Presentation Date:
Location:
Paul von Hippel (UT- Austin) presents
Title: Maximum likelihood multiple imputation: A more efficient approach to repairing and analyzing incomplete data
Abstract: Maximum likelihood multiple imputation (MLMI) is a form of multiple imputation (MI) that imputes values conditionally on a maximum likelihood estimate of the parameters. MLMI contrasts with the most popular form of MI, posterior draw multiple imputation (PDMI), which imputes values conditionally on an estimate drawn at random from the posterior distribution of the parameters. Despite being less popular, MLMI is less computationally intensive and yields more efficient point estimates than PDMI. A barrier to the use of MLMI has been the difficulty of estimating standard errors and confidence intervals. We present three straightforward solutions to the standard error problem.
The paper is available at: https://arxiv.org/abs/1210.0870