Cell Suppression: Experience and Theory [chapter]

Dale A. Robertson, Richard Ethier
2002 Lecture Notes in Computer Science  
Cell suppression for disclosure avoidance has a well-developed theory, unfortunately not sufficiently well known. This leads to confusion and faulty practices. Poor (sometimes seriously flawed) sensitivity rules can be used while inadequate protection mechanisms may release sensitive data. The negative effects on the published information are often exaggerated. An analysis of sensitivity rules will be done and some recommendations made. Some implications of the basic protection mechanism will
more » ... explained. A discussion of the information lost from a table with suppressions will be given, with consequences for the evaluation of patterns and of suppression heuristics. For most practitioners, the application of rules to detect sensitive economic data is well understood (although the rules may not be). However, the protection of that data may be an art rather than an application of sound concepts. More misconceptions and pitfalls arise. Cell suppression is a technique for disclosure control. It is used for additive tables, typically business data, where it is the technique of choice. There is a good theory of the technique, originally developed by Gordon Sande [1,2| with important contributions by Larry Cox [3] . In practice, the use of cell suppression is troubled by misconceptions at the most fundamental levels. The basic concept of sensitivity is confused, the mechanism of protection is often misunderstood, and an erroneous conception of information loss seems almost universal. These confusions prevent the best results from being obtained. The sheer size of the problems makes automation indispensable. Proper suppression is a subtle task and the practitioner needs a sound framework of knowledge. Problems in using the available software are often related to a lack of understanding of the foundations of the technique. Often the task is delegated to lower level staff, not properly trained, who have difficulty describing problems with the rigour needed for computer processing. This ignorance at the foundation level leads to difficulty understanding the software. As the desire for more comprehensive, detailed, and sophisticated outputs increases, the matter of table and problem specification needs further attention. Our experience has shown that the biggest challenge has been to teach the basic ideas. The theory is not difficult to grasp, using only elementary mathematics, but clarity of thought is required. The attempt of the non-mathematical to describe things
doi:10.1007/3-540-47804-3_2 fatcat:ycq3hlmtijepnel7u6txwv5iha