Disclosure risks in odds ratios and logistic regression

Derrick, Ben; Green, Elizabeth; Ritchie, Felix; White, Paul

Abstract

When publishing statistics from confidential data, there exists a risk that the statistic might inadvertently reveal confidential information. Statistical disclosure control (SDC) aims to reduce that risk to an acceptable level. Most SDC theory is concerned with simple tables; there are very few guidelines covering complex statistics, and those mostly rely upon assumptions of negligible risk.
This paper examines the disclosure risk in odds and risk ratios and logistic regression. We demonstrate that calculated ratios should be classified as ‘unsafe’ statistics and modelled ratios as ‘safe statistics, in the jargon of the field. This allows us to develop best practice guidelines for checking outputs before release.

• Presentation Conference Type: Conference Paper (unpublished)
• Conference Name: Scottish Economic Society Annual Conference 2022: Special session 'Protecting confidentiality in social science research outputs'
• Start Date: Apr 25, 2022
• End Date: Apr 27, 2022
• Deposit Date: Aug 11, 2022
• Keywords: privacy, confidentiality, statistical disclosure control, odds ratio, logistic regression
• Public URL: https://uwe-repository.worktribe.com/output/9853160
• Publisher URL: https://www.res.org.uk/events-page/scottish-economic-society-2022-annual-conference.html