Sensitivity of Multiply Imputed Results to Quantity and Differential of Missingness

• Activity Number: 202
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 10:30 AM to 11:15 AM
• Sponsor: Section for Statistical Programmers and Analysts
• Author(s): Chad Evans*
• Companies: University of Pennsylvania
• Keywords: missingness ; simulation ; differential ; sensitivity ; regression ; inference

Abstract

More and more, missing data in sample surveys impugn the inferences that can be made about populations. High missingness in all but rare circumstances leads to inefficiency and often biases results as well. One of the most effective approaches to handling missing data is the technique of multiple imputation. While the advantages of multiple imputation are generally well known, its performance is less understood when missing data are not missing at random (NMAR). In an effort to "make better decisions" in statistics, this paper explores the sensitivity of regression results when data with different features of missingness are multiply imputed. I utilize simulations towards this end. Contrary to expectations, I argue that results based on multiple imputation may be most sensitive to higher levels of missingness; moreso than they are for the skewness of this missingness (differential missingness). Caveats are discussed in the discussion section.