Abstract

INSEE has built a Statistical Register of Individuals and Dwellings (Résil) in order to modernise its information system of demographic and social statistics by making greater use of administrative data. Given the growing importance of administrative data in the statistical production, rationalising the way it is processed is becoming a matter of concern. The data, received as configured in the administrative sources, is then transformed into statistical data, i.e. data organised according to the statistical units (individuals, households, dwellings, etc.) and statistical concepts. There are of course a number of prerequisites for setting up such a system: the need to use different sources in order to aim for exhaustiveness; the need to apply innovative statistical processing methods at scale to link and process multiple data sources while ensuring statistical quality; the need for a legal framework to protect the data processed; and finally, the need for a "social acceptance", i.e. a legitimacy that goes beyond technical or legal capacity.

To build the statistical register of dwellings, Résil uses the tax administration’s files on premises and buildings. Regarding individuals, Résil is first initialized by an administrative register (the national register for the identification of natural persons), which is then supplemented with various administrative data sources, such as tax data, data on social protection benefits, paid employment data and data concerning replacement income. Each data source is used as an indicator of the presence of individuals in the national territory (signs of “life” model) and a localization algorithm is then mobilized to determine the most likely dwelling/address of usual residence of households.

Built from multiple administrative sources integrated at different times, the content of Résil is constantly evolving. Furthermore, the rich but raw information may be inconsistent from one source to another. Therefore, producing a reference dataset requires both freezing the content of the register on an annual basis and applying a series of statistical procedures aimed at reconciling the different sources. The outcome of this process is the reference universe : a « snapshot » dated January 1st of each year aiming to provide exhaustive lists of individuals living on the national territory, habitable dwellings, and households.

After five years of IT development, Résil is now used in production to set up the sampling frame for household surveys, to check the coverage of administrative data sources, and to manufacture enriched files by matching different sources in a simpler and more reliable way, via a non-significant statistical code computed for each individual. The usual resident population counts are still produced yearly from census surveys. Nevertheless, the population coverage of Résil and the census surveys will be assessed by using the dual system estimation method.

A lot of work still needs to be done, especially to improve the model for estimating the signs of “life” and to localize individuals in their usual home, which in turn enable us to deduce the compostion of households. Data from the yearly census surveys will be used as benchmarks to tune these models over time.