introduction to the basic
probability distributions and distribution theory, and covers the fundamentals
of classical inference before reaching Bayesian methods around page 200.
The coverage of Bayesian theory is extensive, and includes a discussion
of information and entropy, and of the notion of "uninformative" priors,
as well as model assessment and model averaging. There follows a chapter
on the EM algorithm, which forms a prelude to a substantial introduction
to MCMC. Turning then to quantitative genetics applications, the authors
cover Bayesian formulations of (and Gibbs sampler algorithms for) the
basic linear models with t-distributed errors, as well as categorical
response and longitudinal data, finishing with an introduction to models
for segregation and quantitative trait locus (QTL) analysis.
There are some simple data examples distributed through the text, and
occasional outline algorithms for computational implementation. However
this is not a "practical" book, it is about the ideas motivating the theory.
Some detailed derivations and proofs are given. With so much material
packed into one volume, it is inevitable that the reader will need some
mathematical sophistication to be comfortable with the formulas and derivations.
However, the mathematical level is not high -- the reader need only be
familiar with elementary calculus and matrix notation to glean the most
important ideas. Even previous familiarity with probability and statistical
inference is not strictly needed, but is in practice a prerequisite --
there is too much for a reader to absorb without some head start.
I found the coverage of material to be excellent: well chosen and well
written, and I didn't spot a single typographical error. There is little
that I would have omitted: the discussion of information and entropy is
perhaps unnecessary at this level, and possibly the entire chapter devoted
to the EM algorithm is excessive, given the prominence of MCMC in the
applications. Conversely I can think of little to ask for that is not
provided. The coverage of MCMC convergence is thin and would have benefited
from examples. Similarly, problems of model mispecification and its diagnosis
could have been given more attention, with examples. The authors do not
discuss software, arguing that developments happen too quickly. I find
this justification weak: some guidance is better than nothing, and software
web sites are becoming more stable.
There are no exercises, and the book isn't particularly suited as the
basis for a course text. It can serve as a resource book for masters-level
taught courses, but will be most useful for PhD students and other researchers
who need to fill in gaps in their knowledge, grasp the intuition behind
statistical techniques, models, and algorithms, and find pointers to more
extensive treatments. Overall, I find that the authors have succeeded
admirably in their goals. I highly recommend this excellent book to any
researcher seeking a graduate-level introduction to the modern statistical
methods applied in quantitative genetics.
David Balding, Imperial College, UK |