Applying scoring method on national surveys for local estimations: an example with the ICT-Household Survey

In national surveys, the size of the regional sample is often too small to directly produce reliable sub-regional estimates. However, small domain estimation methods can fix this problem. Among them, the scoring method gathers all the individuals in the national sample to estimate, thanks to an econometric model, a relationship between a given phenomenon and a set of socio-demographic explanatory variables (sex, age, occupational situation, etc.). From this estimation, individual probabilities to be faced with this phenomena are calculated.
Then, these probabilities are projected on the population of a region, thanks to an external source, such as the census. It is only possible if two important hypotheses are respected. On the one hand, the available individual data in the survey and external source must correctly grasp the analysed phenomenon causes. On the other hand, the analysed phenomena must be relatively similar between the region of interest and the whole national sample.
The purpose of this working paper is to give a concrete example by using the 2019 ICT-Household Survey. In particular, it was tested and applied on the illectronism rate in 12 regions in France.