Review

Author(s): Keith Hirst

Review by: Keith Hirst
Source: The Mathematical Gazette, Vol. 77, No. 480 (Nov., 1993), p. 391
Published by: Mathematical Association
Stable URL: http://www.jstor.org/stable/3619806
Accessed: 12-02-2016 05:24 UTC

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.

Mathematical Association is collaborating with JSTOR to digitize, preserve and extend access to The Mathematical Gazette.

http://www.jstor.org

This content downloaded from 130.15.241.167 on Fri, 12 Feb 2016 05:24:57 UTC
All use subject to JSTOR Terms and Conditions
REVIEWS 391

The book underreview aims to present"the bootstrapas a technique with a sound and
promising theoretical basis". The first two chapters are concerned respectively with the
bootstrap and Edgeworth expansion, which are brought together in the third chapter.
Chapter4 discusses the estimation of curves. Some points of mathematicalrigour occupy
chapter 5, and there are five appendices, the most substantialof which is concerned with
Monte Carlo ideas.
The original intention of the bootstrap was to replace the analytical problem
mentioned earlier with the power of the modem computer. As computing power becomes
cheaper, this is a sensible concept, though is does not address the problem of what should
be computed. A further and considerable advantage of the method is that it makes few
restrictionson the class of probabilitymechanisms and applies to many types of data. For
example, it is well adapted to non-parametricideas. It is a pity therefore that the present
book is almost entirely theoretical. Even more seriously, in my view, is that it makes no
mention whatsoeverof data.This is statistics strippedof its raison d'etre.
Even within the theory the book is confined. "Thus, while the present monograph
describes an importantaspect of the theory for the bootstrap,it does not present a complete
picture."One fine feature is the recognition given to the importanceof pivots. A function of
the data and parameters is a pivot if its distribution is the same for all populations
considered. (x--)/cr is a pivot if x is normally distributedwith mean p and variance 0a2,it
being normal with zero mean and unit variance, for all p and a. It is demonstratedhow the
Edgeworthmethod works much betterwith pivots, or with quantitiesthat are nearly pivotal,
than with others. Though the next aspect is not mentioned, the use of pivots blurs the
distinction between the frequentist and Bayesian methods; behaviour in the data space
being mimicked in the parameter space. Consequently pivotal methods have double
authority.The book is entirely frequentist,the Bayesian preferringto use computing power
to integrate,not to sample. Appendix 3 is the only place where Hlailshows any awareness
of the criticisms of confidence intervals;even then failing to addressthem.
This book deals admirably with a part of a part of statistics. It shows extraordinary
competence in analysis. My impression is that it has been hastily written. The text rushes
along with many infelicities: for example, the extraordinaryprelude to the basic equation
(1.1). But as a referenceto this comer, it will long remainthe standardtext.
D.V. LINDLEY
2 Periton Lane, Minehead,SomersetTA24 8AQ

Probability, by Leo Breiman. Pp.421. $34.50. 1992 ISBN 0-89871-296-3 (SIAM)

This is a reprint, in the "Classics in Applied Mathematics"series, of a book first
published in 1968. The title "Probability"gives no indication of the level or coverage. In
fact it would serve as a graduateor final year undergraduatetext in stochastic processes. A
knowledge of the elements of measure and integrationis assumed, and although the author
claims that no prior study of probabilityis needed, he does acknowledge the usefulness of
an acquaintancewith the ideas, such as would come from a first course in combinatorial
probabilityfor example.
Being a book at this level, an extensive review is inappropriatefor the Gazette.
Suffice it to say that there is a considerableemphasis on topics that might well not appearin
a first course on stochastic processes, such as the ergodic theorem, martingales,the renewal
theorem, convergence in distribution, multidimensional central limit theorems and
Brownian motion. The foundationsof the subject are not neglected however, with chapters
in independence, conditional probability and conditional expectation, and Markov chains
contributingto this aspect.
The author is well known for the quality of his exposition, and the re-issue of this
book is welcome, and to be thoroughly recommended for detailed study in a graduate
programme.
KEITHHIRST
MathematicsDepartment,SouthamptonUniversity,SouthamptonS09 5NH

This content downloaded from 130.15.241.167 on Fri, 12 Feb 2016 05:24:57 UTC
All use subject to JSTOR Terms and Conditions