A parametric approach to nonparametric statistics (Q721293)

According to the authors, the book bridges the gap between parametric and nonparametric statistics. The authors propose an embedding which focuses on local properties more in line with the concept of exponential tilting. The authors obtain a new derivation of the well-known Friedman statistic as the locally most powerful test in an embedded family of distributions. In addition, they exploit Hoeffding's change of measure formula which provides an approach to obtaining locally most powerful tests based on ranks of various multi sampling problems. The book presented consists of three parts. \par In the first part, the authors of the book present fundamental concepts of probability theory and list various probability distributions together with their principal numerical characteristics. The various modes of convergence, the central limit theorem and the multivariate central limit theorem are also discussed. In the section on statistical inference, the authors review some terminology and basic concepts: parametric estimation, hypothesis testing, contiguity. The exponential family of distributions, Cramer-Rao inequality, maximum likelihood estimator, three Le Cam's lemmas and Bayesian methods are also discussed. The third section of the first part is dealt with tools of the nonparametric statistics. Nonparametric statistics develops the distribution free methods to solve various statistical problems. The linear rank statistics, $U$ statistics and their main properties are presented in the third section. \par In the second part of the book, the authors present a large number of the nonparametric statistical methods. They describe Neyman's smooth test in connection with goodness of fit problems and they obtain test statistics for some common nonparametric problems. In Chapter 5, the use of this concept is discussed to the usual one-sample or two-sample tests. The authors begin by embedding the nonparametric problem into a parametric paradigm. Then they derive the score test statistics and find its asymptotic distribution. The construction of the parametric paradigm involves often the use of composite likelihood. In such a case, it is necessary to use either linear rank statistics or $U$ statistics in order to determine the asymptotic distribution of the test statistic. The parametric paradigm provides new insights into well-known problems. In Chapter 6, the authors present a unified theory of hypothesis testing and application of this theory to study the multi sample problems of location. In Chapter 7, a new method is presented to obtain the asymptotic distribution of the Friedman statistic. It is shown that this statistic is locally most powerful. Chapter 8 deals with locally most powerful tests, and the last chapter of this part of the book is devoted to the concept of efficiency. \par In the third part of the book, the authors consider some modern applications of the nonparametric statistics. Section 10 is devoted to the multiple change point problem. Here the problem is described, properties of the new model are discussed, and a composite likelihood function is constructed which permits to conduct test of hypotheses based on a score statistic to asses the significance of the change points. In Section 11, the authors propose a new Bayesian approach to study of ranking problems. The last twelfth section of the book is devoted to describe the application of the methodology to the analysis of censored data. \par Overall, the book is interesting and well written. Theoretical results and formulas derived are illustrated by various numerical examples. The majority of chapters are equipped with interesting exercises for the readers.