ABSTRACT
A typical multiple imputation framework involves two parties or individuals:
An imputer and the analyst. An imputer working with a set of variables de-
velops predictive models for the missing values conditional on the observed
values and generates imputations. On the other hand, the analyst, perhaps
uses a subset of variables or even a subset of subjects for the analysis using
a model that may differ from the predictive model used by the imputer. It is
possible that the analyst uses the imputer model but settles on a statistically
less efficient procedure for convenience. These differences between the im-
puter and analyst introduce “uncongeniality.” The goal is to assess how such
uncongeniality affects the inferences for the analyst using multiply imputed
data sets.