On a disclosure probability statement for the 5 safes framework

Article type: Research Article

Authors: Tam, Siu-Ming

Affiliations: National Institute of Applied Statistical Research, University of Wollongong, Northfields Ave, Wollongong, NSW 2522, Australia | E-mail: [email protected]

Correspondence: [*] Corresponding author: National Institute of Applied Statistical Research, University of Wollongong, Northfields Ave, Wollongong, NSW 2522, Australia. E-mail: [email protected].

Abstract: With increasing demand from the research community for more frequent and unrestricted access to data, national statistical offices (NSOs) are adopting the 5 safes framework to manage the disclosure risk for releasing such data. In this paper, under some mild conditions, we show that the probability of disclosure, given the controls in the 5 safes, is not greater than the product of the smallest conditional disclosure probability amongst the 5 controls and the Risk Ratios of the remaining four safe controls. By computing the disclosure probabilities of all possible configurations of the controls in each of the 5 dimensions of the framework, one can select the set which has the least control on data, but which also meet the confidentiality and privacy requirements of the NSO. Where the required assumption of unconditional independence of the safes cannot be met, the paper proposes a merger of some of the controls to overcome the violation.

Keywords: Disclosure probability, disclosure risks, risk ratio

DOI: 10.3233/SJI-200677

Journal: Statistical Journal of the IAOS, vol. 37, no. 2, pp. 693-698, 2021