Missing Data

Missing data occur almost universally in research, especially applied research. This can give rise to problems of both statistical estimation and interpretation. The work of Donald Rubin in the 1970s was highly influential on researchers’ theoretical and practical understanding of ‘missingness’ – the state of being missing. Previously, approaches to missingness tended to exclude all participants with missing responses (e.g., complete case analysis) or to fill missing data with arbitrary values, such as sample means (e.g., mean imputation). These seemed to solve issues of estimation but merely deferred problems of interpretation. Contemporary approaches to missingness begin by classifying the missingness as either ignorable (missing at random or completely at random) or non-ignorable (missing not at random). Consequently, imputation and maximum likelihood approaches can be employed that minimise problems of estimation and maximise the validity of interpretation. However, when missingness is not at random, problems remain. The chapter concludes with recommendations for researchers.