The dissemination of statistics at fine and This suppressive multiple geographical levels adds an additional difficulty in preserving statistical confidentiality. In order to always better meet the needs of local and national stakeholders? INSEE disseminates statistical information on a growing set of so-called “infra-municipal” zonings: Îlots regrouped pour l’information statistique (IRIS)? QPV or even 200 m squares .
However? the risk of disclosure is increased
The case of statistical dissemination nepal phone number library on geographically close territories. By “geographic differentiation”? it is indeed possible? from additive indicators This suppressive (for example the number of households below the poverty line)? to deduce the characteristics of the population residing in the territory corresponding to the geographical difference between two close territories.
This is what Figure 1 shows on an example in Compiègne: the priority district La Victoire was slightly extended (blue parts) during the update on January 1? 2024.
The simultaneous dissemination of statistics on the old perimeter (in gray) and the new one (gray and blue part) makes it possible? in the absence of protective measures? to reveal the characteristics of the population in the blue zone by difference.
Figure 1 – Example of risk of geographic differentiation in Compiègne
Sources: IGN (base map);open data(QPV contours).
The solution of hiding boxes reaches its limits
The method traditionally applied at INSEE to how using smartphones in non-traditional protect statistical confidentiality is “suppressive”: the value of the cells in a data table that poses a risk of breach of confidentiality is replaced by a “missing value”? i.e. the initial value is simply deleted from the file (this is referred to as “whitening” the cells). In addition? so that the hidden information cannot be reconstructed? other cascading deletions may then be necessary under the secondary confidentiality or geographical differentiation mentioned above.
To masking a large number of boxes and the aero leads loss of information can then be significant. As an illustration? for the dissemination of job seeker statistics in 2021 at the scale of urban policy districts? the processing of statistical confidentiality required whitening more than 3?000 boxes in the data table out of approximately 70?000 boxes? i.e. an overall loss of information of around 5%.