Online Program Home
My Program

Abstract Details

Activity Number: 237 - SPEED: Missing Survey Data: Analysis, Imputation, Design, and Prevention
Type: Contributed
Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
Sponsor: Survey Research Methods Section
Abstract #329267 Presentation
Title: Degrees of Freedom in Multiple Imputation: The Original vs. The Adjusted in 2015 National Hospital Ambulatory Medical Care Survey
Author(s): Qiyuan Pan* and Rong Wei
Companies: CDC/NCHS/DHCS and National Center for Health Statistics
Keywords: Adjusted degrees of freedom; Fraction of missing information; Missing data; Multiple imputation; National Hospital Ambulatory Medical Care Survey

The original degrees of freedom (DFori) for multiple imputation (MI) was defined by Rubin in 1987. In 1999, Barnard and Rubin proposed the adjusted degrees of freedom (DFadj) to correct the potential problems caused by the DFori's being larger than the complete-data degrees of freedom (DFcom). The adjustment that generates DFadj is a mechanism that forces DFadj to be always smaller than DFcom. The difference between DFori and DFadj can be extremely large. For example, a DFori of 1,000 may correspond to a DFadj of 20. At present DFadj is essentially the default for all types of MI analyses. However, the gains and biases from using DFadj and DFori have not been adequately studied. This research evaluates the impacts of DFadj and DFori on t tests and parameter estimations. Bootstrapping samples of various DFcom values were formed by resampling the 2015 National Hospital Ambulatory Medical Care Survey. MI simulations were carried out on these bootstrapping samples. The results suggest that there are situations under which the use of DFori is more appropriate than DFadj even though DFori > DFcom has occurred.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2018 program