By Olaf Behnke, Kevin Kröninger, Grégory Schott and Thomas Schörner-Sadenius (eds)
Wiley
Paperback: £60 €72
E-book: £48.99 €61.99
Also available at the CERN bookshop
This book is actually 11 books in one, with 16 authors, four of whom are also editors. All are high-energy physicists, including one theorist, and all are experts in their assigned areas of data analysis, so the general level of the book is excellent. In addition, the editors have done a good job putting the 11 chapters together so that they work as a single book, and they have even given it a global index. Still, each chapter has its own author(s) and its own style, and I will comment on the individual contributions that I found most interesting.
Roger Barlow (“Fundamental Concepts”) gives a good introduction to the foundations, but surprisingly he has some trouble with frequentist probability, which is the one that physicists understand best because it is the probability of quantum mechanics. Instead of taking an example from physics, where experiments are repeatable and frequentist probability is applicable, he uses life insurance and finds problems. But his example for Bayes’s theorem works fine with frequentist probabilities, even if they are not from physics.
Olaf Behnke and Lorenzo Moneta (“Parameter Estimation”) have produced a useful practical guide for their chapter. The treatment is remarkably complete and concise. I especially liked figure 2.9, which illustrates the fit of a typical histogram to a single peak, showing the value of chi-square as a function of peak position across the whole range of the abscissa, with a local minimum at every fluctuation in the data.
Luc Demortier (“Interval Estimation”) displays an impressive knowledge of both frequentist and Bayesian methodologies, and is careful to list the good and bad features of both in a level of detail that I have seen nowhere else, and did not expect to find in a “practical guide”. He succeeds in presenting a balanced view overall, even though his personal prior shows through in the first sentence, where the point estimate is intuitively defined as “in some sense the most likely value”, instead of the more tangible “in some sense the value closest to the true value”.
The most remarkable aspect of this book is found in the chapters devoted to topics that are not usually covered in books on statistics. Therefore “Classification” (by Helge Voss) is treated separately from “Hypothesis Testing” (by Grégory Schott), describing techniques that are common in data analysis but not used in traditional statistics. In “Unfolding”, Volker Blobel reminds us that statistics is really an inverse problem, although it is not usually treated as such. There are two separate chapters on “Theory Uncertainties” and other “Systematic Uncertainties”, a chapter on “Constrained Fits” and two chapters on “Applications”, some of which duplicate subjects treated elsewhere, but of course from a different point of view. In the concluding chapter, Harrison Prosper, in his inimitable style, takes the reader on “a journey to the field of astronomy”.
In summary, this ambitious project has produced a useful book where experimental physicists will find expert knowledge about a range of topics that are indispensable to their work of data analysis.
