Disclosure control for regression outputs

Disclosure detection and control for analytical outputs is an almost unexplored field. However, with the increase in access to detailed microdata, it is becoming increasingly important to be able to quantify exactly what the risks are from allowing, for example, regression coefficients to be released.

This paper looks in detail at the risks of linear regressions, and demonstrates that, even in the best-case scenario for an intruder, analytical results are fundamentally safe, and can be made utterly non-disclosive by the application of simple rules. Estimation of the risk of likely disclosure is also considered, and it is shown that the NSI can carry out its own safety tests easily, and can also prevent intruders generating meaningful fitted values by application of the same rules. Some comments on more general functional forms are provided.