Review of methods for data sets with missing values and practical applications

Abstract

The aim of this paper is to revise the traditional methods (complete-case analysis, available-case analysis, single imputation) and current methods (likelihood-based methods, multiple imputation, weighting methods) for handling the problem of missing data and to assess their usefulness in statistical research. The paper provides the terminology and the description of traditional and current methods and algorithms used in the analysis of incomplete data sets. The methods are assessed in terms of the statistical properties of their estimators. An example is provided for the multiple imputation method. The review indicates that current methods outweigh traditional ones in terms of bias reduction, precision and efficiency of the estimation.