Abstract:
|
When survey data are modified for disclosure limitation, the results of statistical analyses may be different from those that would be obtained based on the original data. One of the methods frequently applied to categorical variables is random swapping method and its variations. When categorical variables are swapped the relationships between different variables in the data can be considerably affected and the correlation structure damaged. It is especially problematic when these variables are design variables or the variables defining important subgroups in the population that are of a particular interest to the data analyst. In this paper we present a new disclosure limitation method which is similar to swapping; however, the important difference is that it takes into account other variables, continuous and categorical, thus reducing the impact caused by disclosure limitation on the overall data and within important subgroups.
|