The use of multiple imputation for the analysis of missing data.

Abstract

This article provides a comprehensive review of multiple imputation (MI), a technique for analyzing data sets with missing values. Formally, MI is the process of replacing each missing data point with a set of m > 1 plausible values to generate m complete data sets. These complete data sets are then analyzed by standard statistical software, and the results combined, to give parameter estimates and standard errors that take into account the uncertainty due to the missing data values. This article introduces the idea behind MI, discusses the advantages of MI over existing techniques for addressing missing data, describes how to do MI for real problems, reviews the software available to implement MI, and discusses the results of a simulation study aimed at finding out how assumptions regarding the imputation model affect the parameter estimates provided by MI.

Localizing age-related individual differences in a hierarchical structure

“…Several imputation techniques are available (e.g., Schafer & Graham, 2002;Sinharay, Stern, & Russell, 2001), but there appears to be a consensus that the best in terms of accurately reproducing important characteristics of the parent population is a stochastic regression imputation in which the estimated values are obtained from a regression-based procedure that contains a stochastic or random residual term to better approximate the actual variance. The uncertainty introduced by the stochastic component can then be incorporated in the parameter estimates by generating m plausible values for each missing value, and analyzing these m data sets separately, with the values from the individual data sets combined and reported in terms of both the mean and standard error of each parameter (e.g., Schafer & Graham, 2002;Sinharay et al, 2001). …”

Worksite stress management training: Moderated effects and clinical significance.

“…There were no significant baseline differences on the GHQ between those who responded only at Time 1, and those who responded on more than one occasion. Moreover, the impact of participant attrition was greatly reduced by the multiple imputation (MI) procedure, which is now widely employed in the clinical literature (Sinharay et al, 2001). The pattern of results was similar in the MI datasets and in the reduced completer sample, further increasing confidence in the reliability of the findings.…”

“…MI is considered superior to other approaches for analyzing incomplete datasets as it takes into account the uncertainty due to missing information (Schafer, 1999;Schafer & Graham, 2002;Sinharay, Stern, & Russell, 2001). We employed Schafer's (1999) NORM software to generate five imputed datasets, and performed the same outcome and moderation analyses on each of these datasets; parameter estimates and standard errors were then pooled to obtain final estimates.…”

Advanced Statistics: Missing Data in Clinical Research—Part 1: An Introduction and Conceptual Framework

“…8,10 That is, whether a value is censored has no relationship with other known or unknown values in the data set, even if the pattern of censoring (see below) is not random. 3 Thus, for this assumption to exist, subjects with censored data are required to be a random sample of the study population, and those subjects without censored data are required to be a random sample of the source population.…”

“…8 Unfortunately, the MAR assumption cannot be tested and must be assumed unless censoring is explicitly introduced into the design of the study. 8,10,14 …”