Multiple Imputation Framework for Combining Data from Multiple Sources

• Activity Number: 9
• Type: Invited
• Date/Time: Sunday, August 1, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Trivellore Raghunathan*+
• Companies: University of Michigan
• Keywords: Combining Information ; Complex Surveys ; Missing Data ; Bayesian approach ; Propensity score ; Sequential Regression

Abstract

Increasingly no single study can provide data to fully understand the role of biological, social and environmental factors on health. However, there are surveys/studies that collect data focusing on specific scientific phenomenon and others collect data for more general purposes with some common sets of variables. For example, suppose that three sources are: 1)A population based genetic study on genotypes(X) and disease(D); 2)A population-based epidemiologic study on environmental factors and biological markers(Y) and Genotypes(X); and 3) A general purpose health survey collecting risk factors(Y) and Disease(D). By concatenating these data sets, the missing pieces in each survey/study can be treated as missing data and handled through multiple imputation. This talk will use an example with several data sets to illustrate the methodology, discuss challenges, and evaluate the inferences.