Handbook of Missing Data Methodology

Book Preface

Missing data are a common and challenging problem that complicates the statistical analysis of data collected in almost every discipline. Since the 1990s there have been many important developments in statistical methodology for handling missing data. The goal of this handbook is to provide a comprehensive and up-to-date summary of many of the major advances. The book is intended to have a broad appeal. It should be of interest to all statisticians involved in the development of methodology or the application of missing data methods in empirical research.

The book is composed of 24 chapters, collected into a number of broad themes in the statistical literature on missing data methodology. Each part of the book begins with an introductory chapter that provides useful background material and an overview to set the stage for subsequent chapters. The first part begins by establishing notation and terminology, reviewing the general taxonomy of missing data mechanisms and their implications for analysis, and providing a historical perspective on early methods for handling missing data. The following three parts of the book focus on three alternative perspectives on estimation of parametric and semi-parametric models when data are missing: likelihood and Bayesian methods; semi-parametric methods, with particular emphasis on inverse probability weighting; and multiple imputation methods. These three parts synthesize major developments in methodology from the extensive statistical literature on parametric and semi-parametric models with missing data. Because inference about key parameters of interest generally requires untestable assumptions about the nature of the missingness process, the importance of sensitivity analysis is recognized. The next part of the book focuses on a range of approaches that share the broad aims of assessing the sensitivity of inferences to alternative assumptions about the missing data process. The final part of the book considers a number of special topics, including missing data in clinical trials and sample surveys and approaches to model diagnostics in the missing data setting.

In making our final selection of topics for this handbook, we focused on both established and emerging methodology for missing data that are advancing the field. Although our coverage of topics is quite broad, due to limitations of space, it is certainly not complete. For example, there is a related statistical literature on the broader concept of “coarsened” data (e.g., rounded, censored, or partially categorized data), of which missing data are a special case, that is not emphasized. In addition, we do not include the growing body of literature on causal inference as a missing data problem; other problems that can be put in the missing data framework, such as measurement error and record linkage, are only touched upon lightly. Nevertheless, we have tried to produce a handbook that is as comprehensive as possible, while also providing sufficient depth to set the scene for future research. We hope that this book provides the framework that will allow our readers to delve into research and practical applications of missing data methods.