Revisiting interval protection, a.k.a. partial cell suppression, for tabular data

Abstract

Interval protection or partial cell suppression was introduced in “M. Fischetti, J.-J. Salazar, Partial cell suppression: A new methodology for statistical disclosure control, Statistics and Computing, 13, 13–21, 2003” as a “linearization” of the difficult cell suppression problem. Interval protection replaces some cells by intervals containing the original cell value, unlike in cell suppression where the values are suppressed. Although the resulting optimization problem is still huge—as in cell suppression, it is linear, thus allowing the application of efficient procedures. In this work we present preliminary results with a prototype implementation of Benders decomposition for interval protection. Although the above seminal publication about partial cell suppression applied a similar methodology, our approach differs in two aspects: (i) the boundaries of the intervals are completely independent in our implementation, whereas the one of 2003 solved a simpler variant where boundaries must satisfy a certain ratio; (ii) our prototype is applied to a set of seven general and hierarchical tables, whereas only three two-dimensional tables were solved with the implementation of 2003.