摘要：One of the most common forms of data release by National Statistical
Institutes (NSIs) are frequency tables arising from censuses and surveys and
these have been the focus of statistical disclosure limitation (SDL) techniques for
decades. With the need to modernize dissemination strategies, NSIs are
considering web-based flexible table builders where users can generate their
own tables of interest without the need for human intervention. This has led to a
shift in traditional disclosure risks of concern and a move towards inferential
disclosure risk where statistical data can be manipulated and combined with
other data sources to reveal sensitive information with a high degree of
certainty. To protect against inferential disclosure risk, perturbative methods
with more formal privacy guarantees are necessary. We examine three posttabular
confidentiality protection methods of additive random noise that can
easily be applied ‘on-the-fly’ in a flexible table builder for generating survey
weighted frequency tables: the computer science approach guaranteeing a
formal privacy model called differential privacy and two SDL approaches of
post-randomization and a new technique called drop/add-up-to-q. We
demonstrate and compare their application in a simulation study based on
survey weighted counts in tables.
