|Friday, February 22|
|CS05 Theme 2: Data Modeling and Analysis #2||
Fri, Feb 22, 10:45 AM - 12:15 PM
Multiple Imputation: An Introduction and Applications (302473)
Keywords: Missing data, inference, measurement error, health data, surveys.
Multiple imputation is a technique for handling missing data that fills in (i.e., imputes) the missing values several times from a predictive distribution to allow subsequent analyses to account for the uncertainty due to missing data in a straightforward way. This talk will outline the basic theory and methods for creating multiple imputations and analyzing multiply imputed data, and it will discuss considerations for the validity of inferences with imputed data. Several software packages for implementing multiple-imputation methods will also be mentioned, as will several references on the topic. Following this introduction, various recent or potential applications of multiple imputation at the National Center for Health Statistics, Centers for Disease Control and Prevention will be described to illustrate different types of problems to which the technique can be applied. The applications will include a traditional missing-data problem (e.g., due to nonresponse), a problem of “bridging” the transition between two different data reporting systems, and a problem of measurement error.