ABSTRACT
Murdoch Children’s Research Institute & The University of Melbourne, Melbourne, Australia
This chapter provides an introduction to the method of multiple imputation, encompassing an overview of its historical origins, theoretical foundations and application in practice. In presenting such a broad overview there are inevitable gaps and a lack of depth on many topics, but subsequent chapters provide more detail in many areas. The first section provides a simple description of the method, accompanied by a detailed illustrative example, and a brief historical account of its origins in large-scale sample survey methodology. The following section then gives an explanation of the formal basis of the method in Bayesian statistical inference, as developed by Rubin in the 1970s, and a review of some of the key theoretical concepts. The application of multiple imputation to practical data analysis has developed within the framework of “ignorable” models (see Chapter 3), and Section 12.3 reviews the
tools that appeared from the mid-1990s to implement the approach within this framework. These tools have sparked much more widespread uptake of the method in practice. The final section reviews the range of problems to which multiple imputation has been applied, beginning with several large computational efforts in the late 1980s and early 1990s and concluding with a brief overview of the current range of applications and outstanding methodological research problems.
