9312143 Duncan This proposal is to develop and evaluate methods to measure and limit disclosure risk in statistical and multilevel relational databases. It addresses the ability of database users to infer sensitive information to which lack authorized access, based on responses to authorized queries. A decision-theoretic approach is proposed to allow analysis as an optimization problem of maximizing access to information subject to constraints on disclosure risk. Building on previous work (e.g., Adam and Wortmann (1989), Denning (1982), Duncan and Mukherjee (1991, 1992), Garvey, Lunt, and Stickel (1991) and Mukherjee, Krishnan, and Duncan (1992), the proposal is to accomplish the following tasks: 1. develop hybrid methods which combine existing disclosure limitation methods for statistical databases, and specify appropriate measures to evaluate their performance; 2. provide a probabilistic framework to evaluate disclosure limitation methods for statistical databases; 3. introduce a methodology that could be used at the time of schema design to detect and eliminate possible inference channels in multilevel relational databases, particularly in the presence of uncertainty about the external information used in the inference process.
