Short Book Reviews On-Line 1995

MONOTONE STRUCTURE IN DISCRETE-EVENT SYSTEMS.	P. Glasserman and D.D. Yao.
LECTURES ON DISCRETE TIME FILTERING.	R.S. Bucy with the assistance of B.G. Williams. C.S. Bussus Consulting Editor.
AI AND COMPUTER POWER: The Impact on Statistics.	D.J. Hand (Ed.).
SHADOWS OF THE MIND. A Search for the Missing Science of Consciousness.	R. Penrose.
CONDORCET: ARITHMETIQUE POLITIQUE, TEXTES RARES OU INEDITS (1767-1789).	B. Bru and P. Crépel (Eds.).
LE JEU DE LA SCIENCE ET DU HASARD.	D. Schwartz.
PHILOSOPHICAL ESSAY ON PROBABILITY.	P.S. Laplace.
LOOKING FOR THE LAST PERCENT - The Controversy over Census Undercounts.	H. Choldin.
FUNDAMENTALS OF EARTHQUAKE PREDICTION.	C. Lomnitz.
QUANTITATIVE METHODS IN BIOLOGICAL AND MEDICAL SCIENCES. A HISTORICAL ESSAY.	H.O. Lancaster.
STATISTICS IN MEDICAL RESEARCH. DEVELOPMENTS IN CLINICAL TRIALS.	E.A. Gehan and N.A. Lemak.
ANALYSING SURVIVAL DATA FROM CLINICAL TRIALS AND OBSERVATIONAL STUDIES.	E. Marubini and M.G. Valsecchi.
EDA. EXPLORATIVE DATENANALYSE. Einführung in die Deskriptive Statistik. Zweite, Neubearbeitete und Erweiterte Auflage, 2nd edition.	W. Polasek.
AN INTRODUCTION TO REGRESSION GRAPHICS.	R.D. Cook and S. Weisberg.
MODERN APPLIED STATISTICS WITH S-PLUS.	W.N. Venables and B.D. Ripley.
PRACTICAL METHODS FOR DESIGN AND ANALYSIS OF COMPLEX SURVEYS.	R. Lehtonen and E.J. Pahkinen.
EXACT STATISTICAL METHODS FOR DATA ANALYSIS.	S. Weerahandi.
CONTINUOUS UNIVARIATE DISTRIBUTIONS. Volume 1, 2nd edition.	N.L. Johnson, S. Kotz and N. Balakrishnan.
ASYMPTOTIC APPROXIMATIONS FOR PROBABILITY INTEGRALS.	K.W. Breitung.
PROBABILISTIC CAUSALITY IN LONGITUDINAL STUDIES.	M. Eerola.
PRACTICAL METHODS FOR RELIABILITY DATA ANALYSIS.	J.I. Ansell and M.J. Phillips.
THE BAYESIAN CHOICE: A Decision-Theoretic Motivation.	C.P. Robert.
APPLIED BAYESIAN FORECASTING AND TIME SERIES ANALYSIS.	A.Pole, M. West and J. Harrison.
KERNEL SMOOTHING.	M.P. Wand and M.C. Jones.
APPLIED DISCRIMINANT ANALYSIS.	C.J. Huberty.
DIFFUSIONS, MARKOV PROCESSES, AND MARTINGALES. Volume 1: FOUNDATIONS, 2nd edition.	L.C.G. Rogers and D. Williams.
STOCHASTIC ORDERS AND THEIR APPLICATION.	M. Shaked and J.G. Shanthikumar.
HIERARCHICAL DECISION MAKING IN STOCHASTIC MANUFACTURING SYSTEMS.	S.P. Sethi and Q. Zhang.
HIDDEN MARKOV MODELS. ESTIMATION AND CONTROL.	R.J. Elliott, L. Aggoun and J.B. Moore.
CASE STUDIES IN BIOMETRY.	N. Lange, L. Ryan, L. Billard, D. Brillinger, L. Conquest and J. Greenhouse, (Eds.).
AIDS Epidemiology. A Quantitive Approach.	R. Brookmeyer and M.H. Gail.
PLANNING PHARMACEUTICAL CLINICAL TRIALS: BASIC STATISTICAL PRINCIPLES.	W.M. Wooding.
STATISTICS FOR THE ENVIRONMENT 2. Water Related Issues.	V. Barnett and K.F. Turkman (Eds.).
STATISTICS IN ENGINEERING. A Practical Approach.	A.V. Metcalfe.
ANALYSIS OF TWO-WAY LAYOUTS.	J. Mandel.
THEORY OF THE COMBINATION OF OBSERVATIONS LEAST SUBJECT TO ERROR.	C.F. Gauss. Translated by G.W. Stewart.
STATISTICS AND THE EVALUATION OF EVIDENCE FOR FORENSIC SCIENCE.	C.G.G. Aitken.
CLINICAL BIOSTATISTICS: AN INTRODUCTION TO EVIDENCE-BASED MEDICINE.	G. Dunn and B. Everitt.
STATISTICAL DESIGN AND ANALYSIS IN PHARMACEUTICAL SCIENCE	S. Chow and J. Liu.
META-ANALYSIS, DECISION ANALYSIS, AND COST- EFFECTIVENESS ANALYSIS: METHODS FOR QUANTITATIVE SYNTHESIS IN MEDICINE.	D.B. Petitti.
THE PLEASURES OF PROBABILITY.	R. Isaac.
CONTINUOUS UNIVARIATE DISTRIBUTIONS, Volume 2, 2nd edition.	N.L. Johnson, S. Kotz and N. Balakrishnan.
PROBABILISTIC TECHNIQUES IN ANALYSIS.	R.F. Bass.
MULTILEVEL STATISTICAL MODELS.	H. Goldstein.
RECENT ADVANCES IN DESCRIPTIVE MULTIVARIATE ANALYSIS.	W. Krzanowski (Ed.).
RELIABILITY ASSESSMENT OF ELECTRICAL POWER SYSTEMS USING MONTE CARLO METHODS.	R. Billinton and W. Li.
SYSTEM RELIABILITY THEORY. Models and Statistical Methods.	A. Høyland and M. Rausand. New York: Wiley, 1994, pp. x + 518, £59.00.
THE WEIGHTED BOOTSTRAP.	P. Barbe and P. Bertail.
PARAMETRIC STATISTICAL THEORY.	J. Pfanzagl. With the assistance of H. Hamböker.
DISTRIBUTION-FREE STATISTICAL METHODS, 2nd edition.	J.S. Maritz.
GROWTH CURVES.	A.M. Kshirsagar and W.B. Smith.
INTEGRATION AND PROBABILITY.	P. Malliavin. In cooperation with H. Airault, L. Kay and G. Letac.
QUEUEING NETWORKS WITH BLOCKING: EXACT AND APPROXIMATE SOLUTIONS.	H.G. Perros.
LIMIT THEOREMS OF PROBABILITY THEORY: SEQUENCES OF INDEPENDENT RANDOM VARIABLES.	V.V. Petrov.
CYCLE REPRESENTATIONS OF MARKOV PROCESSES.	S.L. Kalpazidou.
RANDOM WALKS OF INFINITELY MANY PARTICLES.	P. Révész.
RANDOM WALKS AND RANDOM ENVIRONMENTS. Volume 1: RANDOM WALKS.	B.D. Hughes.
STOCHASTIC ORDERING AND DEPENDENCE IN APPLIED PROBABILITY.	R. Szekli.
TOPICS ON REGENERATIVE PROCESSES.	V. Kalashnikov.
STOCHASTIC VISIBILITY IN RANDOM FIELDS.	S. Zacks.
RANDOM SUMS AND BRANCHING STOCHASTIC PROCESSES.	I. Rahimov.
CHOQUET-DENY TYPE FUNCTIONAL EQUATIONS WITH APPLICATIONS TO STOCHASTIC MODELS.	C.R. Rao and D.N. Shanbhag.
IMAGE ANALYSIS, RANDOM FIELDS AND DYNAMIC MONTE CARLO METHODS. A MATHEMATICAL INTRODUCTION.	G. Winkler.
HARMONIC ANALYSIS OF PROBABILITY MEASURES ON HYPERGROUPS.	W.R. Bloom and H. Heyer.
FROM BROWNIAN MOTION TO SCHRÖDINGER EQUATION.	K.L. Chung and Z. Zhao.
KENDALL'S ADVANCED THEORY OF STATISTICS, Volume 2A: MULTIVARIATE ANALYSIS. Part I: Distributions, Ordination and Inference.	W.J. Krzanowski and F.H.C. Marriott.
KENDALL'S ADVANCED THEORY OF STATISTICS, Volume 2B: BAYESIAN INFERENCE.	A. O'Hagan.
MULTIDIMENSIONAL SCALING.	T.F. Cox and M.A.A. Cox.
MULTIVARIATE STATISTICAL MODELLING BASED ON GENERALIZED LINEAR MODELS.	L. Fahrmeir and G. Tutz.
PIVOTAL MEASURES IN STATISTICAL EXPERIMENTS AND SUFFICIENCY	S. Yamada.
MULTISTAGE SELECTION AND RANKING PROCEDURES. Second-Order Asymptotics.	N. Mukhopadhyay and T.K.S. Solanky
STATISTICAL MODELS FOR ORDINAL VARIABLES.	C.C. Clogg and E.S. Shihadeh.
STATISTICAL FACTOR ANALYSIS AND RELATED METHODS: Theory and Applications.	A. Basilevsky.
SERIES APPROXIMATION METHODS IN STATISTICS.	J.E. Kolassa.
INTRODUCTION TO FACET THEORY. Content Design and Intrinsic Data Analysis in Behavioral Research.	S. Shye, D. Elizur and M. Hoffman.
PERTURBATION THEORY IN MATHEMATICAL PROGRAMMING AND ITS APPLICATIONS	E.S. Levitin.
ROBUST ASYMPTOTIC STATISTICS.	H. Reider.
THEORETICAL PROBABILITY FOR APPLICATIONS.	S.C. Port.
ELEMENTS OF QUEUEING THEORY. Palm-Martingale Calculus and Stochastic Recurrences.	F. Baccelli and P. Brémaud
STOCHASTIC MODELS: AN ALGORITHMIC APPROACH.	H.C. Tijms.
FRACTALS, RANDOM SHAPES AND POINT FIELDS. METHODS OF GEOMETRICAL STATISTICS.	D. Stoyan and H. Stoyan.
MARKOV DECISION PROCESSES: DISCRETE STOCHASTIC DYNAMIC PROGRAMMING.	M.L. Puterman.
EXPONENTIAL STABILITY OF STOCHASTIC DIFFERENTIAL EQUATIONS	X. Mao.
STABLE NON-GAUSSIAN RANDOM PROCESSES. Stochastic Models with Infinite Variance.	G. Samorodnitsky and M.S. Taqqu.

Contents:
1. Introduction
2. Some basic concepts
3. Antimatroid structure: Monotonicity
4. Lattice structure: Convexity and concavity
5. Links to other models
6. Monotone optimal control
7. Subadditivity and stability
8. Association and optimal coupling
9. Perturbation analysis

Discrete-event systems is an emerging area of importance in systems science. These are systems com-prising a number of sub-systems which interact at discrete intervals. Examples of such systems can be found in manufacturing systems, communicating networks and computer resource allocations. This research monograph combines logical/qualitative issues with temporal/ quantitative analysis. The tools used include Markov models and the notion of monotonicity. Monotonicity essentially implies that the future evolution of the system not be radically dependent on the event sequence. This leads to important simplifications in optimization since short term gains are not off-set by long-term losses.
The book is quite mathematical but contains a number of insightful illustrative examples. It is recommended to anybody wishing to find out about recent research in discrete event systems using statistical Markovian representations.

Contents:
Introduction
1. Review
2. Random noise generation
3. Historical background
4. Sequential filtering theory
5. Burg technique
6. Signal processing
7. Classical approach
8. A priori bounds
9. Asymptotic theory
10. Advanced topics
11. Applications
12. Phase tracking
13. Device synthesis
14. Random fields
15. Bibliography

This book is in the form of a series of lectures on linear and nonlinear sequential filtering theory. Perhaps as a result, it is rather uneven in difficulty and coverage when viewed as a book. The first two chapters contain very well-known material which might be better placed in an appendix. After a good discussion of the Kalman-Bucy filter and the Ricatti equation (used extensively in this book), Burg's predictive methods are covered. While reference is made to applications of the Burg algorithm in geo-physics through the work of Claerbout (not Clairbout as in the book), I feel the discussion is too sketchy to be fully informative. The coverage of nonlinear filtering surprisingly does not reference Mallows (1980) Some theory of nonlinear smoothers, Annals of Statistics, 8, 695-715. The bibliography only has references to a few papers in the 1980's or later, and the book lacks an index. While the book could prove useful to an experienced researcher, its rather selective content and uneven approach would not be ideal for research students.

Contents:
1. Statistics and computing: The promise and the risk, D.J. Hand
2. AI and simulation, R.J. Paul
3. Towards a statistic of metadata for knowledge analysis, E. Diday
4. Simulation of uncertainty: Decision support in complex incompletely defined environments, M. von Rimscha
5. Are there any lessons to be learnt from the building of GLIMPSE? C.M. O'Brien
6. Handling imprecisely-known conditional possibilities, S. Amarger, D. Dubois and H. Prade
7. Combining symbolic and numerical methods for reasoning under uncertainty, P.J. Krause and J. Fox
8. Computationally intensive methods in the design of experiments, A.C. Atkinson
9. FRIL: A support logic programming system, J.F. Baldwin, T.P. Martin and B.W. Pilsworth
10. Computational models of diagnostic reasoning, A. Gammerman
11. A general numerical approach to the benchmark problems in defeasible reasoning, S.F. Roehrig
12. On the path to practical probabilistic reasoning, S.O. Kimbrough and S.F. Roehrig

This book consists of papers presented at a UNICOM conference of the same name, together with a few additional ones. It should be of interest to statisticians and computer scientists working with complex statistical problems requiring AI techniques and enhanced computer power to reach a solution; however, the conference was held in 1991 so the papers do not necessarily reflect the current state of the art. A wide variety of topics is covered, including representation of uncertainty, knowledge-based systems, genetic algorithms and computationally-intensive methods.

Contents: (A selection of the subsections of the Chapters is also given)
Prologue
PART I : Why We Need New Physics to Understand the Mind. The Non-Computability of Conscious Thought.
1. Conscience and computation.
1.1 Mind and science;
1.2 Can robots save this troubled world
1.3 The A.B.C.D, of computation and conscious thinking;
1.4 Physicalism vs mentalism
1.21 Is mathematical imagination non-computational?
2. The Gödelian Case
2.1 Gödel's theorem and Turing machines;
2.2 Computations ...
2.10 Further possible objections to G. Appendix A: An explicit Gödelizing Turing machine
3. The case for non-computability in mathematical thought
3.1 What did Gödel and Turing think?
3.2 Could an unsound algorithm knowably simulate mathematical understanding?
3.3 Could a knowable algorithm unknowably simulate mathematical understanding?
3.4 Do mathematicians unwittingly use an unsound algorithm?
3.5 Can an algorithm be unknowable?
3.6 Natural selection or an act of God? ...
3.28 Conclusions
PART II : What New Physics We Need to Understand the Mind. The Quest for a Non-Computational Physics of Mind
4. Does mind have a place in classical physics?
4.1 The mind and physical laws ...
4.5 Computation and physics
5. Structure of the quantum world
5.1 Quantum theory: Puzzle and paradox;
5.2 5.2 The Elitzur-Vaidman bomb-testing problem;
5.3 Magic dodecahedra;
5.4 Experimental status of EPR type Z-mysteries
5.18 The magic dodecahedron explained;
Appendix B: The non-colourability of the dodecahedron;
Appendix C: Orthogonality between general spin states
6. Quantum theory and reality
6.1 Is R a real process?
6.2 Many-worlds-type viewpoints;
6.3 Not taking |u> seriously ...
6.12 The new criterion
7. Quantum theory and the brain
7.1 Large-scale quantum action in brain function?
7.2 Neurons, synapses, and computers;
7.3 Quantum computation ...
7.12 EPR and time: Need for a new world view
8. Implications?
8.1 Intelligent artificial 'devices';
8.2 Things that computers do well or badly
8.7 Three worlds and three mysteries (EPR = Einstein, Podolsky, Rosen)

Readership: Anyone interested in whether "Artificial intelligence" is possible. Application is needed, not prior knowledge

The author further develops the viewpoint of his earlier book: The Emperor's New Mind [Short Book Reviews, Vol. 10, p.21]. Appropriate physical action of the brain evokes awareness, but this physical action cannot even be properly simulated computationally.
The many puzzles now presented by the physics of gravitation and of quantum phenomena, combined with those presented by Gödel and Turing, call for new developments which may clarify all these issues. Recent studies of brain development and function offer hope of such clarifications.

Contents:
0. Présentation
1. La période de gestation (1767-1783). Correspondance, écrits publiés et activités académiques
2. La période de gestation (1767-1783). Manuscrits inédits
3. L'Essai, le "Mémoire" et la Méthodique (1784-1787)
4. Manuscrits mathématiques inédits complémentaires
5. Autour des Elémens du calcul des probabilités
6. Autour des activités économiques, financières, sociales et politiques à l'aube de la Révolution(1786-1789)
APPENDICE: L'arithmétique politique et la Révolution Conclusion

Readership: Statisticians, political scientists, historians, historians of science

Condorcet (born 1743, died in prison in 1794) is best known in the history of statistics as the author of a pioneering 1785 book on mathematical pol-itical science, Essai sur l'application de l'analyse à la probabilité des décisions rendues à la pluralité des voix. As the secretary of the Académie des Sciences, as a mathematician, and as an engaged pol-itical philosopher in the most tumultuous period in French history, Condorcet was involved in or touched upon an amazingly diverse set of topics in social and mathematical science, including the mathematics and philosophy of probability, the law, inoculation, astronomy, demography, elections, economics, lotteries, games of chance, and insurance. Bru and Crépel's volume is a model of the highest order of historical scholarship; they not only present a wealth of Condorcet's hitherto unknown or inaccessible writing on these topics, they also present and discuss the historical context with exceptional care and insight and an unrivalled knowledge of the archival sources. For all those interested in Condorcet or in the history of mathematical social science in the 18th century, this is an essential volume.

Table des matières:
1. La variabilité
2. La probabilité
3. Intermède
4. La description
5. La recherche
6. L'imputation causale
7. Un nouveau regard sur la démarche statistique et sur ... "la science du particulier"
8. Statistique et médecine
9. Un combat d'issue ... incertaine

Les concepts de base de la pensée statistique sont exposés en mots, sans faire appel aux formules mathématiques (sauf quelques pages dans l'annexe). Comme l'auteur est spécialisé dans le domaine de la médecine, la plupart des illustrations viennent de la recherche médicale. Ce petit livre n'est pas très original pour les statisticiens mais pourrait servir comme culture générale pour le grand public.

Contents:
Philosophical Essay on Probability
Notes
Appendix: Editions of Essai
Bibliography
Glossary

Laplace's Essai philosophique sur les probabilités is one of the classics of our field. Published in its first version two centuries ago, it is an elegant treatment of mathematical statistics that is at once a non-technical popularization and an in-fluential philosophical treatise. The only previous English translation, widely available in a Dover edition, was prepared in 1902 by a professor of German and a professor of mechanics, an unfortunate pairing, since they evidently had limited facility with either French or probability. Dale's new translation is clearly superior both linguistically and scientifically to its predecessor, and it contains many bonuses.
This is a semi-variorum edition, translating the fifth edition of 1825 but indicating in copious footnotes the variations in that text from one of the editions of 1814. The editor's hundred pages of scholarly notes are particularly valuable, not least for their presentation of long extracts of sources to which Laplace only alludes. The Appendix gives an incomplete list of editions of the Essai, missing the earliest French version. Readers of French will prefer the scholarly edition in Laplace's own more elegant prose, published with notes and additional material by Bernard Bru in 1986 (Paris: Christian Bourgois), but others will welcome this overdue retranslation, despite its being marred by the inexplicable absence of an index.

Contents:
1. Introduction
2. Science and politics in the US
3. Prelude to the 1980 census: Issues in the 1960s and 1970s

4. Measuring and overcoming the undercount
5. How they did the census: District managers' stories
6. The 1980 Detroit case
7. The New York case, 1980-1987
8. Research towards adjustment, 1980-1987
9. 1987-1988, Three attacks on the census
10. Census year 1990
11. The decision not to adjust
12. Conclusions

This is the story of the conflict in the United States of America over the undercount in the 1980 and 1990 censuses. The focus is on the interplay between statistical science and politics, rather than on the more technical aspects of coverage measurement and adjustment. The book provides a glimpse into the workings of the Bureau of the Census and the political context in which it operates. As the author points out, the form and conduct of the census fit the kind of society we have, and the undercount is as much a feature of society as it is of the census. It is not surprising, therefore, that census adjustment cannot be dealt with as a purely technical issue. While the de-tails are specific to the United States, the book will be of interest to statisticians throughout the world.

Contents:
PART I : The Spiral of Practice
1. Introduction
2. The earthquake hut
3. A blundering oracle
4. Trial by water
5. Scars and healing: The power of seismic gaps
6. The best-laid plans
7. Earthquake hazards
PART II: The Spiral of Theory
8. Disaster theory
9. Theory of strong motion on soft ground
10. Science in ashes: A theory of coincidence
11. Conclusion
APPENDIX 1: Earthquake Disasters by Country
APPENDIX 2: Seismic Moments of Great Shallow Earthquakes 1900-1990

This is a wide-ranging, witty, iconoclastic review of earthquake prediction in the 1990's, not a success story despite some progress. The author speaks from a life-time involvement in seismology and first-hand knowledge of the subject and its practitioners in many different countries. His account is idiosyncratic, not always reliable, unlikely to win him friends in the establishment, but stimulating and entertaining reading. Statistical issues figure prominently throughout the book and form a focus for his criticisms of much recent work. Here also the treatment is controversial, but even where I am sceptical of the details of his discussion I am usually sympathetic to its general drift.

Contents:
1. Greek science
2. Later influences of the Greek authors
3. Microscopic world and the structure of living organisms
4. Genetics
5. Human genetics
6. Death rates and life tables
7. Evolution
8. Infectious diseases and microbiology
9. Puerperal sepsis
10. Wounds and hospital infections
11. Epidemiologic observations
12. Mathematics and epidemiology
13. Epidemiology and noninfectious diseases
14. Metrical characterizations of individuals and populations
15. Quantitative diagnostic and physiological methods
16. Classification of diseases
17. Numerical analysis of clinical experience
18. Modern clinical trials
19. Applications of mathematics to biology and medicine
Epilogue

Readership: Statisticians, medical and biological researchers
Professor Lancaster feels that biology has been neglected in studies of the history of science, that it is an exciting current area of investigation and that quantitative methods have had, and will have, a great impact on the study of biological and medical problems. This book provides considerable evidence in favour of this viewpoint and is full of interesting historical information. For example, there is speculation about an unknown mathematical adviser to Oliver Wendell Holmes in a study of puerperal fever and I was unaware of the work of Gavarret (1809-1890) who, it is claimed, may have given the first formal statement of the principles of medical statistics.
Although described as a historical essay, what has been provided is much more of a series of personal essays on selected topics of interest to the author. Thus I felt the book tended to lack cohesiveness and focus. There was also a selectivity to the aspects of each topic which were discussed. While some information, such as biographical details available else-where, was omitted intentionally, some discussions seemed to lack balance or completeness. Sometimes this derived from the unquestioned adoption of a particular perspective, sometimes from simplistic generalizations. An example of the latter was the grouping of clinical trials and statistical surveys as providing comparable information regarding questions of causation.
Whatever the possible weaknesses of the presentation, it is good that Professor Lancaster has provided his perspective on an important topic. This book will be valuable in motivating historical study and providing background information for teaching and research purposes.

Contents:
1. The dawning of statistics
2. Statistics becomes a distinct discipline
3. Researchers and statisticians come together
4. The awakening of statistics in the United States
5. Clinical trials in the United States
6. Designs and analyses of studies of historical importance

This is a very enjoyable book, both for its balanced contents and rich collections of portraits and photographs. The authors succeed in making light reading of a report of how new statistical ideas and methodologies were introduced and spread through the centuries and through the countries. They start from the early days of the registration of birth and death data, with the first studies in epidemics. They detail the period when mathematical statistics started in England with anecdotes like "Do you wonder where the k in Biometrika comes from?" and they link it to the awakening of statistics in the United States and the theoretical developments of statistics. They finally come to today's concerns on the design and analysis of clinical trials and epidemiological studies.

Contents:
1. The scope of survival analysis
2. Randomized clinical trials: General principles and some controversial issues
3. Estimation of survival probabilities
4. Non-parametric methods for the comparison of survival curves
5. Distribution times for failure time T
6. The Cox regression model
7. Validation of the proportional hazards model
8. Parametric regression models
9. The study of prognostic factors and the assessment of treatment effect
10. Competing risks
11. Meta-analysis
Readership: Biostatisticians, medical researchers, senior undergraduate students

This excellent book is a complete guide to the design, analysis and interpretation of survival data. Drawing examples from clinical trials performed to estimate and compare the survival of subjects receiving different treatment regimes, the authors begin by discussing the issues that need to be considered in the design and recruitment of subjects to such studies, stressing the central role randomization plays in ensuring the statistical validity of the conclusions. In the subsequent chapters, the non-parametric, parametric and regression methods of survival analysis are discussed in sufficient mathematical detail to satisfy most statisticians. These chapters would also be accessible to clinicians since all the methods discussed are illustrated with graphs or by numerical calculations. At every stage the authors point out both the strengths and weaknesses of the methods they discuss. A particularly attractive feature of the book is the very positive advice they offer. When warning about the pit-falls of a technique such as, for example, the over enthusiastic examination of selected subsets of the data, they offer a better way to proceed.
An entire chapter is devoted to the selection of prognostic and confounding variables in a survival regression model, which may be viewed as a special form of model selection and interpretation in the context of clinical trials. A chapter on competing risks ex-tends the survival analysis to multiple failures and the final chapter on a meta-analysis shows how to assess the evidence from a number of independent survival studies.
This book complements that of D. Collet Modelling Survival Data in Medical Research [Reviewed in Short Book Reviews, Vol. 14, p.25] in that it offers a deeper discussion of the design issues, competing risks and meta-analysis. On the other hand Collett gives more detail on the use of statistical packages for survival analysis.
The authors have done an admirable job in bridging the gap between the mathematical theory of survival analysis and its application in practice. This is an important book for statisticians since much of the discussion of practical issues has appeared in the medical literature not often read by statisticians. This is an important book for clinicians since they have a lucid exposition of a valuable tool in clinical research. It is highly recommended.

Sw.fr.45.00.
Contents:
0. Introduction
1. Exploratory and descriptive statistics
PART I : Exploratory Data Analysis
2. Stem and leaf
3. Rank measures
4. Box plots
5. Data transformations
6. Scattergrams
7. Regressograms
8. Time series
9. Two-way tables
PART II : Descriptive Statistics
10. Location parameters
11. Scatter measures
12. Correlation
13. Inequality and concentration
14. Index numbers
PART III: Graphical Techniques
15. 2-dimensional graphics
16. 3-dimensional graphics
17. Projection techniques
18. Postscriptum
Readership: Students and researchers in the social sciences using statistics

The second edition of the present textbook in the German "Springer-Lehrbuch" series has grown by fifty percent from the first edition (1988), mainly by improving on many details and adding more graphical examples. The goal of the book is to teach and demonstrate methods for exploratory, descriptive and graphical data analysis from a similar perspective to enable students to engage in model building without having to study probability theory.
The first part introduces Tukey's (1997) EDA to a German speaking audience in an informal way. Throughout the book, Austrian and Swiss sets of data from social sciences and economics are used. This makes the book particularly appealing to readers from this European Region.
The second part introduces more traditional non-inferential statistics in a similar way. In part three, the author discusses a set of selected topics from the vast field of statistical graphics. The author gives some guidance on practice ("not to lie with statistics"), but this is not done systematically. The chapter on projection techniques, Chapter 17, has nothing to do with modern methods for multivariate analysis. It deals with basic geometrical aspects of perspective representation, The book has been produced from a camera ready copy and this shows. The typesetting of mathematical expressions, for example, is quite poor. There are exercises at the end of each chapter, but no solutions are provided.

Contents:
1. Getting started
2. Simple regression plots
3. Two-dimensional plots
4. Scatter plot matrices
5. Three-dimensional plots
6. Visualizing linear regression with two predictors
7. Visualizing regression without linearity
8. Finding dimension
9. Predictor transformations
10. Response transformations
11. Checking models
12. Assessing predictors
13. Influence and outliers
14. Confidence region
APPENDIX: The R-code

Readership: Undergraduate statistics majors, postgraduates, researchers and lecturers

Ever since computer generated three-dimensional rotating plots became available it has been clear that they would be a powerful tool in data analysis. However, training in the use and interpretation of these plots for final year statistics majors has been out of the question because the software for DOS or Macintosh machines had not been readily available exceptin very expensive packages, but more importantly there was a lack of an accessible and up-to-date account of the subject. The appearance of this excellent introduction to regression graphics redresses these deficiencies and should have a significant impact on the teaching of multiple regression at both undergraduate and postgraduate levels. It comes with its own DOS/Macintosh software, UNIX being also available, software which in addition to the three-dimensional capability includes many other graphical features useful in regression analysis for example fitting of parametric and nonparametric smooths, trans-formations of the predictors or the response, linking, brushing and scatterplot matrices. The explanations in the book are clear, the use of the package is straight-forward and should pose few difficulties for any reasonable statistics majors. Well-chosen exercises at the end of each chapter help consolidate the ideas. The rotating plots are fascinating and play tricks with one's perception; it is possible to see certain points rotating in one direction whilst the rest rotate in the opposite direction! The wealth of information such plots convey can be bewildering. This package provides the reader with the tools necessary for a careful dis-section of this information leading to a better under-standing of the data. There is now no reason for not including these graphical techniques in undergraduate courses in multiple regression.

Contents:
1. Introduction
2. The S language
3. Graphical output
4. Programming in S
5. Distributions and data summaries
6. Linear statistical models
7. Generalized linear models
8. Robust statistics
9. Non-linear regression models
10. Modern regression
11. Survival analysis
12. Multivariate analysis
13. Tree-based models
14. Time series
15. Spatial statistics
APPENDIX A: Datasets and Software
APPENDIX B: Common S-PLUS Functions
APPENDIX C: S versus S-PLUS
APPENDIX D: Using S-PLUS Libraries
APPENDIX E: Command Line Editing
APPENDIX F: Answers to Selected Exercises

Readership: Aspiring, casual, and serious users of S-PLUS. Students and teachers of data analysis and statistics. Statisticians

Venables and Ripley have individually provided advice to the first generation of S-PLUS users via the Internet distribution list s-news and through their respective introductory guides. It is fitting that their collective wisdom, blended and extended, should appear in book form. This volume is an eminently usable introduction to the language and structure of S as well as a user's guide to the many statistical functions included in the S-PLUS package. Computer implementation of each statistical method is presented clearly and logically and is often accompanied by an outline of the underlying statistical idea. Graphs enhance the ex-position; helpful insights abound. Many of the techniques are illustrated by examples of S code (valuable "get started" modules for new users), often in terms of sets of data available to the reader on the ac-companying disk, which also contains supplementary S functions for neural nets, spatial statistics and miscellaneous statistical tasks.
This book is good value. It deserves to become the reference of choice for users of S-PLUS.

Contents:
1. Introduction
2. Basic sampling techniques
3. Further use of auxiliary information
4. Handling missing data
5. Linearization and sample re-use in variance estimation
6. Covariance-matrix estimation of ratio estimators
7. Analysis of one-way and two-way tables
8. Multivariate survey analysis
9. More detailed case studies
APPENDIX 1: Software Review for Survey Analysis
APPENDIX 2: SAS Macro

After a short introduction, Chapter 2 covers the basic sampling methods of simple random sampling, stratified sampling and selection with probability proportional to size. Chapter 3 introduces the use of auxiliary information to form model assisted estimators and compares ratio, regression, post-stratified and design-based estimators. This chapter is completed with a discussion of design effects which are used extensively in the rest of the book. In Chapter 4 methods of dealing with missing data through re-weighting and imputation are described and illustrated with a simple example. Methods of variance and covariance matrix estimation of a ratio estimator appear in the next two chapters including linearization, jackknifing and boot-strapping. The analysis, allowing for cluster effects, of one-way and two-way tables obtained from complex survey designs are illustrated using Rao-Scott adjustment to the Pearson and Neyman test statistics. Logit and linear models for proportions are considered in the penultimate chapter and the text concludes with three further practical examples.
The first half of the book would form a comprehensive introduction to survey sampling methodology at the undergraduate level, while selected topics from the later chapters would be suitable for a more advanced course. The presentation of the book and the use of several small sets of data throughout make this a most readable and instructive text.

Contents:
1. Preliminary notions
2. Notions in significance testing of hypotheses
3. Review of special distributions
4. Exact nonparametric method
5. Generalised p-values
6. Generalised confidence intervals
7. Comparing two normal populations
8. Analysis of variance
9. Mixed models
10. Regression
APPENDIX A: Elements of Bayesian Inference
APPENDIX B: Technical Arguments

Readership: Anyone with elementary knowledge of probability and statistics who is interested in analyzing some data

If, with given n, a binomial B(n,p) X has been observed to equal x0, discreteness makes exact pre-fixed level á testing of H0:p < p0 against H1:p > p0 impossible without resorting to an absurd randomization. Most data analysts wishing to make an objectively based response will in this situation quote the p-value Pr(X > x0|p0), unless additional information justifying a prior density (p) is available. This p-value is exact in the sense of this book's title. In the Behrens-Fisher problem we could find a similarly exact p-value for ì1 < ì2 against ì1 > ì2 if only we knew , the ratio of the unknown variances. The ratio of observed sample variances can be used in an exact pivotal for . A plot of p against , together with a confidence distribution for would constitute an "exact" "generalized" p-value response to the problem, as also would the integral of p over a posterior distribution of were the necessary prior for available.
Weerahandi correctly points out that our founding fathers, the Pearsons, Gosset, Fisher, Yates et al., appeared to sanction pre-fixed level tests only because modern computers were not available to them.
A book addressed to people who hope to extract useful messages from given data without being unduly bothered with considerations of an excessively formal kind makes refreshing reading. Another welcome feature is, that Bayesian methods are not segregated from non-Bayesian methods, while Appendix A sets out the Bayesian versus non-Bayesian issues very clearly. Up-to-date references are made to packages such as StatXact and TESTIMATE, but the "exact" in the book's title does not exclusively refer to these.

Contents:
12. Continuous distributions (general)
13. Normal distributions
14. Lognormal distributions
15. Inverse Gaussian (Wald) distributions
16. Cauchy distribution
17. Gamma distributions
18. Chi-square distributions including chi and Rayleigh
19. Exponential distributions
20. Pareto distributions
21. Weibull distributions

Readership: Pure and applied statisticians, researchers in continuous distribution theory, scientists who use continuous distributions

This is the second book of Johnson and Kotz's classic 1969-72 reference series Distributions in Statistics to appear in a greatly enlarged second edition. For this volume, the co-author is Dr. N. Balakrishnan. The book is about two and one-half times the length of the first edition: the average textual content of each chapter is about doubled, whilst the number of references in the bibliographies is more than trebled. The general style and structure is the same as that of the first edition. As before, the chapter numbering follows on from the 'discrete' book, [Short Book Reviews, Vol. 13, p.17.]. However, the old Chapter 17 topics now fill two chapters (17 and 18); as a result, the old Chapter 21 on extreme value distributions has been transferred to the second 'continuous' volume which is to appear shortly. The order in which different aspects of distributions are dealt with varies from chapter to chapter. Much of the material in the first edition has been reproduced almost verbatim, though with some rearrangement; the very extensive new material appears both as complete additional sections and also as insertions or, more often, lengthy append-ages to the existing material. In several chapters there is now a section devoted to the computer generation of the particular distribution(s). The subject index is a great improvement on its predecessor and the new author index is particularly useful since, as be-fore, there is a separate bibliography for each chapter. The authors are to be congratulated on a very thorough and very substantial update which, with its companion volume, will remain the reference work on univariate continuous distributions for many years.

Contents:
1. Introduction
2. Mathematical preliminaries
3. Asymptotic analysis
4. Univariate integrals
5. Multivariate Laplace type integrals
6. Approximations for normal integrals
7. Arbitrary probability integrals
8. Crossing rates of stochastic processes

This book deals with asymptotic approximation methods for multivariate integrals, with special attention to examples from reliability theory, stochastic optimization and mathematical statistics. Most of these approximation formulas can be found in various existing textbooks on the subject, for example, Watson's lemma on Laplace transforms, Laplace's methods, ... . The analytic approximation formulae brought together in this book could be useful in those problems where more is needed than just a numerical evaluation of an integral.

Contents:
1. Foundations of probabilistic causality
2. Predictive causal inference in a series of events
3. Confidence statements about the prediction process
4. Applications
5. Concluding remarks
APPENDIX 1: Derivatives of the Prediction Probabilities
APPENDIX 2: Results of Estimated Hazard Models

The author extends the classical theory of probabilistic causality to longitudinal settings, suggesting that causal analysis is essentially dynamic analysis, and formulates the sequence of causal events as a marked point process. He adopts the definition that the influence of a cause is the difference between the prediction probability of the effect given that the cause has occurred and the probability given that it has not occurred. He presents a short but interesting discussion of philosophical and probabilistic ideas in causality and distinguishes between explanation ('What were the causes of E?') and prediction (a generic description that 'usually' Type C events precede Type E events) and focuses on the latter, extending it to a causal chain of events.
As has been emphasized in work on expert systems, probability serves to model both the uncertainty about causal relationships as well as the inherent indeterminism of nature; in any particular application, we need to know the general relevant causal laws and the particular situation. Statistical modelling is used to model these to determine which conditions produce which events preceding the outcome. This allows us to calculate the probabilities of different routes to the outcomes and probabilities of the different outcomes.
This book is not merely a theoretical exposition, the author also applies the ideas to real data.
The subject is a deep one, but the author has given a very clear presentation. Even more impressive is that he has done this in a mere one hundred and thirty pages. I certainly recommend the book to anyone interested in the area.

Contents:
1. Introduction to reliability
2. Lifetime distributions
3. Analysis of lifetimes with covariates
4. System reliability
5. Models for repairable systems
6. Analysis of repairable systems
7. Growth and adaptive models
8. Dependency analysis
9. Practical aspects of reliability data analysis
10. Case studies in reliability

This book provides a reasonably comprehensive but brief overview of statistical methods for reliability. It covers much of the same ground as the recent monograph Statistical Analysis of Reliability Data, by M.J. Crowder, A.C. Kimber, R.L. Smith and T.J. Sweeting [Short Book Reviews, Vol. 12, p.6]. The Crowder et al. book has a superior presentation of statistical techniques in my opinion, but the current book has some useful discussion on practical aspects of reliability studies. In terms of its stated aim of providing practitioners with guidance on the use of statistical techniques, it is fairly successful.

Contents:
1. Introduction
2. Decision-theoretic foundations of statistical inferences
3. From prior information to prior distributions
4. Bayesian point estimation
5. Tests and confidence regions
6. Admissibility and complete classes
7. Invariance, Haar measures, and equivariant estimators
8. Hierarchical and empirical Bayes extensions
9. Bayesian calculations
10. A defense of the Bayesian choice
APPENDIX A: Usual Probability Distributions
APPENDIX B: Usual Pseudorandom Generators

Readership: Advanced undergraduate and postgraduate students and teachers of theoretical statistics

This book provides a traditional approach to the subject of Bayesian statistics. It contains many references to results and papers from the period 1985-1994 and, in this way, up-dates the book Statistical Decision Theory and Bayesian Analysis by J.O. Berger [Reviewed in Short Book Reviews, Vol. 6, p.6]. The use of numerical approximations and sampling techniques in Bayesian computation is one important new area covered. Others are of a more theoretical nature. The tone of the book is academic and, as a result, requires the reader to have a stronger general background and sophistication in statistical theory than does the book by Berger. However, its role in the teaching of Bayesian statistics is assured, not simply because of its up-to-date nature, but also for the large number of exercises, mainly theoretical, at the end of each chapter. These will be appreciated by students and teachers alike.

Contents:
PART A: Dynamic Bayesian Modelling
1. Practical modelling and forecasting
2. Methodological framework
3. Analysis of DLM
APPENDIX 3.1 Review of Distribution Theory
APPENDIX 3.2 Classical Time Series Models
4. Application: Turkey chick sales
5. Application: Market share
6. Application: Marriages in Greece
7. Further examples and exercises
PART B: Interactive Time Series Analysis and Forecasting
8. Installing BATS
9. Tutorial: Introduction to BATS
APPENDIX 9.1: Files and Directories
10. Tutorial: Introduction to modelling
11. Tutorial: Advanced modelling
12. Tutorial: Modelling with incomplete data
13. Tutorial: Data management
PART C: BATS Reference
14. Communications
15. Menu descriptions

The book provides a "hands-on" introduction to time series analysis suitable for non-statistics majors. The emphasis is on real sets of data. A key aspect of the book is the provision of the BATS (Bayesian Analysis of Time Series) program for both DOS and Windows environments. A significant proportion of the book is devoted to documenting this program and illustrating its use on time series drawn from different application domains.
The book would be an useful introduction to time series analysis with emphasis on manipulation of real sets of data. Relatively little mathematical sophistication is called for in reading the book and thus it is suitable for majors, for example, in business studies or operations research.

Contents:
1. Introduction
2. Univariate kernel density estimation
3. Bandwidth selection
4. Multivariate kernel density estimation
5. Kernel regression
6. Selected extra topics
APPENDIX A: Notation
APPENDIX B: Tables
APPENDIX C: Facts About Normal Densities
APPENDIX D: Computation of Kernel Estimations

Much of the book is taken up with nonparametric density estimation using the kernel density approach. A substantial amount of detail concerning asymptotic behaviour of these estimates is provided because the authors feel that this also sheds considerable light on the behaviour of kernel smoothing estimates of nonparametric regression functions. One chapter is reserved for local polynomial smoothing of bivariate data. Since, as the authors point out, kernel smoothing in the sense of local averaging has fallen somewhat out of favour, many readers may find that the title of the book is somewhat misleading, but this chapter is nevertheless a welcome contribution.

Contents:
PART I : Introduction
1. Preliminaries
2. Discriminant analysis in research
PART II : Prediction
3. Basic ideas of classification
4. Multivariate normal rules
5. Classification results
6. Hit rate estimation
7. Effectiveness of classification rules
8. Selecting and ordering predictors
9. Two-group classification
10. Nonnormal rules
11. Reporting results of a PDA
12. Applications of PDA
PART III: Description
13. Group separation
14. Assessing effects
15. Describing effects
16. Selecting and ordering variables
17. Reporting results of a DDA
18. Applications of DDA
PART IV : Issues and Problems
19. Issues
20 Special problems
APPENDIX A: Data Set Description
APPENDIX B: Computer Printouts
APPENDIX C: Content of Accompanying Diskette

Readership: Users of multivariate techniques, mainly those in the field of education

The author's interpretation of discriminant analysis allows both for the construction of classification rules, so-called Predictive Discriminant Analysis, as well as multivariate analysis of variance related techniques (Descriptive Discriminant Analysis). Mainly based on data from education, a rather superficial introduction to the various techniques is given. Computer output (more than 100 pages of SAS, SPSS, BMDP output) on the analyses of four sets of data should help the reader in making the transition from theory to practice. Logistic discriminant analysis is only briefly mentioned and topics like neural nets or CART do not appear. The four sets of data presented all come from educational studies. The avoidance of basic mathematical statistics leads to definitions like that on page 81 "E(R) denotes the 'expected value of R'. Suppose that all possible samples are selected and that for each sample, an R value is calculated; the mean of the collection of R values is the expected value of R." I personally like the idea of reviewing a field of statistics using detailed analyses of substantial sets of data based on existing statistical software. I would, however, have preferred a more in depth, as well as more complete, discussion of the techniques used together with greater diversity in the choice of data. The space needed for the former could have been found by restricting the computer output to what is strictly necessary.

Contents:
Chapter I : Brownian Motion
1. Introduction
2. Basics about Brownian motion
3. Brownian motion in higher dimensions
4. Gaussian processes and Lévy processes
Chapter II : Some Classical Theory
1. Basic measure theory
2. Basic probability theory
3. Stochastic processes
4. Discrete-parameter martingale theory
5. Continuous-parameter supermartingales
6. Probability measure on Lusin spaces
Chapter III: Markov Processes
1. Transition functions and resolvents
2. Feller-Dynkin processes
3. Additive functionals
4. Approach to Ray processes: The Martin boundary
5. Ray processes
6. Applications

Readership: All those interested in a more in-depth treatment of stochastic processes

This second edition to the wonderful 1979 first edition by D. Williams alone has changed in many ways. Having taken C. Rogers as co-author on board, a completely new book emerged resulting in an extra one hundred and fifty pages. Especially the Chapters I and II have been extensively rewritten, offering many more detailed results and proofs. I fully agree with the authors when they say in the preface that "Chapter II is now a highly systematic account of what every young probabilist must know." The main feature of the book lies in the fact that the authors nearly always convey to the reader first why new (often difficult) results are needed, then go on by showing how these results are to be used before embarking on a proof. Together with D. Williams' Probability with Martingales [Reviewed in Short Book Reviews, Vol. 11, p.29] and Volume 2 of Itô Calculus [Reviewed in Short Book Reviews, Vol. 8, p.8] by the same authors, these books offer a very concise account of modern probability theory. Every student and indeed researcher in probability should have these volumes within easy reach.

Contents:
1. Univariate stochastic orders
2. Univariate variability orders
3. Univariate monotone convex and related orders
4. Multivariate stochastic orders
5. Multivariate variability and related orders
6. Stochastic convexity and concavity
7. Some applications of multivariate variability orders
8. Statistical inference for stochastic ordering
9. Association and unbiased tests in statistics
10. Information orderings and stochastic orderings
11. Stochastic orderings of epidemics
12. Comparing risk and risk aversion
13. Scheduling
14. Stochastic comparisons in closed Jackson networks
15. Comparison of maintenance policies
16. Stochastic order in system reliability theory

This is a welcome book, for many reasons. Stochastic orders have become important tools in diverse areas of applied probability. These orders, their properties and their interrelationships, are systematically described in the first part of the book. Although the style is somewhat terse, the treatment is comprehensive, and the book will provide an invaluable reference. Fourteen different authors contributed the final ten chapters of the book. Each chapter provides a relatively self-contained treatment of its subject. They are valuable not only in their own right, but in illustrating the stochastic orders and in indicating the breadth of their application.

PART I : Introduction of Models of Manufacturing Systems
1. Concepts of hierarchical decision making
2. Models of manufacturing systems
PART II : Optimal Control of Manufacturing Systems: Existence and Characterisation
3. Optimal control of parallel machine systems
4. Optimal control of dynamic flowshops
PART III: Asymptotic Optimal Controls
5. Hierarchical controls in systems with parallel machines
6. Hierarchical control in dynamic flowshops
7. Hierarchical control in dynamic jobshops
8. Hierarchical production and set up scheduling in a single machine system
9. Hierarchical feedback controls in two-machine flowshops
PART IV : Multilevel Hierarchical Decisions
10. A production of capacity expansion model
11. Production-marketing systems
PART V : Computations and Conclusions
12. Computations and evaluation of hierarchical controls
13. Further extensions and open research problems

Readership: Operations researchers, system and control theorists, applied mathematicians

The book is concerned with manufacturing problems described in a probabilistic or stochastic setting. It is further assumed that some variables, for example demand, change rapidly in comparison with other events, breakdowns, etc. In this case, it is shown that a two-time scale approach to optimization can lead to near optimal results. In this approach the fast changing processes are replaced by their long-run averages to simplify the problem. The book is recommended to anybody who has an interest in applied stochastic processes or manufacturing problems.

PART I : Introduction
1. Hidden Markov model processing
PART II : Discrete-Time HMM Estimation
2. Discrete states and discrete observations
3. Continuous-range observations
4. Continuous-range states and observations
5. A general recursive filter
6. Practical recursive filters
PART III: Continuous-Time HMM Estimation
7. Discrete-range states and observations
8. Markov chains in Brownian motion
PART IV : Two Dimensional HMM Estimation
9. Hidden Markov random fields
PART V : HMM Optimal Control
10. Discrete-time HMM control
11. Risk-sensitive control of HMM
12. Continuous-time HMM control
APPENDIX A: Basic Probability Concepts
APPENDIX B: Continuous-time Martingale Representation

Readership: Mathematicians with interests in stochastic dynamic systems or signal processing

The book uses change-of-probability measure techniques to study estimation and control problems associated with Hidden Markov Models (HMM's). The change-of-probability measure technique is used to convert potentially difficult problems into a simple form such that well-known results for identically and independently distributed random variables can be used. The book begins with a treatment of discrete-time, discrete state HMM's and then proceeds to more difficult problems including continuous-time HMM's and two-dimensional image processing.
The book is a tribute to the power of the change-of-probability measure techniques and is recommended to anybody with an interest in stochastic dynamic systems of HMM's. A reasonable degree of mathematical sophistication is required to read the book.

Contents:
PART I : Environmental Hazards
1. Spatial pattern analyses to detect rare disease clusters
2. Assessing toxicity of pollutants in aquatic systems
3. Prediction models for personal ozone exposure assessment
PART II : Forestry, Fisheries, Genetics
4. Measurement error models for gypsy moth studies
5. Estimating pine seedling response to ozone and acidic rain
6. Geostatistical estimates of scallop abundance
7. Survival analysis for size regulation of Atlantic halibut
8. Mixture fraction and linkage analysis for hybrid onions
PART III: Habitat and Animal Studies
9. Spatial association learning of hummingbirds
10. Habitat association studies of the northern spotted owl, sage grouse, and flammulated owl
11. Time-series analyses of beaver body temperatures
PART IV : Health Care and Public Health Policy
12. Parametric duration analysis of nursing home usage
13. Analysis of attitudes toward workplace smoking restrictions
PART V : Clinical Trials
14. Interpretation of a Leukaemia trial stopped early
15. Early lung cancer detection studies
16. Modeling interrater agreement for pathologic features of Choroidal Melanoma
17. Quality control for bone mineral density scans
PART VI : Epidemiology, Toxicology
18. Modeling the precursors of cervical cancer
19. Patterns of lung cancer risk in ex-smokers
20. Two-stage sampling designs for adolescent depression studies
21. Drug interactions between morphine and marijuana

A glance at the above list of twenty-one topics dealt with in this collection immediately reveals that it is about a wide variety of problems from real life. Each of these practical cases is well described and modern statistical methods are applied to analyze the data and to come to relevant conclusions in the biological, medical or sociological problem. The book is more than just a collection of articles. The editors did a great job in presenting each of the studies in one and the same format with the following sections: 1. Motivation and background, 2. Data, 3. Methods and Models, 4. Results, 5. Conclusions, Software Notes and References - Questions and Problems. A small selection of the methods used in the text are: dose-response curves and surfaces, generalized linear models, resampling, regression, longitudinal data, repeated measures, survival analysis, time series. The original sets of data used by the various authors are included on an accompanying diskette which will be of great use for students.
Case Studies in Biometry is a work of high quality that presents applied statistics in a very attractive way. It is dedicated to the memory of Myrto Letkopoulou. I am sure that she would have enjoyed this book too.

Contents:
1. Introduction
2. Risk factors for infection and the probability of HIV transmission
3. Surveys to determine seroprevalence and seroincidence
4. The incubation period distribution
5. Cofactors and markers
6. Screening and the accuracy of tests for HIV
7. Statistical issues in surveillance of AIDS incidence
8. Back-calculation
9. Epidemic transmission models
10. Synthesizing data sources and methods for assessing the scope of the epidemic
11. Developing and evaluating new therapies and vaccines

The intensive study of the AIDS epidemic that has taken place over the last ten years has involved a wide range of epidemiological issues, many of them involving novel and demanding technical statistical problems. Special considerations arise, for example, from the long and variable incubating period, the difficulty in most cases of identifying the time of infection, as well as from the obviously sensitive nature of the field.
This book, by two workers who have made very important contributions of their own, is an impressive and authoritative account of the subject up to about the end of 1992; the authors' preface is dated January 1993. The emphasis throughout is on the substantive interpretation, but where new statistical ideas are needed these are described concisely and lucidly.
While the book is addressed to specialists, a general statistical reader who has not followed work on AIDS is likely to be interested and impressed by the wide range of ideas, techniques and special methods involved.
It is inevitable and appropriate that the book is very largely based on North American experience, but there is some tendency to overlook work in other countries.

Contents:
PART I : Preliminary Considerations
1. Introduction
2. Nature and purposes of clinical testing
3. The role of the FDA
PART II : Planning Your Trial
4. Planning your trial I
5. Planning your trial II
6. Planning your trial III
7. The protocol and clinical report form
PART III: Some experimental designs, with examples
8. One-factor parallel designs
9. Multifactor parallel designs
10. Parallel factorial designs: Examples
11. Parallel factorial designs: Blocking
12. Changeover designs
PART IV : Basic Statistical Analysis
13. Data analysis: Fundamental ideas
14. Data analysis: Basic statistical resources
PART V : Details of Sample Size Estimation
15. Sample size estimation
Readership: Clinical investigators, clinical research associates, statisticians involved in designing and analyzing clinical trials

This book aims to explain the reasons under-lying the detailed procedures involved in planning a clinical trial. The emphasis is on experimental design rather than analysis and throughout the text the reader is encouraged to understand as much as necessary about the statistical aspects of the trial, but to take ad-vantage of the services of a professional statistician if available. Parts I and II cover practical features of setting up a trial, including obtaining regulatory approval, writing the protocol, selecting a suitable design and arranging the randomization. Problems of multiplicity and the effects on Type I errors and sample size are considered in some detail. Part III examines, by example, a variety of factorial designs, and changeover designs, introducing the concepts of interactions, confounding, fractional replications and blocking. Again the emphasis is on the reasons for choosing a particular design and its practical application, including sample size determination and patient randomization. Simple statistical methods are covered in Part IV, including t-tests, and non-parametric tests such as Wilcoxon, Mann-Whitney and chi-squared tests. The final section, Part V, gives details of sample size estimation for most of the design situations discussed earlier in the text. A most useful set of sample size tables is included in the Appendix.

Contents:
PART 1 : Rainfall and Climate
PART II : Sea Levels and Wave Energy
PART III: Water Quality, Supply and Management
PART IV : Hydrological Modelling

This volume is a sequel to Statistics for the Environment, 1993 [Short Book Reviews, Vol.14, p.12]. It contains nineteen of the papers presented at the SPRUCE 2 environmental conference held at Rothamsted Experimental Station in the United Kingdom during 1993. As with the first volume, this is an integrated and commendably readable set of conference proceedings. It should be of interest both to statisticians looking for sophisticated examples of statistical modelling, and to hydrologists looking for assistance from statistics, although they may be disappointed to find that `Acid Rain' is not featured here. The last paper is by a self-confessed "Confused Hydrolic Modeller" who concludes, amongst other things, that hydrologists must learn to distinguish between mathematical models and the real-world phenomena they model. This seemingly innocent observation induces much heart-searching, although he will have been reassured by the preceding papers in this collection.

Contents:
1. Why understand 'statistics'?
2. Probability in engineering decisions
3. Justifying engineering decisions
4. Modelling variability
5. Combining variables
6. Precision of estimates
7. Asset management plan
8. Making predictions from one variable
9. Making predictions from several explanatory variables
10. Design experiments
11. Modelling variability in time and space

Engineering students will find this a stimulating introduction to statistics with plenty of interesting sets of data having a clear engineering flavour. The material is well motivated and enough theory is covered, together with appropriate exercises, to enable important ideas of regression and experimental design to be understood. The objective is not so much to turn engineering students into statistical consultants; rather it is to give them a certain basic knowledge together with a good idea of the extent and nature of the information that can be extracted from a statistical analysis. This is provided via a series of case studies utilizing MINITAB that may appear in the latter part. The author indicates that the book may be used by an engineer needing to design an experiment or survey and has provided flow charts to identify efficiently those sections of the book which are relevant to the problem at hand. However readers may find that the case studies are too specific to handle their particular problem and one would hope that they would conclude that the next step is to go directly to a statistician. Engineers should come away from this book recognizing that statistical methodology has consider-able relevance for them.

Contents:
Introduction
1. Statistics of the straight line
2. Structures of two-way tables
3. Functional expressions of structure
4. The row-linear and concurrent structures
5. Graphical representation of the row-linear structure
6. Some analysis of variance
7. A general representation of two-way arrays
8. A systematic approach to the analysis of two-way arrays
9. Two-way analysis with replication within cells
10. Interlaboratory testing
11. Missing values
12. Curve fitting
13. Two-way tables with quantitative labels
14. Applications to real data sets
15. One-way tables and the h statistic
16. How useful are models?

The theme of this book is that we should allow the data to suggest a model. This is therefore not a text on analysis of variance. The data in this context are observations on continuous variables arranged in a rectangular array determined by two explanatory variables. There is a wealth of interesting industrial examples which are analyzed in detail, with the minimum of distributional assumptions. A number of potential structures for data arranged in a two-way layout are examined. Of these, the central one is the row-linear structure in which for each row, the expected response is linear in the column effect. There is a very strong emphasis on graphical methods and little in the way of point estimation or significance tests.
Statisticians as well as scientists may well find something of interest here in spite of a few minor irritations: random variables not differentiated from their expected values, estimates of variances confused with their population values and a large sprinkling of misprints.

Contents:
1. Translator's introduction
2. Pars prior/Part one
3. Pars posterior/Part two
4. Supplementum/Supplement
5. Anzeigen/Notices
Afterword

Readership: Statisticians, probabilists, numerical analysts, historians of science

Gauss published his first work on least squares in 1809, but his mature reconsideration of the subject was published over a decade later, in Latin and in parts, as Theoria Combinationis Observationum Error-ibus Minimis Obnoxiae (1823-1828). In 1855 Joseph Bertrand translated this (and Gauss's other works on least squares) into French, and that French translation served as the source for later translations into German (1887) and English (by Hale Trotter in 1957, but only circulated in mimeograph form). G.W. Stewart has per-formed for the profession a signal service by providing an all new translation from the Latin of the Theoria Combinationis.
This work of Gauss's is the source of what has come to be known as the "Gauss-Markov Theorem", as well as many other elegancies (such as a Chebychev-like in-equality that Richard Savage wrote about in 1961 in the U.S. NBS Journal of Research 65 B:211-222). The translation is exceedingly well done, with the Latin original being presented on facing pages. It flows easily for a modern reader, and historians can consult the facing original for reassurance that this flow has not been bought at the cost of accuracy. The Contemporary German 'Notices' (essentially, abstracts) are also included. The regrettable lack of an index is partially recompensed by an excellent, detailed table of contents that amounts to a section-by-section synopsis of the work. Stewart has added an Afterword that sets the work in an historical context. It is a pleasure to see Gauss back in print in such a splendid edition.

Contents:
1. Uncertainty in forensic science
2. The evaluation of evidence
3. Variation
4. Historical review
5. Transfer evidence
6. Discrete data
7. Continuous data
8. DNA profiling

The book is timely: with the recent controversy over the statistical interpretation of DNA pro-file evidence, forensic scientists are becoming more aware both of the role of statistics in their work and of the subtlety of many statistical ideas. Students of statistics will also find that forensic science provides a fascinating field of application, not least because even the most hardened anti-Bayesian must con-cede that the Bayesian approach to inference is well suited here, in contrast with frequentist approaches which lead to substantial difficulties.
Aitken introduces elements of Bayesian statistical inference in an informal way, accessible to those with little mathematical background. The theory is illustrated with examples of transfer evidence such as hair, glass fragments and DNA profiles. It is on the latter topic that the book is most disappointing. The author ignores the genetic correlations which are crucial to an appropriate analysis and is consequently led to the slippery concepts of a "random" person drawn from a "relevant population": these concepts are not needed and cause confusion. The fallacy that the ethnicity of the suspect is irrelevant to inference is asserted repeatedly: a careful analysis shows that the ethnicities of both the suspect and the possible culprits are relevant to inference. Kernel density estimation is advocated inappropriately when the underlying density is, due to population genetics, very spiky. The book has much to recommend itself to forensic scientists as an introduction to relevant statistical ideas, but a much-needed thorough statistical treatment of DNA profile evidence is still awaited.

Contents:
1. Clinical problems and statistical solutions
2. Diagnosis, probability and sampling
3. The variability of clinical measurements
4. Sampling and estimation
5. The search for associations
6. Treatment trials
Postscript: Hypothesis testing and P-values

This book is intended as a main text for a course on the critical appraisal of statistical evidence in a medical context. It is intentionally nontechnical and contains very few formulae. Rather, the focus is on many of the key concepts arising in the interpretation of statistical evidence in medical studies. The chapters are easy to read and thus make this a very approachable book for medical investigators who wish to gain a greater appreciation of the topic.

Contents:
1. Introduction
2. Assay development
3. Assay validation
4. Scaleup design and analysis
5. USP tests and specifications
6. Process validation
7. Quality assurance
8. Stability studies
9. Accelerated testing
10. Design for long-term stability studies
11. Stability analysis with fixed batches
12. Stability analysis with random batches

Readership: Statisticians involved in the pharmaceutical industry; pharmaceutical scientists and pharmacists

This book aims to give a comprehensive presentation and discussion of designs and analyses used in drug development, including those relevant to assay development and validation, process validation, quality assurance, stability assessment, and accelerated testing. Numerous examples are used to illustrate the methods that are presented. A basic knowledge of regression and analyses of variance is assumed. This text appears to be a useful reference for practitioners as well as a good source for statisticians wishing to understand some of the key issues in these aspects of drug development.

Contents:
1. Introduction
2. Overview of the methods
3. Planning the study
4. Information retrieval
5. Data collection
6. Advanced issues in meta-analysis
7. Statistical methods in meta-analysis
8. Other statistical issues in meta-analysis
9. Complex decision problems
10. Estimating probabilities
11. Utility analysis
12. Advanced cost-effectiveness analysis
13. Sensitivity analysis
14. Reporting results
15. Limitations

Readership: Epidemiologists, biostatisticians, researchers working with meta-analysis

The aim of this book is to bring together three research areas, meta-analysis, decision analysis and cost-effectiveness analysis, that are integral parts of policy recommendations in medicine. Each area is described and then used in an integrated way in case studies which are designed as prototype analyses. The descriptions of the three areas are not treated too expansively, since each would require a book-length description; thus some prior knowledge is important. However, the descriptions are helpful in pin-pointing additional reading. Although the book is designed for a medical audience, the methodology can be transferred to other contexts, in order that the methodology will be of interest to a broader audience.

Contents:
1. Cars, goats, and sample spaces
2. How to count: Birthdays and lotteries
3. Conditional probability: From kings to prisoners
4. The formula of Thomas Bayes and other matters
5. The idea of independence, with applications
6. A little bit about games
7. Random variables, expectations, and more about games
8. Baseball cards, the law of large numbers, and bad news for gamblers
9. From traffic to chocolate chip cookies with the Poisson distribution
10. The desperate case of the gambler's ruin
11. Breaking sticks, tossing needles, and more: Probability on continuous sample spaces
12. Normal distributions, and order from diversity via the central limit theorem
13. Random numbers: What they are and how to use them
14. Computers and probability
15. Statistics: Applying probability to make decisions
16. Roaming the number line with a Markov chain: Dependence
17. The Brownian motion, and other processes in continuous time

As the chapter headings show, this is not about the masochistic pleasures of gambling, but is an introduction to the ideas of probability using the examples that we employ to keep our classes alert and entertained. Most of the classics are here: the birth-day problem, the prisoner's dilemma, the coupon collector's problem, Buffon's needle problem, etc., ex-tending as far as the Hardy-Weinberg law. Surprisingly, Simpson's paradox is missing. These examples are clearly explained, the mathematics is informal and almost everywhere simple, calculus being mentioned only briefly. The general reader, who is new to the mathematics of probability, and not interested primarily in its application, will find a feast here. The teacher will have on his shelf a useful collection of examples.
A quibble at the editorial level is that a solidus / is used for conditioning, instead of a vertical rule |, so that we have, for example, P(U/V)=2/3.

Contents:
22. Extreme value distributions
23. Logistic distribution
24. Laplace (double exponential) distributions
25. Beta distributions
26. Uniform (rectangular) distributions
27. F-distributions
28. t-distributions
29. Noncentral 2-distributions
30. Noncentral F-distributions
31. Noncentral t-distributions
32. Distributions of correlation coefficients
33. Lifetime distributions and miscellaneous orderings

Readership: Pure and applied statisticians, researchers in continuous distribution theory, scientists who use distributions

Volume 1 of the second edition of Continuous Univariate Distributions was reviewed in Short Book Reviews, Vol. 15, p.25. As one might expect, all the general comments in that review apply equally well to Volume 2, which has the same style, structure, and approach as Volume 1. In particular, the average chapter length is again about double that of the first edition and the number of references has trebled. The chapter numbering now corresponds rather less well with the first edition, partly as a consequence of changes in Volume 1 and the decision to delay revision of the chapter on quadratic forms to the projected multivariate volume. The final chapter, previously entitled Miscellaneous Distributions, has been very substantially revamped and more narrowly focused; it is now entitled Lifetime Distributions and Miscellaneous Orderings.
These two volumes, together with the 1992 'discrete' revision by Johnson, Kotz and Kemp (A.W.) [Short Book Reviews, Vol. 13, p.17], now form a remarkably comprehensive, up-to-date, and indispensable guide to univariate distributions. We look forward eagerly to the projected 'multivariate' revision. All the authors deserve both our congratulations and our gratitude.

Contents:
1. Probability
2. Potential theory
3. Lipschitz domains
4. Singular integrals
5. Analytic functions

In my review of Stromberg's Probability for Analysts [Short Book Reviews, Vol. 15, p.43] I already stressed the growing number of books treating the interplay between analysis and probability. Here is one more, also written by a distinguished researcher in the field. And again, an interesting text emerges. After giving the necessary background material in Chapters 1 and 2, the author embarks on a discussion of applications of probabilstic techniques to problems in analysis all being established as from 1970. In that sense, the book is perhaps a bit more specialized than some of its competitors. This is however compensated by the fact that this allows for a more in-depth discussion of the treated analysis problems. Numerous exercises enliven the presentation. The research student will welcome the various open problems presented making the text useful for a broad, but necessarily highly dedicated, readership.

Contents:
1. Introduction
2. The basic linear multilevel model and its estimation
3. Extensions to the basic multilevel model
4. The multivariate multilevel model
5. Nonlinear multilevel models
6. Models for repeated measures of data
7. Multilevel models for discrete response data
8. Multilevel cross classification
9. Multilevel event history models
10. Multilevel models with measurement errors
11. Software for multilevel modelling; missing data and multilevel structural equation models

Readership: Statisticians, educational and social researchers, medical researchers, psychologists

This book first appeared in 1987 as Multilevel Models in Educational and Social Research [Short Book Reviews, Vol. 8, p.2]. Then it had ninety-eight pages and seven chapters; this extension has been quite substantial. It reflects the growth of interest in the area since 1987, as well as the considerable research effort over that time. Particular new developments included in this edition are work on discrete response data, time series models, random cross-classifications, errors of measurement, missing data, nonlinear models, and a discussion of available software. As in the previous edition, the theoretical presentation is en-livened with many examples. The dropping of the qualifying phrase from the title of the first edition is fitting since this class of model is of wider interest than merely to educational and social researchers. In summary, this book is an elegant outline of a class of statistical models and ideas which are becoming of increasing relevance and importance. It is nicely produced and clearly written.

Contents:
1. Clustering from the perspective of combinatorial data analysis, by P. Arabie and L.J. Hubert
2. Developments in principal component analysis, by B.D. Flury
3. Canonical discriminant analysis: Comparison of resampling methods and convex hull approximations, by C. Weihs
4. Nonlinear methods for the analysis of homogenicity, and heterogeniety, by W.J. Heiser and J.J. Meulman
5. Principal component models for patterned covariance matrices, with applications to canonical correlation analysis of several sets of variables, B.D. Flury and B.E. Neuenschwander
6. Othogonal and projection Procrustes analysis, by J.C. Gower
7. Graphical modelling, by D. Edwards
8. Convergent computation by iterative majorization: Theory and applications in multidimensional analysis, by W.J. Heiser
9. Biplot display of multivariate categorical data, with comments on multiple correspondence analysis, by K.R. Gabriel
10. MANOVA biplots for two-way contingency tables, by K.R. Gabriel
11. Some tools for the multivariate analysis of functional data, by J.O. Ramsey
12. A general theory of biplots, by J.C. Gower

This book arose from a seminar series, sponsored by the University of Exeter and Shell Research Limited, in which leading researchers in various areas of descriptive multivariate analysis each presented several talks reviewing recent work in their area. The aim of the book is to bring statisticians who want to learn about the current frontiers of research in the area of descriptive multivariate analysis up to the frontier. In this, the book succeeds.
Inevitably, in reviewing a book with the aims of this one, one will seek topics which have been omitted. Two such topics are modern exploratory methods and visualization methods. However, it turns out that both of these were covered by the seminar series, by C. Weihs and F. Young, respectively, but have been published elsewhere. In terms of the completeness of coverage of the present volume, this is a pity.
Also inevitably, there is some unevenness in depth of coverage and range of the topic treated by each of the contributors. For example, Weihs describes 'the examination of criteria for judging the predictive power of canonical discriminant analysis'. Contrast this with Edwards' overview of the whole of the now large field of graphical models and Gower's 'general theory of biplots'.
The book is the second in the Royal Statistical Society Lecture Notes Series. It is a worthy addition to that series. It would make a good supplementary text for a course on modern multivariate descriptive methods.

Contents:
1. Introduction
2. Basic concepts of power system reliability evaluation
3. Elements of Monte Carlo methods
4. Generating system adequacy assessment
5. Composite system adequacy assessment
6. Distribution system and station adequacy assessment
7. Reliability cost/worth assessment

With the development of faster and faster computers, the sky has really become the limit, at least, this is what the authors seem to be suggesting. Power system behaviour has traditionally been analyzed and designed on the basis of a deterministic framework. It is only recently that the inherent stochastic nature of power provision has been accounted for in design. Clearly, Monte Carlo methods are nicely suited to evaluate reliability properties and other related network characteristics. Therefore, the authors devote a substantial part of their book to Monte Carlo techniques, starting from first principles. The remainder of the book focuses on various types of analyses for different types of systems. It is supplemented by numerical examples and plenty of attractive graphs. This book is not designed for statisticians: readers with interest in probabilistic modeling would be disappointed not to find the expected in-depth material and/or references. However, the book will be valuable as an introduction to stochastic methods for practising electrical engineers.

Contents:
1. Introduction
2. Failure models
3. Qualitative system analysis
4. Systems of independent components
5. Component importance
6. Markov models
7. Counting processes
8. Dependent failures
9. Life data analysis
10. Accelerated life testing
11. Bayesian reliability analysis
12. Reliability data sources

This book provides a quite comprehensive treatment of system reliability theory and methods. It is especially good in its discussion of stochastic models and of structural analysis of systems and contains many practical illustrations and useful information on software. This material covers eight chapters and about three hundred and fifty pages. There are also three chapters (9-11) on statistical analysis, com-prising about one hundred and ten pages, and a short final chapter on reliability data sources. The book is a valuable addition to the reliability literature.

Contents:
Introduction
1. Asymptotic theory for the generalised bootstrap of statistical differentiable functionals
2. How to choose the weights
3. Some special forms of the weighted bootstrap
4. Proofs of results of Chapter 1
5. Proofs of results of Chapter 2
6. Proofs of results of Chapter 3
Eight appendices

With the standard Efron bootstrap, members of a bootstrap sample are drawn by randomly sampling with replacement from the original sample of data and are present in proportions which are determined by a uniform multinomial distribution on the original sample values. A generalized or weighted bootstrap is obtained by requiring only that the number of times that values of data are resampled form an exchangeable sequence. This monograph is concerned with conditions for asymptotic consistency of the weighted bootstrap and with questions related to appropriate choice of the ex-changeable sequence in particular contexts. The presentation is clear, but as might be expected of a research monograph, the treatment is highly technical and dense, with over half the monograph devoted to detailed proofs of the main results. It is a book for the specialist only. Though the book will be found useful by some, the overall level of finish is rather weak, with numerous misspellings, errors in the references, etc.

Contents:
1. Sufficiency and completeness
2. The evaluation of estimators
3. Mean unbiased estimators and convex loss functions
4. Testing hypotheses
5. Confidence procedures
6. Consistent estimators
7. Asymptotic distribution of estimator sequences
8. Asymptotic bounds for the concentration of estimators and confidence bounds
9. Miscellaneous results and asymptotic distributions
10. Asymptotic test theory

Among the numerous books that appear on parametric statistical inference, the present one is of an outstandingly high mathematical level. The author presents the theory of estimation and testing in parametric models assuming good knowledge of measure theory and probability theory. Some, but not all, of these prerequisites are summarized in appendices to some of the chapters. They include for instance: conditional expectation, uniform integrability, uniform stochastic convergence, measurable selection, weak convergence. Chapter 1 of the book gives the basic concepts of the whole theory: sufficiency, completeness, exponential families, equivariance, invariance. From there, the theory of parametric inference for independent and identically distributed observations is developed and the optimality of the estimators and tests is discussed. Both exact and first-order asymptotic results are given. A nice thing about the book is that throughout there are many examples and some exercises which bring the reader from some abstract space back into the familiar parameter spaces of gamma and normal densities. The bibliography contains more than two hundred and fifty items. Some of these go back to the early history and are commented on by the author.

Contents:
1. Basic concepts in distribution free methods
2. One-sample location problems
3. Miscellaneous one sample problems
4. Two sample problems
5. Straight line regression
6. Multiple regression and general linear models
7. Bivariate problems
8. Miscellaneous complements (linearization representation; asymptotic relative efficiency; estimating equations and the smoothing of statistics; least squares smoothing; kernel gradient estimates; bootstrap estimation of standard errors; conditional standard errors)

Readership: Those who want an introductory course in these methods. Nominally, undergraduate students

The author points out early (p.2) that "Few of the so-called distribution-free methods are truly distribution-free." They do require at least mild assumptions. Randomization is a unifying notion here and there is emphasis on problems of location and location shift.
Remarkably, the second edition is about ten pages shorter than the first! This is due to a smaller print size, and the text is slightly expanded with more examples and data. Also, the author has added a new Chapter 8. The table of contents is now very detailed, but the main topics of Chapters 1 to 7 remain the same.
A few exercises have been added and a few removed. Altogether, this is a nice revision. If you liked the first edition, this one will please you too!

Contents:
1. Introduction
2. The growth curve model
3. A multidimensional growth curve model
4. The sum of profiles and time moving covariates model
5. A growth curve model with exchangeably distributed errors
6. Structured covariance matrices, model selection, prediction from growth curves and other topics
7. Growth curves with incomplete or unbalanced data
8. Potthoff-Roy growth curve model: Derivations of main results
9. Bayesian analysis of the Potthoff-Roy model
10. Nonparametric methods in growth curve analysis

Growth curves are used to model data in which there are observations on the same experimental units taken over time, so that the observations are correlated. A model's coefficients and the variance-covariance structure are estimated from the data. This book provides an attractive, sophisticated treatment of the area. Knowledge of a matrix algebra, including the Kronecker product, is needed for an easy comprehension. The presentation is clear and terse. The three case studies come with sets of data and SAS programs, including three-dimensional graphic programs. This is an attractive course text.

Contents:
1. Measurable spaces and integrable functions
2. Borel measures and Radon measures
3. Fourier analysis
4. Hilbert space methods and limit theorems in probability theory
5. Gaussian Sobolev spaces and stochastic calculus of variations
6. Appendix I : Hilbert Spectral Analysis
7. Appendix II: Infinitesimal and Integrated Forms of the Change-of-Variables Formula
8. Exercises for Chapter 1
9. Exercises for Chapter 2
10. Exercises for Chapter 3
11. Exercises for Chapter 4
12. Exercises for Chapter 5

If I were asked to recommend texts to research students who need a grounding in integration theory this book would be on the list. Professor Malliavin gives a systematic development of the integration theory which ties together the abstract theory, 'the Halmos way', with the Riesz representation approach. He then moves on to applications and it is here that the richness of this subject area is displayed. The book is excellent!

Contents:
1. Basic concepts
2. Numerical methods for queueing networks with blocking
3. Two-node open queueing networks with blocking
4. Approximate analysis of open tandem queueing networks with blocking
5. Approximate analysis of arbitrarily linked open queueing networks with blocking
6. Closed queueing networks with blocking with product-form solution
7. Closed queueing networks with blocking with non product-form solution
8. Applications

Readership: Graduate students, researchers, practitioners involved in performance evaluation of computers, communication networks and production systems

This book collects together and surveys exact and approximate methods for queueing networks with blocking, at a level suitable for graduate students, one queueing theory course being the stated prerequisite. The first two chapters include practical introductions to techniques that are applicable in many areas of stochastic modelling, such as the approximation of distributions by phase-type distributions, and numerical methods for continuous-time Markov chains. The remaining chapters analyze particular queueing networks with blocking, giving exact solutions where possible, and illustrating how to apply a wide variety of approximation techniques in other cases. Special attention has been given to providing a de-tailed historical overview of the relevant literature, together with a bibliography, at the end of each chapter. This is in addition to the extensive bibliography at the end of the book. There are no exercises.

Contents:
1. Some basic concepts and theorems of probability theory
2. Probability inequalities for sums of independent random variables
3. Weak limit theorems: Convergence to infinitely divisible distributions
4. Weak limit theorems: The central limit theorem and the weak law of large numbers
5. Rates of convergence in the central limit theorem
6. Strong limit theorems: The strong law of large numbers
7. Strong limit theorems: The law of the iterated logarithm

Limit theory for sums of independent random variables plays a central role in probability and statistics, and this text provides a carefully prepared and up-to-date coverage of the basic material. It is a book in the spirit of the classic one written by Gnedenko and Kolmogorov some forty-five years ago. It is a scholarly work based on four hundred and eighty-seven references. The author index is fine but the two-page subject index should have been expanded; for example, a finer indexing of 'central limit theorem' and 'infinitely divisible distributions' would be natural in a book of this title. A casual reading revealed no typographical errors. The outline of the material is good. Its presentation is succinct. Each of the seven chapters ends with a subsection entitled, Addenda, in which two hundred and forty-four additional and relevant results are stated with references. A few of these could possibly be used as problems for advanced classes provided hints were included. Their main purpose is to make the book's coverage of recent developments more complete. In summary, this text should be a useful reference for many years to come.

Contents:
PART I : Fundamentals of the Cycle Representations of Markov Processes
1. Directed circuits
2. Genesis of Markov chains by circuits: The circuit chains
3. Cycle representations of recurrent denumerable Markov chains
4. Circuit representations of finite recurrent Markov chains
5. Continuous parameter circuit processes with finite state space
6. Spectral theory of circuit processes
7. Higher-order circuit processes
PART II : Applications of the Cycle Representations
1. Stochastic properties in terms of circuits
2. Lévy's theorem concerning positiveness of transition probabilities
3. The rotational theory of Markov processes

Readership: Research mathematicians in probability theory, algebraic topology (network theory), algebra, convex analysis, theory of algorithms and stochastic processes

The purpose of this book is to give a systematic and unified exposition of Markovian stochastic processes, which, under an additional assumption concerning the existence of invariant measures, can be defined by directed cycles or circuits. These processes are called cycle (or circuit) processes, and the corresponding collections of weighted cycles are called cycle representations. A circuit or a cycle is a geo-metric concept that can be defined either by geometric or by algebraic considerations. This book is an excel-lent state-of-the-art survey of the principal trends to cycle processes theory. The first part deals with the basic concepts and equations of cycle representations. The second part is concerned with the application of the theory to the study of the stochastic properties of Markov processes.

Contents:
Introduction
PART I : Random Walk of a Random Field
1. Brownian motion of a Poisson field
2. Extreme value problems
3. Changing the initial process and the motion
PART II : Branching Random Walk
4. Branching random walk starting with one particle
5. Branching random walks of a random field
6. Branching Wiener process starting with one particle
7. Critical branching random walk starting with one particle
8. Critical branching random walks of a random field
9. Multitype branching random walk
PART III: Strassen Type Theorems
10. Infinitely many dependent particles
11. Branching random walk
Historical Overview

This book is a sequel to the author's Random Walk in Random and Non-Random Environments (1990) [Short Book Reviews, Vol.11, p.48], which considered the motion of a single particle. Here the focus is on the motion of countably infinitely many particles in d-dimensional Euclidean space. Asymptotic properties as time tends to infinity are studied. In Part I the particles have independent and identically distributed motions. Part II is concerned with branching random walks where the particles initially move independently but die after a random time and are replaced by off-spring according to a Galton-Watson process. These continue to move independently according to the same law as their ancestors. Part III is concerned with functional laws of the iterated logarithm for the path properties of many independent Wiener processes.

Contents:
1. Introduction
2. Random walks and random flights
3. Random walk on a lattice
4. Random walks in the continuum limit
5. Continuous-time random walks
6. Exploration and trapping
7. The self-avoiding walk

In the last ten years a number of books appeared on random walks showing the increasing interest in the subject. In fact statistical physics produces new questions day-by-day. The present book differs from the others in the following facts: (i) each new problem is started by an interesting, detailed historical overview, (ii) concentration on problems suggested by physics, (iii) more emphasis on fractals (showing the fractal properties of random walks and studying the random walks on fractals), (iv) studies of many special classes of random walks, (v) less emphasis on strong laws and limit theorems; in many problems the author is satisfied by giving the expectation and variance.

Readers with some background in mathematics will be able to read this pleasant monograph on stochastic ordering and dependence. Those familiar with stochastic processes will enjoy riding on an 'ordering' horse while watching their pet topics along the side-walks: coupling, conditioning, Markov processes, point processes, queues, networks, etc.
The bibliography covers some one hundred and eighty papers. Examples, problems and useful remarks are generously spread over the text.

Contents:
1. Regenerative processes
2. Crossing and coupling
3. Ergodicity and comparison
4. First occurrence times
5. Applications
Commentaries

The notion of regeneration appears in many fields of applied probability such as Markov chains, queues, storage models, reliability models, ... . For example, the successive passage times to a definite state in a finite Markov chain form a sequence of re-generation times. At each of these random times, the process starts again and the trajectory is divided into independent identical distributed cycles. The precise definition of a regenerative process, in the strict sense and the wide sense, is given in Chapter 1 after a brief overview of the needed concepts from probability and stochastic process theory. Many examples are provided. The important techniques of coupling and crossing are described in Chapter 2 and used in Chapter 3 for the analysis of ergodicity and continuity. Probability metrics play an important role here. Chapter 4 is devoted to the estimation of the distribution function of first occurrence times. Some practical algorithms are proposed. Applications are given in the final chapter. This is a well-written book on a very specialized topic. Many of the theorems are given with the detailed proofs which makes it useful for students and researchers in this area.

Contents:
0. Introduction
1. Probability models
2. Geometric probability, coverage and visibility in random fields
3. Visibility probabilities
4. Visibility probabilities II
5. Distributions of visibility measures
6. Distributions of visible and invisible segments
7. Problems and solutions
Readership: Military scientist, communications scientist, statistician

This monograph is concerned with the use of coverage processes in geometric probability for model-ling one or more observers viewing partially obscured targets in the plane or on the sphere. The obscuring elements are generally taken as random discs, or spheres with random radii, whose centres are randomly located in a specified region. The motivation for this work has come primarily from military applications of hunting obscured targets, and its focus reflects this, but the results are relevant for a wide diversity of other contexts. Problems, with solutions, are provided for each chapter.

Contents:
Introduction
1. Sums of a random number of random variables
2. Branching processes with generalized immigration
3. Branching processes for time-dependent immigration
4. The asymptotic behavior of families of particles in branching processes

Consider particles that live independently of one another for some random time before generating a random number of new particles; these new particles undergo analogous transformations. Moreover there is a possibility that new particles not only emerge through reproduction but also through immigration depending on the reproduction. The monograph studies the number of such particles in the process by developing methods for the behavior of random sums of random variables.
The somewhat technical monograph provides an important addition to and a welcome survey of the broad literature on branching processes.

Contents:
1. Probability tools and preliminary results
2. Simple integral equations: Versions of the integrated Cauchy functional equation
3. A version of Deny's theorem and its extensions: A Martingale approach
4. Multiple integral equations and stability theory
5. Mean residual life function and hazard measure
6. Properties based on Fourier and Mellin transformations
7. Damage models and partial independence
8. Order statistics, record values and properties in applied probability
9. Characterizations based on regression and related statistical properties

Characterization problems is a fascinating area of mathematical statistics. Books to survey recent results have been missing, but here is a good one, essentially concerned with findings from the last three to four decades. The earliest of the many papers referred to by the authors are from 1947 and 1970 respectively; this book is a result of a long love. What Rao and Shanbhag have compiled is a cornucopia of analytical probability theory.

Contents:
PART I : Bayesian Image Analysis: Introduction
PART II : The Gibbs Sampler and Simulated Annealing
PART III: More on Sampling and Annealing
PART IV : Texture Analysis
PART V : Parameter Estimation
PART VI : Supplement

Readership: Research workers interested in rigorous mathematical treatments of the subjects covered by the monograph's title

This book presents a rigorous, self-contained treatment of the mathematical theory of random fields and Monte Carlo methods in the context of image analysis. This includes, for example, recent results on approximating the second largest eigenvalue of a Markov chain and the work of Comets and Gidas on asymptotics of maximum likelihood and maximum pseudolikelihood estimation in Markov random fields. The book also touches on much applied material, such as texture analysis and tomography, but the emphasis is still on precise mathematical formulations rather than detailed consideration of applications. The book could be very successful as a course text or for background reading among those who want to study the mathematical back-ground of this whole rapidly expanding field. It will not be so popular among those whose primary interest is computation and applications. There are various
quirky features, for example, the consistent mis-spelling of spatial, but these will not deter the committed reader.

Contents:
1. Introduction
2. Hypergroups and their measure algebras
3. The dual of a commutative hypergroup
4. Some special classes of hypergroups
5. Positive and negative definite functions and measures
6. Convolution semigroups and divisibility of measures
7. Transience of convolution semigroups
8. Randomised sums of hypergroup-valued random variables
9. Further topics

Although this book deals mainly with com-mutative topological hypergroups it offers a comprehensive introduction to hypergroups. It aims to apply the "hypergroup method" to problems in probability theory. The reader unfamiliar with this area would do well to read this book with a text about harmonic analysis on groups to hand, because, from the viewpoint of this book, harmonic analysis on hypergroups amounts to harmonic analysis on certain measure algebras.
A good test of the quality of the writing of a mathematics text is to pick a few of the proofs, at random, and read them through looking for style and clarity. This book scores well.

Contents:
1. Preparatory material
2. Killed Brownian motion
3. Schrödinger operator
4. Stopped Feynman-Kac functional
5. Conditional Brownian motion and conditional Gauge
6. Green functions
7. Conditional Gauge and q-Green function
8. Various related developments
9. The case of one dimension

This book is a self-contained monograph on potential theory of Schrödinger operators, to which the authors contributions in the last decades are well-known. Each chapter contains "Notes" by one of the authors, K.L. Chung, who provides an historical over-view of the subject treated. This book is an excellent contribution to potential theory and stochastic pro-cesses, and recommended to researchers and graduate students of mathematics and mathematical physics.

Contents:
1. Introduction
2. Multivariate distributions
3. Initial data analysis
4. Projections and linear transformations
5. Distance methods and ordination
6. Inference: Estimation and hypothesis testing
7. Multivariate linear models
8. Nonlinear methods

This is the first of two volumes of multivariate analysis, forming part of the updating of Kendall's Advanced Theory of Statistics, and presented in the familiar format of that work. This book reveals the breadth of research that is taking place in the area; topics covered include mention of simulated annealing, neural networks, genetic algorithms and Markov-chain Monte Carlo. Inference is covered in several places, but only briefly. The facilities of several main computer packages are described, but not in an exhaustive fashion. Throughout there is valuable critical discussion.
This book is a scholarly work which is likely to be the essential reference for statisticians interested in multivariate analysis research until the next update of Advanced Theory. Part 2 of Volume 2A will contain material on discriminant analysis, cluster analysis, latent variable techniques, graphical model-ling and path analysis, and repeated measures analysis.

Contents:
1. The Bayesian method
2. Inference and decisions
3. General principles and theory
4. Subjective probability
5. Non-subjective theories
6. Subjective prior distributions
7. Robustness and model comparison
8. Computation
9. The linear model
10. Other standard models
Readership: Statisticians and students of statistics who have been exposed to a fair amount of calculus-based statistical theory

This very well-written book has been designed to complement the 'Kendall' series by presenting there-in the Bayesian point of view. Its contents and its presentation make it a suitable addition to the Kendall family. The format is the same as the other Kendall volumes, to which it makes reference when appropriate.
Chapter 1 is an excellent and brief introduction to the Bayesian method. Chapter 2, containing details about how to make inferences from the posterior distribution, is perhaps longer than necessary. Chapters 3 to 6 cover the rest of the generalities well; Chapter 5 contains information on the relationship between admissibility and Bayes/generalized Bayes rules and on James-Stein estimation. Chapter 7 on robustness is extensive, reflecting the attention given in the literature to this area; topics are sensitivity, epsilon contamination, local and limiting sensitivity and a good deal of material on Bayes factors for com-paring models. Regarding computation, one very notable feature is that Gibbs sampling, the importance of which has skyrocketed since 1990, is not only the main topic of the second half of Chapter 8, but is also referred to in various places throughout the text. Chapters 9 and 10 deal with specific models.
Although this reader would have liked to see more material on the very important topic of hierarchical models, a total of five section headings being devoted to it here, the author has skilfully managed to cover a great deal of ground in this volume, and readers will find few topics of interest to be missing.

Contents:
1. Introduction
2. Metric multidimensional scaling
3. Nonmetric multidimensional scaling
4. Further aspects of multidimensional scaling
5. Procrustes analysis
6. Monkeys, aeroplanes, yoghurts and bees
7. Unfolding
8. Correspondence analysis
9. Individual differences models
10. ALSCAL and SMACOF
11. Further m-mode, n-way models

Multidimensional scaling is the name for a class of statistical techniques which seek a low dimensional representation for a multivariate sample. The representation sought is the one which best re-presents the similarities between the sample objects by the distances between the representing points. Freedom to choose what is meant by 'best', 'similarity', and 'distance', as well as freedom to choose the algorithm which finds the best, leads to many different methods. This book reviews such methods, beginning with early 'classical' methods (originating in the 1930's) and culminating with some of the latest research, for example, the tunnelling methods of Groenen. The chapters present the theory and then illustrate the methods by way of real data examples. An envelope at the back of the book contains a disc which includes the sets of data and programs for applying the methods described in the text.
The book is clearly written and would be well suited for use as the text for a course on multidimensional scaling.

Contents:
1. Introduction
2. Modelling and analysis of cross-sectional data: A review of univariate generalized linear models
3. Models for multicategorical responses: Multivariate extensions of generalized linear models
4. Selecting and checking models
5. Semi-and nonparametric approaches to regression analysis
6. Fixed parameter models for time series and longitudinal data
7. Random effects models
8. State space models
9. Survival models

Readership: Applied statisticians, graduate students of statistics, and students and researchers with a strong interest in statistics and data analysis from areas like econometrics, biometrics and the social sciences

This book is not an introductory text on generalized linear models, nor a research monograph on the latter. The basic aim of the authors is to bring together and review a large part of recent advances in statistical modelling of multivariate and multicategorical models within the generalized linear models framework. Based on well-chosen sets of data, these new developments are introduced to a not necessarily expert audience. Completeness was not an aim. The result is a self-contained, well-written text offering the applied researcher a useful insight into the applicability of the general linear model methodology.

Contents:
0. Introduction
1. Undominated experiments
2. PSS, pivotal measure, and Neyman factorization
3. Structure of pairwise sufficient subfield and PSS
4. The Rao-Blackwell theorem and UMVE
5. Common conditional probability for PSS and its applications
6. Structure of pivotal measure

Readership: Persons interested in giving maximally general mathematical expression to the Fisher-Neyman concept of sufficiency

"In these notes we present a theory of sufficiency which covers undominated experiments as well as dominated ones. The familiar topics in the dominated case, such as pairwise sufficiency, Neyman factorization, minimal sufficient statistics, the Rao-Black-well theorem, are treated from a more general viewpoint than in the Halmos-Savage-Bahadur scheme ... while the more well-known "pathologies" of Pitcher (1957) and Burkholder (1961) are averted."
"The concepts of pivotal measure and PSS (pairwise sufficiency with supports) play a fundamental role in this theory. The former is an extension of the "pivotal measure" in the dominated case, a special type of dominating measure defined by Halmos and Savage (1949) and called as such by Badahur (1957)."
It is perhaps unfortunate that the term "pivotal" was used by Fisher in connection with his fiducial argument in a sense completely different from its use by Bahadur, although both theories involve attempts to express the notion of something which "carries all the information".

Contents:
1. Introduction
2. Theory of sequential and multistage procedures
3. Selecting the best normal population
4. Selecting the best negative exponential population
5. Estimation of ordered parameters
6. Selecting the best component in a multivariate normal population
7. Estimation after selection and ranking
8. Additional topics

After a long period without any new books on ranking and selection, it is a pleasure to welcome the appearance of this scholarly book, but not without some critical comments.
This is a scholarly work that requires a deep extended study in view of the constant reference to published papers; hence it is for specialists in the field of ranking and selection. It is not appropriate as a textbook on the subject or as an introduction to the field, although there is an attempt by the authors to make it appear that way. The tabular comparisons of different procedures (a Herculean job) are all presumably Monte Carlo results (but never titled as such) and they are not easy to read or interpret. A twenty-five page bibliography at the end of the book is heavily weighted with the papers of one of the authors and is admittedly incomplete for the field as a whole.
The book has a nice feature that multistage (and sequential) procedures compete with single-stage procedures whenever both are possible. The authors could have made their point clearer and stronger by bringing in at least one example of a multistage procedure in the introduction and making early comparisons and conclusions right there. As it stands, the earliest comparison comes after 100 pages of text (cf. pages 128, 139). Here we see two tables with K=5 and 10 populations and it takes quite a bit of digging (cf. Pages 113-117, 138), to understand what is in the two tables. An early synopsis would have been useful; an illustration of the second-order asymptotics could also have been included in the introduction. The fact that the sequential procedure is better (although not always practical to implement) is an important conclusion, also not easy to dig out from the book.
If a table is based solely on Monte Carlo sampling, one expects to see (but rarely sees in the book), titles like "Monte Carlo Comparisons..."
The authors should be congratulated on their thorough treatment of a large number of topics, all related to ranking and selection. These include elimination procedures, estimation of ordered parameters, estimation after selection, three-stage procedures, accelerated sequential procedures, with special emphasis on normal and negative exponential populations. These help to give some perspective on the vastness of the field and its unlimited potential usefulness. What comes to mind, however, is the oft-repeated question of Professor Herbert Robbins: "Why hasn't ranking and selection caught on like the analysis of variance?" and the associated question, "Will this new book have a serious influence on that?" Hardly!

Contents:
1. Preliminaries
2. The linear-by-linear interaction model
3. Association models for two-way tables
4. Other models for two-way tables: Symmetry type models
5. Multiple dimensions of association
6. Bivariate association in multiple groups
7. Logit type regression models for ordinal dependent variables

Readership: Social scientists and other researchers who wish to analyze ordinal variables

The book aims to survey a large class of models for the analysis of discrete ordinal variables, focusing mainly on association models. It is not a theoretical discourse, but is aimed at those who wish to apply the ideas. It assumes familiarity with log-linear models and logit regression. There are no exercises, but there are periodic parenthetic injunctions to the reader, for example to 'prove', 'show', or 'verify' results to test the reader's understanding.
Chapter 1 includes a concise but clear review of criteria for assessing goodness-of-fit, defines the notation, and discusses some basic issues such as measures of association and the independence model. Chapter 2 describes how fixed scores can be used for the levels of categorical variables and the consequences this has for the analysis. Later chapters permit the scores to be parameters which have to be estimated. More sophisticated and flexible models are gradually introduced as the book proceeds. Most of the book is concerned with association models and Chapter 7 is rather different from the other chapters in that it discusses the situation in which one of the variables is an ordinal 'response' variable. This material has been covered extensively elsewhere, and in this book the emphasis is in large measure to link it to the earlier discussion of association models.
Overall I found this volume an accessible unification of work in the area. I recommend it.

Contents:
1. Preliminaries
2. Matrices, vector spaces
3. The ordinary principal components model
4. Statistical testing of the ordinary principal components model
5. Extensions of the ordinary principal components model
6. Factor analysis
7. Factor analysis of correlated observations
8. Ordinal and nominal random data
9. Other models for discrete data
10. Factor analysis and least squares regression

Readership: Statisticians, researchers in the empirical sciences, postgraduate students

This is another book [Reyment and Jöreskog, Short Book Reviews, Vol. 13, p.40] which uses the phrase "factor analysis" in an informal sense: it encompasses principal component analysis, canonical correlation analysis, correspondence analysis, multidimensional scaling and many other eigenvalue/eigen-vector-based techniques in addition to the more usual interpretations of factor analysis and latent structure analysis. While one can appreciate the reasons behind such relaxation of language, nevertheless the frequent interchangeability of "factor" with "component" or "dimension" can be confusing to the uninitiated.
Nonetheless, the book displays a number of strengths: it is very comprehensive, covering many less-familiar techniques from the psychometric and econometric literature as well as the traditional ones cited above; it has an extensive bibliography, over eight-hundred references, and there are many sets of data, data-based illustrations and examples alongside the formal algebraic development of the theory. Unfortunately, even a cursory reading reveals an alarmingly high level of (obvious) typographical errors and a number of careless statements, for example, the definition of distance between two points along a circle, p. 279. Any recommendation must therefore be a qualified one.

Contents:
1. Asymptotics in general
2. Characteristic functions and the Berry-Esseen theorem
3. Edgeworth series
4. Saddlepoint series for densities
5. Saddlepoint series for distribution functions
6. Multivariate expansions
7. Conditional distribution approximations
8. Applications to likelihood ratio and maximum likelihood statistics
9. Other topics
10. Computational aids

There exists already quite a number of excel-lent books on the topic of asymptotic expansion methods in statistics. The present monograph inevitably has much overlap with the publications and it is hard to find out what is really new. The careful way in which regularity conditions are stated and discussed is certainly a plus for this book. Also the inclusion of the lattice case is interesting.
Chapters 2 and 3 deal with characteristic function techniques and their use in deriving Berry-Esseen bounds and Edgeworth expansions (via the smoothing theorem). Chapters 4 and 5 give the saddle-point methodology for approximating densities and tails of distribution functions. Reading this requires knowledge of complex analysis. Multivariate and conditional versions are also dealt with and applications are given. There are some illustrations indicating the quality of the approximations in concrete examples.

Contents:
1. Facet theory: a strategy for scientific theory building
PART I : Facet Design: The Shaping of Research Contents
2. Mappings and assignments
3. The range facet: An image of reality
4. Formalization of research content design: Domain facets
5. Definitions and hypotheses
6. The mapping sentence
PART II: Intrinsic Data Analysis: The Straight Way To Handle Multivariate Data
7. Behavioral theories: Picturing concepts by faceted SSA
8. Behavioral measurement: Multiple scaling by POSAC/LSA

Facet theory represents statements and quest-ions as mappings of Cartesian products of sets, facets, to range sets. Mapping sentences specify the domain and range of such mappings, and the connections between facets. A variable is a focused question and observed behavioural and psychological variables are regarded as points on a continuum of variables, rather than as discrete entities. The observed variables are but a sample from this continuum and the aim is to infer the structure of the continuum. This is done by a particular kind of multidimensional scaling called smallest space analysis, based on observed similarities between the variables.
Part I of the book introduces and defines the notions of the theory, and Part II describes how to apply the methods in practice. There are many simple examples.
As an introduction to facet theory, this book is good. What one thinks of the methods themselves, is up to the reader to decide.

Contents:
Introduction
PART I : Foundations of Theory and Methods of Finite-Dimensional Optimization
1. Basic concepts, problems and fields of applications of perturbation theory
2. Perturbation theory of smooth mathematical programming problems (main results)
3. Examples of studies in parametric optimization problems. Applications of perturbation theory
PART II : Basic Perturbation Theory in Finite- Dimensional Optimization
4. Elements of convex and nonconvex analysis
5. Optimality criteria in generating problems
PART III: Perturbation Theory in Mathematical Programming
6. General perturbation theory
7. Perturbation theory for smooth mathematical programming problems
8. Theory of minimax perturbations under bound constraints

In this book the author presents his work on a general perturbation theory of finite-dimensional extremum problems. The principle aim is to state strictly almost all of the results of perturbation theory and to illustrate these results in analyzing a set of parametric extremum problem classes. Many of the three-hundred references are to articles published in Russian journals. This text is for the mathematically strong who are doing research in perturbation theory.

Contents:
1. Von Mises functionals
2. Log likelihoods
3. Asymptotic statistics
4. Nonparametric statistics
5. Optimal influence curves
6. Stable constructions
7. Robust regression

This book contains two parts. The first, which is of interest also to people not working in robust-ness, provides a self-contained treatment of several important results in asymptotic statistics. This material is presented in Chapters 1 to 4, in combination with Appendix A which gives the necessary back-ground on weak convergence. The author's approach is based on Von Mises functionals, for which the main ideas of differentiation and asymptotic normality are outlined in Chapter 1. Chapter 4 stresses the estimation of functionals instead of parameters.
The second part of the book is of specific interest to robustness researchers. Chapters 5 and 6 discuss the selection of robust functionals for estimation and testing, based on the principles of minimax asymptotic variance, minimax oscillation (bias) and minimum distance. Chapter 7 considers the regression model, mainly focusing on bounded-influence equivariant estimators.
The book forms a bridge between the general field of asymptotic statistics and robustness research, and provides a rigorous framework for further developments in both areas.

Contents:
PART I : Fundamentals of Probability Theory
PART II : Discrete Models
PART III: Nondiscrete Models
PART IV : Multivariate Normal Models
PART V : Limit Concepts

Readership: Undergraduate and graduate students needing a broad coverage of probabilistic models and techniques

Based on the knowledge of real analysis and linear algebra, this book aims at providing a unifying and encyclopedic treatment of probability theory. Measure theory is introduced 'along the way', depending on the needs of the developing probabilistic theory. The book goes beyond most introductory texts in scope and breadth. Examples of more advanced material are chapters on large deviation theory and central limit theorems for dependent random variables. The basic probabilistic calculus, distribution theory, convergence concepts, generating functions, etc. are treated very much in detail throughout the text. Chapters on random walks, Markov chains and discrete renewal pro-cesses offer a first excursion into the world of stochastic processes. Each chapter contains numerous examples and exercises. Both undergraduate as well as graduate students may find the fact of having most of the theory under 'one roof' useful. The price one pays for the latter is a text which occasionally looks too much like an encyclopedia. I missed somehow the spark of enthusiasm which a really good textbook on the subject should have.

Contents:
Introduction
1. The Palm-martingale calculus of point processes
2. Stationarity and coupling
3. Formulas
4. Stochastic ordering and comparison of queues

One of the purposes of the present book is to present a unified view of what can be said when only very weak assumptions are made on the input to a queueing system. While the book does not give a systematic treatment of queueing theory, it presents beautifully the elements of the more general methods of point processes and stochastic recurrences which can be directly useful to queueing theory. The final outcome is impressive in that the book not only weakens the usual renewal assumptions but also provides unified and rigorous proofs of an abundance of classical
results.

Contents:
1. Renewal processes with applications
2. Markov chains: theory and applications
3. Markovian decision processes and their applications
4. Algorithmic analysis of queueing models
APPENDIX A: Useful Tools in Applied Probability
APPENDIX B: Useful Probability Distribution Functions
APPENDIX C: Laplace Transforms and Generating Functions
APPENDIX D: Numerical Solutions of Markov Chain Equations

There are three main strands to the approach of this book. First it covers theoretical properties of various stochastic models; secondly, it illustrates their applications through a wide range of examples; and thirdly, it includes detailed discussion of numerical and computational aspects of stochastic modelling. Overall the book has similar aims and objectives to those of the author's earlier book Stochastic Model-ling and Analysis: A Computational Approach, [Short Book Reviews, Vol. 6, p.28] the main difference being that the new book is aimed at students rather than researchers.
From the preface: "the prerequisites of the book are only calculus and a first course in probability." Students with no previous exposure to a course involving stochastic models could find the development tough in places, although the appendices should provide some help. There are many exercises. Each chapter ends with a short discussion of the literature and a list of references. A solution manual and a supporting software package are available.

,39.95/US$63.95.
Contents:
PART I : Fractals and Methods for the Determination of Fractal Dimensions
1. Introduction
2. Hausdorff measure and dimension
3. Deterministic fractals
4. Random fractals
5. Methods for the empirical determination of fractal dimensions
PART II : The Statistics of Shapes and Forms
6. Fundamental concepts
7. Representation of contours
8. Set theoretic analysis
9. Point description of figures
10. Examples
PART III: Point Field Statistics
11. Fundamentals
12. Finite point fields
13. Poisson point fields
14. Fundamentals of the theory of point fields
15. Statistics for homogeneous point fields
16. Point field models

Readership: Scientists working with planar structures, statisticians advising these scientists

This is a translation from German of the "Fraktale-Formen-Punktfelder. Methoden der Geometrie-Statistik" Akademie Verlag, Berlin, 1992. It introduces three topics on geometric structures in the plane. Scientists investigating these structures, and statisticians called on to analyze them, will find much use-ful background. Dietrich Stoyan is one of the creators of stochastic geometry, and the section on point fields is the best here. The sections on fractals and shapes seem more to reflect the authors' reading than their involvement. I would like to have heard their views on the merits of the various procedures, and would have welcomed much more about their statistical properties.
The book is well illustrated, but presents some difficult material very early on (formulae for Hausdorff measure on page 11) which could be off-putting. In contrast, we are given a subroutine for uniform variates ("X=á*RND(0)") on page 198. The translation can be awkward ("quadratic window"), and the writing can be confusing, as with the Delaunay triangulation, and occasionally carelessness intrudes ("Note ö " A=ö!"). The appendices are too brief, with just two pages on Measure and Content, and one on Basic Ideas in Topology, for instance.
Despite these critical remarks, I found the book stimulating and anticipate a second edition incorporating developments inspired by it.

Contents:
1. Introduction
2. Model formulation
3. Examples
4. Finite-horizon Markov decision processes
5. Infinite-horizon models: Foundations
6. Discounted Markov decision problems
7. The expected total-reward criterion
8. Average reward and related criteria
9. The average reward criterion - multichain and communicating models
10. Sensitive discount optimality
11. Continuous-time models
APPENDIX A: Markov Chains
APPENDIX B: Semicontinuous Functions
APPENDIX C: Normed Linear Spaces
APPENDIX D: Linear Programming

The subject of Markov decision processes began forty years ago along with dynamic programming and it grew first in North America. Substantial European contributions helped it to maturity and added many ramifications. This book gives a unified and rigorous treatment of the theory and it also covers computational and applied research. It will serve as a graduate level course in operations research, but I think the mathematical requirements will make it less accessible to
readers from other disciplines. The theme is models in discrete time and problems of maximizing expected rewards over an infinite horizon, with or without discounting. Several methods, value iteration, policy iteration and linear programming, are described, and there is a detailed study of the theoretical subtleties generated by the average reward criterion. The book provides a comprehensive reference; I found the bibliographic remarks at the end of each chapter very helpful for their historical insight.

Contents:
General Notations
1. Semimartingales with spatial parameters and stochastic integrals
2. Stochastic differential equations
3. Stochastic differential delay equations
4. Exponential stability of stochastic differential equations
5. Almost sure exponential stability of stochastic differential delay equations
6. Moment exponential stability of stochastic differential delay equations
7. Exponential stability of stochastic differential equations with small time lag
8. Exponential stability of large-scale stochastic differential delay systems

Readership: Probabilists, stochastic modellers in engineering, biology, finance, etc.

The book is concerned with the solution of stochastic (delay) equations driven by nonlinear integrators and their exponential stability. (Exponential stability of a solution is the requirement of a bound which decreases at negative exponential rate as time goes to infinity and both a.s. and pth moment versions are considered.) A brief introduction to stochastic calculus is provided as a prelude. Emphasis is given to discussion of existence and uniqueness properties. Various applications are provided, for example concerning stochastic stabilization and destabilization, and stochastic flows and stochastic oscillators, to illustrate the theory. Nevertheless, the treatment is a little austere.

Contents:
1. Stable random variables on the real line
2. Multivariate stable distributions
3. Stable random processes and stochastic integrals
4. Dependence structures of multivariate stable distributions
5. Non-linear regression
6. Complex stable stochastic integrals and harmonizable processes
7. Self-similar processes
8. Chentsov random fields
9. Introduction to sample path properties
10. Boundedness, continuity and oscillations
11. Measurability, integrability and absolute continuity
12. Boundedness and continuity via metric entropy
13. Integral representation
14. Historical notes and extensions
APPENDIX A: Tables of symmetric á-stable fractals

Readership: Researchers in probability, applied probability and statistics, graduate students

This is the first book that is devoted solely to the topic of stable processes. The aim of the authors is to make this important branch of probability widely accessible and provide both an introduction and a basic reference. They succeed admirably on both counts and deserve to be congratulated on their achievement. They emphasize the probabilistic approach over the analytic one, featuring tails, moments when any exist, dependence structures, and focusing on multivariate and sample path properties. A first year graduate course in probability is the only background required for studying this excellent monograph. Proofs are presented in detail. Each chapter begins with a brief summary and concludes with exercises at varying levels of difficulty. Detailed hints are given on the more challenging problems. The first four chapters provide a good general overview of the subject. The remaining chapters contain more specialized material, while Chapter 14 contains informative historical notes, references and extensions to the topics covered in the text.

Title	MONOTONE STRUCTURE IN DISCRETE-EVENT SYSTEMS.
Author	P. Glasserman and D.D. Yao.
Publisher	New York: Wiley, 1994, pp. xiii + 297.

Reviewer:
Institute	The University of Newcastle
Place	Newcastle, Australia
Name	G.C. Goodwin

Title	LECTURES ON DISCRETE TIME FILTERING.
Author	R.S. Bucy with the assistance of B.G. Williams. C.S. Bussus Consulting Editor.
Publisher	New York: Springer-Verlag, 1994, pp. xv + 156, DM.98.00/ÖS.764.40/Sw.fr.98.00.

Reviewer:
Institute	Imperial College of Science, Technology and Medicine
Place	London, U.K.
Name	A.T. Walden

Title	AI AND COMPUTER POWER: The Impact on Statistics.
Author	D.J. Hand (Ed.).
Publisher	London: Chapman and Hall, 1994, pp. vii + 212, US$114.95.

Reviewer:
Institute	BP Exploration
Place	Sunbury-on-Thames, U.K.
Name	D.E. Wolstenholme

Reviewer:
Institute	University of Essex
Place	Colchester
Name	G.A. Barnard

Reviewer:
Institute	University of Chicago
Place	Chicago, U.S.A.
Name	S.M. Stigler

Reviewer:
Institute	Limburgs Universitaire Centrum
Place	Diepenbeek, Belgium
Name	N. Veraberbeke

Reviewer:
Institute	University of Chicago,
Place	Chicago, U.S.A.
Name	S.M. Stigler

Reviewer:
Institute	---------------------
Place	Ottawa, Canada
Name	R.G. Carter

Reviewer:
Institute	Victoria University
Place	Wellington, New Zealand
Name	D. Vere-Jones

Reviewer:
Institute	University of Waterloo
Place	Waterloo, Canada
Name	V.T. Farewell

Reviewer:
Institute	London School of Hygiene and Tropical Medicine
Place	London, U.K.
Name	B.L. De Stavola

Reviewer:
Institute	University of Cape Town
Place	Ronderbosch, South Africa
Name	J.M. Juritz

Reviewer:
Institute	ETH-Zentrum
Place	Zürich, Switzerland
Name	M. Maechler

Reviewer:
Institute	Macquarie University
Place	Sydney, Australia
Name	J.R. Leslie

Reviewer:
Institute	Queen's University
Place	Kingston, Canada
Name	J.T. Smith

Reviewer:
Institute	Southampton University
Place	Southampton, U.K.
Name	P. Prescott

Reviewer:
Institute	University of St. Andrews
Place	St. Andrews, U.K.
Name	C.D. Kemp

Reviewer:
Institute	Limburgs Universitair Centrum
Place	Diepenbeek, Belgium
Name	N. Veraverbeke

Reviewer:
Institute	The Open University
Place	Milton Keynes, U.K.
Name	D.J. Hand

Reviewer:
Institute	University of Newcastle
Place	Newcastle, Australia
Name	G.C. Goodwin

Reviewer:
Institute	McGill University
Place	Montreal, Canada
Name	J.O. Ramsay

Reviewer:
Institute	Nuffield College
Place	Oxford, U.K.
Name	D.R. Cox

Reviewer:
Institute	University of Southampton
Place	Southampton, U.K.
Name	P. Prescott

Reviewer:
Institute	University of Manchester Institute of Science and Technology
Place	Manchester, U.K.
Name	P.J. Laycock

Reviewer:
Institute	Imperial College of Science Technology and Medicine
Place	London, U.K.
Name	L.V. White

Reviewer:
Institute	Queen Mary and Westfield College
Place	London, U.K.
Name	D.J. Balding

Reviewer:
Institute	Harvard University
Place	Boston, U.S.A.
Name	S.W. Lagakos

Reviewer:
Institute	Stanford University
Place	Stanford, U.S.A.
Name	I. Olkin

Reviewer:
Institute	University of Calgary
Place	Calgary, Canada
Name	M.A. Maes

Reviewer:
Institute	University of Cambridge
Place	Cambridge, U.K.
Name	G.A. Young

Reviewer:
Institute	University of Wisconsin
Place	Madison, U.S.A.
Name	N.R. Draper

Reviewer:
Institute	University College London
Place	London, U.K.
Name	S.M. Pitts

Reviewer:
Institute	University of Washington
Place	Seattle, U.S.A.
Name	R. Pyke