Making the Release of Confidential Data from Multi-Way Tables Count

tatistical disclosure limitation (SDL) and confidentiality have often been shrouded with a nonstatisti-cal veil and the methodology for protecting confidential data has produced problematic outcomes for research data users. Here we describe one possible statistical approach to SDL for data in the form of multidimensional contingency tables that illustrates the following points: • For categorical data the traditional form of reporting has been marginal tables and conditionals. • Releasing such partial information is compatible with and useful for statistical methods for log-linear models and directed acyclic graphs. • Interesting new research problems arise in this area. From the early part of the 20 th century , confidentiality has been an important element of the mantra of statistical agencies, and it became embedded in the culture of the U.S. Census Bureau with the protections associated with the 1929 Census Act (now known as Title 13 of the U.S. Code). But the term confidentiality was always thought about in terms of the protection of individual and establishment data and not the release of data to policy makers, researchers, and the public. Moreover, confidentiality represented for many an " absolute " concept and it was not until the 1970s that there was movement toward making statistical thinking more central to the operational implementation of confidentiality protection. More specifically, the President's Commission on Federal Statistics, issued in 1971, placed special emphasis on confidentiality and subsequently , when the Office of Management and Budget's Statistical Policy office created the Federal Committee on Statistical Methodology, one of its first activities was a study of confidentiality and statistical disclosure protection (Working Paper No. 2). Working paper No. 2 was of special interest in part because it signaled for the first time the importance of the The confluence of research on disclosure methods and of research on log-linear model theory uncover the essential elements needed by analysts working with discrete data. trade-off between access and confidentiality and presented Tore Dalenius's probabilistic notion of disclosure: " If the release of the statistics S makes it possible to determine the value [of confidential statistical data] more accurately than is possible without access to S, a disclosure has taken place. " [Working Paper No. 2, pp. 7 and 9] The past 25 years have seen the growth of disclosure limitation as a statistical subdiscipline, and the term itself which was a change from that used in the 1970s, recognized that the …