DG Update: December 2001

Digital Government:

Project Update - December, 2001

1. RESEARCH UPDATE: OPTIMAL TABULAR RELEASES. At NISS, Adrian Dobra, Ashish Sanil and Alan Karr are adapting table server technology (for release of marginal tables over time with dynamic risk evaluation) to the design of optimal static releases of sets of marginals of large contingency tables. In the initial versions,

Released sets of marginals are restricted to correspond to the minimal sufficient statistics of decomposable models, which allows use of project-developed methods for explicit calculation of bounds on cell entries in the full table;

Risk is measured by tightness of bounds (computed as above) for sensitive, small-count cells in the full table;

Various optimality criteria are available, including the number of marginal tables released and the number of cells contained in the released marginals.

The methods appear to scale effectively; to date 14-way tables have been treated. The results are very interesting: in one case of a 13-way where releasing all two-way marginals would violate the risk criterion, optimal tabular release technology allows release of some 5-way marginals.

2. RESEARCH UPDATE: CYCLIC PERTURBATION. At CMU, George Duncan and Stephen Roehrig are working on a new method for disclosure protection in tables, currently called cyclic perturbation. It is an extension of work started in the 1998 Duncan and Fienberg paper "Obtaining information while preserving privacy; a Markov perturbation method for tabular data" given at SDP `98. Cyclic perturbation leaves table margins unchanged, but applies -1, 0, 1 moves to interior cells in a principled way. So principled, in fact, that we can analyze precisely disclosure risk and data utility, from a Bayesian perspective. Unlike rounding and cell suppression, cyclic perturbation allows a data user to reach concrete, accurate statistical conclusions from the published table. The method is applicable to higher-dimensional tables, as well as hierarchical tables.

3. PRESENTATIONS

National Center for Education Statistics, Washington:

George T. Duncan, Sallie Keller-McNulty and S. Lynne Stokes and Alan F. Karr: "Disclosure Risk vs. Data Utility: The R-U Confidentiality Map"

Haifa Winter Workshop on Computer Science and Statistics, Haifa Israel

Stephen E. Fienberg: "Conditional Distributions, Log-linear Models, and Disclosure Limitation Methods" (December 17, 2001)

Workshop on Privacy Protection by Misclassification, Erasmus University Rotterdam, The Netherlands:

Stephen E. Fienberg: "Conditional Distributions, Log-linear Models, and Disclosure Limitation Methods" (December 20, 2001)

4. NISS NEWSLETTER. Volume 1, Number 2 (January, 2002) is available on-line, at http://www.niss.org/newsletter.html

5. HAPPY NEW YEAR! All of us at NISS wish you and your families a safe, peaceful new year.

Navigation: www.niss.org > Digital Government > Updates > December 2001