The Sin of Missing Data: Is All Forgiven by Way of Imputation?

Abstract

Missing data can seriously compromise inferences from clinical research studies. One method for handling missing data is to substitute each missing value with a reasonable guess, and then carry out the analysis as if there were no missing values (imputation). Single imputation replaces a missing value based on a predefined rule and includes last observation carried forward, worst observation carried forward, and simple mean imputation. In general, single imputation methods are not recommended. Multiple imputation is a statistical technique that creates multiple complete datasets, substituting missing with imputed values. The datasets are analyzed separately, and the results pooled into a final result. Whether to use multiple imputation for missing data depends on the missing data mechanism (reason for missing data). To properly use multiple imputation, the missing data should be MAR.