Missing data are frequently encountered in many applied statistical problems, especially in biostatistics. Three major approaches to handling missing data have emerged in the literature. These approaches are (1) Maximum Likelihood, (2) Multiple Imputation, and (3) Weighted Estimating Equations. There is much uncertainty about many aspects of missing data, including both the similarities and differences among the three major approaches and the risks involved with ignoring the missing data entirely.In this talk, we introduce the three methodologies given above and discuss the advantages and disadvantages of each from a practical perspective. We also discuss computational implementation and compare the performance of the methods in an example involving real data.
This is joint work with Joe Ibrahim at the Harvard School of Public Health and Stu Lipsitz at the Medical University of South Carolina.
Return to Biostatistics Working Group