Operationalising ‘safe statistics’: the case of linear regression

The recent growth in research access to confidential government microdata has prompted the development of more general 'output-based statistical disclosure control' (OSDC) methods which go beyond tabular protection. Central to OSDC is the concept of 'safe/unsafe statistics', allowing researchers and facility owners to make informed judgments about the types of research output that pose a disclosure risk. While increasingly accepted in specialist environments, in the wider community this novel approach causes some concern: how can 'safe' be unconditional? This paper therefore demonstrates the new approach using linear regression, a key research output, as an example. In doing so, the paper reconsiders the objectivity of SDC decision-making, arguing that ‘safety’ be explicitly acknowledged as a relative concept.