Pseudo-Population Bootstrap Methods for Imputed Survey Data

• 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): Zeinab Mashreghi*+ and Christian Léger and David Haziza
• Companies: Université De Montréal and Université de Montréal and Université de Montréal
• Keywords: Bootstrap ; Imputation ; Imputation model approach ; Non-response model approach ; Variance estimation

Abstract

In the absence of non-response, pseudo-population bootstrap procedures have been extensively studied in the literature; see Gross (1980), Booth et al. (1994), Chauvet (2007) and Wang and Thompson (2012), among others. This type of bootstrap procedures consists of creating a pseudo-population from the original sample and selecting bootstrap samples from this pseudo-population using the same sampling design that was utilized to select the original sample. We extend these procedures to the case of item non-response, where linear regression imputation is used to compensate for the missing values. Our procedures can be used even if the sampling fractions are large. In the presence of imputed data, two inferential frameworks are used in order to study the properties of point and variance estimators: the non-response model approach and the imputation model approach. We present two pseudo-population bootstrap schemes: the first leads to consistent bootstrap variance estimators with respect to the non-response model approach, whereas the second scheme leads to consistent estimators with respect to the imputation model approach. Results from a limited simulation study will be presented.