Preserving Relationships Between Variables with MIVQUE-Based Imputation for Item Nonresponse in Surveys

• Activity Number: 434
• Type: Contributed
• Date/Time: Tuesday, August 6, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Survey Research Methods Section
• Author(s): Brigitte Gelein*+ and David Causeur and David Haziza
• Companies: ENSAI and Agrocampus Ouest and Université de Montréal
• Keywords: item nonresponse ; imputation ; MIVQUE ; relationships

Abstract

Nonreponse may affect the quality of the estimates produced by statistical agencies when the respondents and the nonrespondents do not share the same characteristics with respect to the study variables. In this work, we focus on item nonresponse. We study the problem of preserving the relationship between items requiring imputation. Shao and Wang (2002) proposed a joint random regression imputation procedure and showed that it leads to asymptotically unbiased estimators of coefficients of correlation. We propose a calibrated imputation procedure, which consists of two steps: in the first step, missing values are imputed using the Shao and Wang procedure. In the second step, the imputed values derived in the first step are modified so that the imputed estimators of the first and second moments as well as the imputed estimator of the cross-product are calibrated on Minimum In Variance QUadratic Estimators (MIVQUE). More specifically, we seek a new set of imputed values in step 2, as close as possible to the original set of imputed values, so that appropriate constraints are satisfied. Results from a simulation study suggest that the resulting imputed estimators are efficient.