Chapter twenty-four - Missing Data Analysis

Summary

This chapter provides a nontechnical overview of missing data issues, with a particular emphasis on multiple imputation and maximum likelihood estimation. It describes missing data mechanisms such as missing completely at random (MCAR), missing at random (MAR), and not missing at random (NMAR) that explain why data are missing. The chapter discusses planned missing data designs that can reduce respondent burden and lower the cost of data collection. It explains traditional missing data handling methods commonly used by researchers, and explains why these approaches produce biased parameter estimates and reduce the statistical power to detect effect. Multiple imputation consists of three phases: the imputation phase, the analysis phase, and the pooling phase. The chapter also discusses practical considerations that influence the choice between multiple imputation and maximum likelihood estimation. It explains the class of analysis models that are designed to address the difficult case of not missing at random data.