A Framework for Integrating Multiple Imputation and the Bootstrap

• Activity Number: 184
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Susan Shortreed*+ and Russell Steele
• Companies: Group Health Research Institute and McGill University
• Keywords: Missing Data ; Multiple Imputation ; Bootstrap ; Resampling methods

Abstract

Imputation is a common approach to avoiding potential biases and losses of efficiency that can accompany missing data. Multiple imputation methods create multiple completed data sets by resampling from a predictive distribution. However, standard multiple imputation formulae require a reliable estimate of the standard error of the complete data parameter. This is not always available. The bootstrap is a resampling method that is often used to estimate sampling variability when closed form solutions for standard errors do not exist or are biased. In this talk we will introduce and discuss approaches for integrating multiple imputation and the bootstrap. We will present the results of applying these methods in simulations settings and to estimate the effect breastfeeding on cognitive development.