ON THE STATIONARY DISTRIBUTION OF ITERATIVE IMPUTATIONS

By Jingcheng Liu, Andrew Gelman, Jennifer L. Hill, Yu-Sung Su. 2010

130 citations recorded by the source corpus.

Iterative imputation, in which variables are imputed one at a time conditional on all the others, is a popular technique that can be convenient and flexible, as it replaces a potentially difficult multivariate modelling problem with relatively simple univariate regressions. In this paper, we begin to characterize the stationary distributions of iterative imputations and their statistical properties, accounting for the conditional models being iteratively estimated from data rather than being prespecified. When the families of conditional models are compatible, we provide sufficient conditions under which the imputation distribution converges in total variation to the posterior distribution of a Bayesian model. When the conditional models are incompatible but valid, we show that the combined imputation estimator is consistent.

Subfields: Artificial Intelligence, Management Science and Operations Research, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/d7ce745d427d2bc0ad4bf73fce50418db297af2a

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Management Science and Operations Research

In Statistics and Probability

Citations in the corpus

Multiple imputation for continuous variables using a Bayesian principal component analysis† (2014, background): “More generally, the theoretical properties of chained equations are not well understood and they are a current topic of research.[8] Both the joint and conditional methods have their own advantages and drawbacks as investigated recently in [9].”
Multiple imputation of covariates by fully conditional specification: Accommodating the substantive model (2012, unknown): “contains the true probability distribution), we say (again following Liu et al.(13)) that the set of models fhj; j 1⁄4 1, .”