The ability to quantifying certainty of an estimated or predicted quantity is key to practical and realistic decision-making. Hence, realistic and correct statistical intervals are essential to all statistical inferences. The book by Meeker, Hahn, and Escobar is a must-have on the bookshelves of practitioners, researchers and students as it offers by far the most complete and up-to-date collection of statistical methods and computing tools for statistical intervals.
The authors have done a remarkable job of making a large collection of important methods accessible and usable. The book also provides a good framework for identifying the right methods that match a given problem. The first edition of this book was published in 1991, which was timely to meet the desire of a single book wholly designated for a comprehensive overview of statistical intervals for guiding practitioners and statisticians working in diverse areas. Twenty-six years later, this much-anticipated second edition is now out with an extensive revision and substantial extensions to include more modern computational-driven techniques as well as up-to-date computing resources for calculating statistical intervals.
The major new additions in this new edition include:
Six new chapters with a comprehensive exposition of methodologies that were newly developed in the past two decades for calculating statistical intervals. These include the likelihood-based, bootstrap and simulation-based, and Bayesian approaches.
A new chapter (Chapter 18) which offers a diverse set of advanced case studies for illustrating the applicability, flexibility and efficiency of the modern methods for constructing statistical intervals for complex problems which involve the use of nonlinear regression models, random effects models and/or incomplete or correlated data.
An R package StatInt that offers additional functions for computing statistical intervals for distributions not available in R (such as the smallest and largest extreme distributions, the Frechet, loglogistic, log-Cauchy, log-uniform, beta-binomial and negative-hypergeometric distributions). This is a quite valuable addition to existing computing recourses for handling broader applications.
A collection of resources which are available at the companion website and provide the up-to-date information about the availability and capability of various easy-to-access statistical software (such as JMP, Minitab, SAS, and SPSS) for calculating statistical intervals. Many of the modern methods rely on the largely enhanced computing power and are generally computationally intensive. The R code and computing resources offered in this book are valuable for practitioners to easily and efficiently implement the methods without requiring more advanced analytical or computing skills.
An extensive collection of technical appendices which offers in-depth explanations about the underlying theory for statistical intervals to meet the needs of more advanced users and researchers. In addition to the general methods, this book also covers more advanced topics such as generalized pivotal methods and distribution-free intervals which are often not available in most introductory statistical textbooks.
The book contains 18 chapters. The first two chapters offer an introduction and an overview of the commonly used different types of statistical intervals. Chapters 3 and 4 discuss simple tabulation and other methods for constructing statistical intervals based on a normal distribution. Chapter 5 discusses distribution-free intervals. Chapters 6 and 7 focus on statistical intervals for discrete (Binomial and Poisson) distributions. Chapters 8–10 talk about how to determine the sample size requirements for various statistical intervals. Chapter 11 provides a large collection of case studies for illustrating the methods described in Chapters 3–10. The first 11 chapters are more similar to the first edition with modifications and modernization to reflect more update-to-date implementation of the methods with some older versions of chart-based approaches being moved to the companion website.
Chapters 12–18 are completely new additions. Chapter 12 discusses likelihood-based intervals for a broad class of distributions. Chapters 13 and 14 describe nonparametric and parametric bootstrap intervals as well as other simulation-based approaches. Chapters 15–17 introduce Bayesian credible intervals for continuous and discrete distributions, as well as statistical intervals for Bayesian hierarchical models. Chapter 18 is the complement to Chapter 11 with additional advanced case studies using the modern computing intensive methods described in Chapters 12–17.
I found this book to be well-written and easily accessible for readers with a broad variety of technical backgrounds. The book offers a nice balance between the depth and breadth of the topics. The chapters are well organized and cover a wide spectrum of topics with a clear progression from basic to more advanced techniques, from general to specialized approaches, and from model-based to model-free methods. The main chapters are clearly and concisely written for practitioners with interest in implementation for real applications. The large number of case studies offers a versatile collection of important real-world applications to attract broad interests and connect with diverse users. The addition of the technical appendices offers more details for statisticians and savvy researchers, and the comprehensive list of references provides additional resources for methods that were not covered in depth in the book.
In addition, the book has included many state-of-art techniques, which is of great help to researchers for inspiring cutting-edge research and innovations. I am a good example of a researcher who has greatly benefited from this book. For example, I learned about the fractional random weight bootstrap approach from this book, which is a better alternative to the regular nonparametric bootstrap approach that I was already familiar with. I have applied this new technique in several of my research projects to achieve improved results.
In concluding, I highly recommend this book to all the statisticians, practitioners and researchers who use data to aid understanding, decision-making and problem-solving. The book serves as a comprehensive handbook for constructing statistical intervals for broad groups of users and also catalyzes research in advanced statistical inferences and uncertainty quantification.
University of South Florida