Multiple Imputations Quality Assessment for Survey Data

• 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): Jingchen Liu*+ and Xiao-Li Meng
• Companies: Columbia University and Harvard University

Abstract

Imputation is a convenient tool for the analysis of incomplete data set via complete data procedures. Checking the imputation quality is generally a very challenging task especially for the data users. The imputations are usually produced based on statistical models, for which the user has little or no knowledge. Therefore, no "Gold standard" is available to check against. In this talk, we cast this problem into the setting of two-group comparisons: observed data versus imputed data. Typically, the imputed data are highly dependent on the observed data. This makes it hard to assess how close is close to enough for a good fit. More precisely, the challenge lies in the variance calculation due to the high dependence between the observed and imputed data.