Kernel density smoothing as a means for the construction of anonymized regional maps

Abstract

The smearing effect of kernel estimates of the local density, local proportions and local means is used as a means for the construction of anonymized maps. The standard anonymization criteria were derived for the display of case numbers of a predefined area system. However, for kernel estimates there does not exist such a defined area system. We discuss the resulting difficulties of the application of these criteria for kernel estimates. Besides, there are some de-anonymization risks which are specific for kernel estimates. We discuss these topics for data from 1.9 million Berlin taxpayers with known exact address and taxable income. In the conclusions we vote for a much stronger emphasis on the output format of a map and the labelling of the displayed values in the map.