PRAM and Disclosure Risk

Statistical Disclosure Control is part of the usual process, necessary for the dissemination of microdata. Usually, direct identifiers are removed and variables that can be used to re-identify certain records are recoded or suppressed. For certain surveys these methods are inadequate to produce safe microdatasets with enough detail, to satify both the supplier and the user of these sets. To that end, data perturbation techniques were developed. These techniques are meant to maintain more detail, while at the same time enough disclosure control is achieved. Basically, these methods can be devided into two groups. The first group consists of methods that attempt to ensure anonymity of the respondents during the interview, e.g., randomised response. The second group consists of methods that can be considered as part of the editing process, e.g., resampling, dataswapping and noise addition. PRAM (Post Randomisation Method) is a perturbation technique that has connections to both groups of methods. In one way it can be considered to be similar to randomised response. However, contrary to standard randomised respons techniques, PRAM is applied after the interview has been completed. Moreover, it can be considered to be some sort of missclassification as well, with known missclassification probabilities. Disclosure risk in case of the traditional methods like recoding and suppression, is usually defined in terms of the number of occurrences of certain combinations of categories. Often a simple rule is applied: whenever the frequency of such a combination in the population is above a certain threshold, that combination is considered to be safe. Intuitively this means, that whenever there are enough respondents satisfying that particular combination of categories, the uncertainty when matching such records with another file is still large enough. In case of PRAM, this rule can not be applied. In this paper we will propose a different approach to quantify the disclosure risk when applying PRAM.