Using microdata noise to protect tabulations
by
Frances Krsinich
Statistics New Zealand

The Statistics Act of 1975 requires Statistics New Zealand (SNZ) to ensure that individual reponses cannot be derived from published data. Cell suppression is the predominant confidentiality method used to achieve this for business survey data, both in New Zealand and overseas.

Another approach, currently being considered by the US Census Bureau and SNZ, involves the addition of random unbiased noise to microdata. All tabulations can then be produced from this perturbed microdata, without any further confidentialising required. This method could replace cell suppression as a confidentialising method for business survey data, and initial results indicate that the resulting information loss is less than that from cell suppression. A consequence of the method is that non-sensitive cells will have a small amount of noise added, while sensitive cells receive larger amounts of noise. All cells with greater than a given amount of noise can be 'flagged' to the user as unreliable if necessary. The big advantage over the cell suppression method is that only the sensitive cells get perturbed significantly, whereas cell suppression can result in the loss of non-sensitive cells through secondary suppressions which are necessary to protect against derivations of suppressed sensitive cells from marginal totals.

Date received: September 13, 2001