Multiple imputation for multivariate missing-data problems: A data analyst's perspective

Analyses of multivariate data are frequently hampered by missing values. Un til recently, the only missing-data methods available to most data analysts have been relatively ad hoc practices such as listwise deletion. Recent dr amatic advances in theoretical and computational statistics, however, have produced a new generation of flexible procedures with a sound statistical b asis. These procedures involve multiple imputation (Rubin, 1987), a simulat ion technique that replaces each missing datum with a set of m > 1 plausibl e values. The Nz versions of the complete data are analyzed by standard com plete-data methods, and the results are combined using simple ru les to yie ld estimates, standard errors, and p-values that formally incorporate missi ng-data uncertainty. New computational algorithms and software described in a recent book (Schafer, 1997a) allow us to create proper multiple imputati ons in complex multivariate settings. This article reviews the key ideas of multiple imputation, discusses the software programs currently available, and demonstrates their use on data from the Adolescent Alcohol Prevention T rial (Hansen & Graham, 1991).