Analytically Valid Discrete Microdata Files and Re-identification
William E. Winkler
KEY WORDS: Data Quality, Loglinear Model Fit,
Missing Data, Convex Constraints
ABSTRACT
Loglinear modeling methods have become quite
straightforward to apply to discrete data X. A
good-fitting loglinear model can be used to
generate synthetic copies of X1, …, Xn of X that
preserve analytic properties but may allow reidentification
of small cells. With fitting
algorithms that use more general convex constraints
and are designed to deal with missing data, we are
able to disperse the counts associated with small
cells over other cells in a manner that reduces reidentification
risk while still maintaining most
analytic properties.
CITATION: