Statistical Disclosure Limitation with Multiple Imputation

Agencies wishing to release confidential microdata in public use files
for use by researchers and other agencies must do so without disclosing
confidential identities and attributes.  The proliferation of
record-matching software and publicly available databases that can be
used to re-identify records makes it increasingly difficult for agencies
to do so and they have become leery of releasing data at all.   When
they do release data it may be in a form of limited utility to the user.
Synthetic data sets, where some values are replaced with multiple
imputations, limit the risk of disclosure and allow for valid inferences
to be made.  I will review the use, generation, analysis of these data
and discuss some ongoing work.