Estimation in Surveys with Nonresponse

Abstract

In writing this book, we found inspiration in the extensive literature on nonresponse. Although further development is bound to follow, we believe that our book has something to offer to all categories of professionals engaged in surveys and statistics production, survey managers, subject-matter specialists, as well as specialists in statistical survey methodology. The background knowledge required to assimilate the contents of this book varies between the chapters. Roughly the first one-third of the book is easy to follow for all categories; the rest requires more of a technical preparation. Chapters 2 and 3 give a wholly nontechnical overview. They can be read with only a minimal background in statistical science. They emphasize general issues in the treatment of nonresponse and provide a general orientation in the field. Technical arguments and formulae are essentially absent. Chapters 4–14 form a logical sequence in which one chapter builds on the preceding ones. The theory of estimation in the presence of nonresponse is developed from Chapter 6 onwards. Many of the notions of the classical survey theory are preserved. There is a finite population of identifiable elements with which one can associate individual probabilities of being selected and of responding. The classical randomization theory is obtained as a special case, namely, when the nonresponse is reduced to nil. The contents of Chapters 4–14 can be described as moderately technical. To fully appreciate these chapters, readers should have been exposed to at least one thorough course in statistical inference, including principles of point estimation and confidence intervals. Familiarity with linear models and regression analysis, including some linear algebra, is important background. It is a considerable advantage also to have followed one or more courses in modern survey sampling theory and techniques. Calibration is the unifying concept that keeps the chapters together. Calibration is a technique for computing weights to be used in estimation, given an input of auxiliary information. Persons working actively with survey methodology, in statistical agencies or survey institutes, should be able to easily follow the whole book. Readers with practical experience with surveys and statistics production are in a favoured position. They can identify with many issues raised in the book. Their practical background will facilitate their reading. A reader who is a first-year graduate student at university may start from a background derived mainly from a set of undergraduate statistics courses and with little or no exposure to the reality of surveys. Nevertheless, the book should find use in a graduate course within a university statistics curriculum. Some more technical material – notably derivations and lengthier proofs of certain propositions – is placed in end-of-chapter appendices. These need not be read or understood if the primary objective is to see the practical utility of the methods. This book evolved from a Current Best Methods (CBM) manual which we wrote in 1999–2001 for Statistics Sweden. Its title is Estimation in the Presence of Nonresponse and Frame Imperfections. A far-reaching revision and extension PREFACE xi of the contents of that manual was necessary to meet the objective that we