Comparison of different sensitivity rules for tabular data and presenting a new rule -- the interval rule

Abstract

Statistical disclosure control (SDC) is a set of methods that are used to reduce the risk of disclosing information on individuals, businesses or other organisations. The focus of this paper is on sensitivity rules, which deal with how to define whether a cell in tabular data has the risk of disclosing information or not. The current popular sensitivity rules include the dominance rule and the P% rule. There is a weakness with these rules and a new rule - the interval rule is presented. The main argument for this new rule is that the rule should only be based on the information that the intruder knows, not on the information that the statistical institution knows. Based on simulated data, the P% rule tends to classify a dataset to be “sensitive” when it contains only one observation with a very large value. In this respect, and the dominance rule and the P% rule share a lot in common. Meanwhile the interval rule tends to classify a dataset to be “sensitive” when it contains two observations with large values.