Random Disturbance of Unit Record Weights to Confidentialise Tabular Data
by
Frances Krsinich
Statistics New Zealand

The Statistics Act, Section 37(4), specifies that statistical information published by Statistics New Zealand shall be arranged such that no individual response is identifiable.

Common methods used to confidentialise tabulated data are random rounding (for census data) and dominance rules such as the (n,k) rule (for business survey data). Both of these methods have limitations.

Random rounding can result in non-additive tables. Dominance rules identify sensitive cells simply enough and these can be suppressed. But there is a further requirement to protect these sensitive cells from being derived from marginal totals - this is called 'secondary suppression'. Performed manually, secondary suppression is non-optimal and time consuming, while specialised software is required for its automation.

Both random rounding and dominance rules are applied to the table, rather than the underlying unit record data, which may result in some disclosure risk if many related, and independently confidentialised, tables are combined.

Unit record weights are used to adjust for the probability of sample selection so that the sample is representative of the population. Census records can be thought of as having weights of 1.

Random disturbance of unit record weights is a new idea motivated by one of Statistics New Zealand's clients. The US Census Bureau has recently been exploring an equivalent method for business survey data. The method looks promising, and may solve the problems experienced with current confidentiality methods.

Date received: August 17, 2000

Copyright © 2000 by the author(s). The author(s) of this document and the organizers of the conference have granted their consent to include this abstract in Atlas Conferences Inc. Document # cadt-25.