What is PAGE?

We represent a community with a shared interest in data analysis using the population approach.


2001
   Basel, Switzerland

Multiple Imputation for Missing-Data Problems

Joe Schafer

Penn. state Univ, USA, PA 16802, Unversity Park, 325 Joab L Thomas Building

Multiple imputation (MI) is a general-purpose technique for handling missing data. In MI, each missing value is replaced by M > 1 simulated values drawn from a predictive distribution, producing M different versions of the complete data set. Each version is analyzed in the same fashion, and the results are combined to yield parameter estimates and standard errors that account for uncertainty due to missing data. This presentation will provide an overview of MI, including its advantages over other commonly used missing-data methods. We will answer frequently asked questions about MI, discuss computational methods and software for creating imputations, and present an example application of MI to longitudinal data.

Software for Multiple Imputation

Many statistical packages do not handle missing data well. Cases with missing values are typically discarded, resulting in substantial loss of information and perhaps biasing the results in unpredictable ways. Multiple imputation (MI) is a valuable alternative which allows the user to retain all observed data and, at the same time, obtain proper estimates and standard errors. In this presentation, we will demonstrate software for creating multiple imputations in the Windows 95/98/NT environment. These programs, which are distributed free of charge, make it easy for users to generate imputations under a variety of multivariate models. The programs are not designed to replace well-established statistical packages like SAS or SPSS; they do not perform statistical analyses for you. Rather, they act as pre-processors, filling in the missing values so that other statistical packages can make full use of your data. They also act as post-processors, combining the output from the m analyses to produce a single set of results.



Top