Measuring Confidentiality Risks in Census Data

Two trends have been on a collision course over the recent past. The first is the increasing demand by researchers for greater detail and flexibility in outputs from the decennial Census of Population. The second is the need felt by the Census Offices to demonstrate more clearly that Census data have been explicitly protected from the risk of disclosure of information about individuals. To reconcile these competing trends the authors propose a statistical measure of risks of disclosure implicit in the release of aggregate census data. The ideas of risk measurement are first developed for microdata where there is prior experience and then modified to measure risk in tables of counts. To make sure that the theoretical ideas are fully expounded, the authors develop small worked example. The risk measure purposed here is currently being tested out with synthetic and a real Census microdata. It is hoped that this approach will both refocus the census confidentiality debate and contribute to the safe use of user defined flexible census output geographies.