Chapter 3 Univariate missing data
Statistics is a missing-data problem.
— Roderick J.A. Little
Chapter 2 described the theory of multiple imputation. This chapter looks into ways of creating the actual imputations. In order to avoid unnecessary complexities at this point, the text is restricted to univariate missing data. The incomplete variable is called the target variable. Thus, in this chapter there is only one variable with missing values. The consequences of the missing data depend on the role of the target variables within the complete-data model that is applied to the imputed data.
There are many ways to create imputations, but only a few of those lead to valid statistical inferences. This chapter outlines ways to check the correctness of a procedure, and how this works out for selected procedures. Most of the methods are designed to work under the assumption that the relations within the missing parts are similar to those in the observed parts, or more technically, the assumption of ignorability. The chapter closes with a description of some alternatives of what we might do when that assumption is suspect.