Maximum Likelihood Multiple Imputation: Faster Imputations and Consistent Standard Errors Without Posterior Draws

Multiple imputation (MI) is a method for repairing and analyzing data with missing values. MI replaces values sample of random drawn from an model. The most popular form MI, which we call posterior draw multiple (PDMI), draws the parameters model Bayesian distribution. An alternative, maximum likelihood (MLMI), estimates using (or equivalent). Compared to PDMI, MLMI faster yields slightly more efficient point estimates. A past barrier was difficulty estimating standard errors We derive, implement evaluate three consistent error formulas: (1) one combines variances within between imputed datasets, (2) uses score function (3) bootstrap two imputations each bootstrapped sample. Formula modifies formula that has long been used under while formulas can be without modification either PDMI or MLMI. have implemented estimators in mlmi bootImpute packages R.