Robust Statistics, 2nd edn

Robust statistical techniques are widely used by statisticians for estimating the parameters of a model if the assumptions regarding the model are not satisfied. These techniques are not unduly affected by outliers or departures from model assumptions and thus increase the accuracy of the inference. The first edition of this seminal book appeared in 1981 and since then it has been the leading book on this topic, providing an authoritative and systematic treatment of robust statistics for researchers and practitioners both. However, significant development has taken place in this field in the last 30 years which has warranted this second edition. This book provides an up-to-date exposition and comprehensive treatment of robust statistics. It will be helpful to graduate students and applied statisticians working in the area of time series and statistical and econometric modelling as well as researchers in the areas of engineering, medicine and biology where robust statistics are widely used. The new edition incorporates all the important developments on this topic since the publication of the first edition. Four new chapters have been included in the second edition. Chapter 11 deals with a robust method for the finite sample breakdown point, which can cause an estimator to take on arbitrarily large aberrant values. Chapter 13 discusses robust tests. Many classical tests do not satisfy the criterion of robustness of validity (the test should be stable under small arbitrary departures from the null hypothesis) and robustness of efficiency (the test should have good power under a small, arbitrary departure from specific alternatives). Robust tests for general parametric models in the multivariate case, and robust tests for regression and generalized linear models, are also discussed in this chapter. Another new chapter (Chapter 14) on small sample asymptotics has been added, which describes how saddle point techniques can be used to derive accurate approximations of the density and tail probabilities of robust estimators. Finally a new chapter on Bayesian robustness (Chapter 15) has been included in this new edition, which discusses some robustness issues in a different light by the Bayesian approach. Chapter 7 on robust regression has been expanded. The other chapters, which are not much different from those in the first edition, provide the mathematical foundation of robust statistics and their application in statistical inference, covariance and correlation, regression and robustness of design. All the chapters in this book open with ‘General remarks’ which have been expanded and provide readers with the motivation to understand the technique and tools that are discussed in the chapter. Overall the book is well organized and clearly written. This book cannot be recommended as a textbook as it lacks practice exercises. It is a good mixture of theory and practical applications. It may not be a suitable book for beginners but definitely graduate and research students will enjoy this book. Also, practitioners may find this book quite useful. I shall also recommend it for library purchase.