Benchmarking the Effect of Cell Adjustment on Tabular Outputs: The Shortcomings of Current Approaches

Paul Williamson

Abstract:
Current assessments of the effect of disclosure control lack rigour and are too far removed from the real implications for user analyses. This article uses a case study to illustrate these shortcomings and suggest a remedy. The proposed solution has three main components. First, the use of a set of benchmark counts providing representative population and geographical coverage. Second, consideration of the effect of cell adjustment not only upon counts, but also upon rates derived from them. And third, the use of more transparent statistical summary measures, such as indicative confidence intervals. The benchmark counts and benchmarking software used in the writing of this article have been placed in the public domain, in order to encourage the wider adoption of the proposed evaluation strategy. The general approach outlined, however, could be pursued independently of the use of either.

Keywords:
Statistical disclosure control, random rounding, small cell adjustment, goodness of fit, fitness for purpose