Bayes and Multiple Imputation

Roderick J. A. Little,Donald B. Rubin
DOI: https://doi.org/10.1002/9781119013563.ch10
2002-01-01
Abstract:When sample sizes are small, a useful alternative approach to multiple imputation (ML) is to add a prior distribution for the parameters and compute the posterior distribution of the parameters of interest. As with ML estimation with a general pattern of missing values, Bayes simulation requires iteration. The iterative simulation methods discussed eventually create draws from the posterior distribution of θ. However, the five draws of Ymis can be quite adequate for generating MI inferences, provided the fraction of missing information is modest, as when the fractions of cases with Y1 or Y2 missing are limited.
What problem does this paper attempt to address?