Confidentiality control of tabulated data: Some practical network models

The problem of confidentiality control is a significant issue for statistics gathering organisations. Most existing methods suffer from a variety of difficulties for practical implementation. This paper presents techniques for confidentiality control of tabulated data using network formulations to overcome many of the weaknesses of the current methods. These techniques are based on the established methodology of controlled rounding, which preserves additivity in the resulting solution, while adding a number of pieces of new material. Firstly, a controlled rounding formulation is introduced that is designed to provide an unbiased solution. Then we construct three variations of a controlled perturbation technique using different forms of bounds on the cells. Finally, a methodology is explained for disturbing only the small-valued cells of the table using each of the four basic methods. A simple example of the methods and results from experimentation on the use of the methods show that they satisfy the needs of confidentiality and are computationally viable.