Fine or rare characteristics of the population? the Finally in deletion rate can be much higher. In addition? the presence of “whitened” boxes at the QPV level prevents the user from aggregating data at higher geographical levels? for example all the QPVs of the same Public Intermunicipal Cooperation Establishment (EPCI)? which is the administrative level of reference for the implementation of urban policy.
The usual whitening method raises several Finally in technical difficulties. In particular? the detection of the boxes to be whitened under secondary secrecy and geographical differentiation can prove particularly complex? especially when the dissemination is carried out on several geographical levels which are not nested within each other? Finally in such as IRIS? QPV or 200 m squares.
“Disruptive” methods
A different way of addressing the protection-utility trade-off
In order to overcome these limits? INSEE is netherlands phone number library now using a new family of methods for protecting statistical confidentiality? known as “disruptive” methods: instead of whitening certain cells? these methods slightly “noise” the original data? i.e. slightly modify the value of some of them (or even all of them) with a (small) random disturbance.
Like whitening methods? the choice of a disruptive a thread that connects to method results from a trade-off between data protection and utility: but whereas with whitening methods the result is binary (keep the exact value or whiten)? disruptive methods allow to refine this trade-off by degrading more or less the information contained in individual data. This degradation must be sufficient to guarantee the protection of the data while maximizing their utility for the public (minimizing the loss of original information).
Note that an important advantage of
Perturbative methods that modify all the cells of a table is the protection against the risk of geographical differentiation: since all the cells are (slightly) noisy? it is no longer possible in practice to determine aero leads with certainty the characteristics of individuals who would be at the intersection between several independently disseminated zonings. For example? if all the statistics are disseminated to ±3 units? the risk that the intersection of two (or more) zonings reveals certain information on fewer than 5 people becomes low.
