Eliminating small cells from census counts tables: empirical vs. design transition probabilities

Abstract

The software SAFE has been developed at the State Statistical Institute Berlin-Brandenburg and has been in regular use there for several years now. It involves an algorithm that yields a controlled cell frequency perturbation. When a microdata set has been protected by this method, any table which can be computed on the basis of this microdata set will not contain any small cells, e.g. cells with frequency counts 1 or 2. We compare empirically observed transition probabilities resulting from this pre-tabular method to transition matrices in the context of variants of microdata key based post-tabular random perturbation methods suggested in the literature, e.g. Shlomo, N., Young, C. (2008) and Fraser, B.,Wooton, J. (2006)