Title: A Web--Based Query System for Disclosure-Limited Statistical Analysis of Confidential Data

Project Director: Alan F. Karr, NISS

Senior Investigators: George T. Duncan, Carnegie Mellon University; Stephen E. Fienberg, Carnegie Mellon University; Sallie Keller-McNulty, Los Alamos National Laboratory; Andrew Moore, Carnegie Mellon University; Stephen Roehrig, Carnegie Mellon University; Alan Saalfeld, Ohio State University; Latanya Sweeney, Carnegie Mellon University

Postdoctoral Fellows: Ashish Sanil, NISS

Graduate Research Assistants: Karen Brady, NISS; Adrian Dobra, Carnegie Mellon University; Christopher Holloman, Duke University

System Development: Susan Karimi, MCNC; Karen Littwin, MCNC; Bonnie Parrish, MCNC; Syam Sundar, MCNC

Federal Agency Partners: Bureau of Labor Statistics, Census Bureau, National Agricultural Statistics Service, National Center for Education Statistics, National Center for Health Statistics

Project Summary: This is a project to develop and build systems that expand access to Federal data but preserves confidentiality of the data and privacy of subjects. The systems will respond to queries from networked users of a Federal data bases by performing and reporting statistical analyses that extract knowledge from the data but preserve confidentiality. Their distinguishing characteristic is history-dependence: the response to each query will depend on the history of previous queries and responses.

The systems respond directly to the Federal government's role as the nation's largest gatherer and consumer of data. They will support core missions of Federal agencies as well as commercial and other public uses of Federal data, and will leverage limited agency resources that simply cannot meet burgeoning demands to access data for research purposes. The contexts include formulation and evaluation of policy at national, state and local levels, crisis management, and protection of the nation's infrastructure.

The research advances in computer science, social science and statistical science necessary to make such systems work are the thrust of the proposal. The specific issues are deep, but concrete and finite. The disciplines merge throughout, especially in regard to the metadata needed for the system to function, the multiplicity of roles of the query history data base, and the iteration between computation of disclosure risk and selection of risk reduction strategies.

Scale is another cross-cutting theme: existing techniques are untried at the scale (complexity of analyses, dimension and size of data sets, number of users) of the proposed system. One key challenge of the research is to make the system work.

The research is being performed by a cross-disciplinary, multi-institution research team, led by NISS. The team comprises researchers from NISS, Carnegie Mellon University, Kansas State University, Los Alamos National Laboratory, MCNC, Ohio State University and the University of Maryland College Park. Leading Federal statistics agencies, as partners in the proposal, provide access to data and personnel to participate in development and evaluation of the systems.