Improving the accuracy of annual census surveys using tax sources

The last population census was conducted in 1999. Since 2004, France has renewed its census operation, which is now based on a five-year cycle: each year, only part of the population is enumerated in the annual census survey (ACS). A population census is thus based on the aggregation of data from five ACSs.

The rotation groups, on which the ACSs are built, are sometimes unbalanced in terms of number of dwellings and population. While this does not affect the quality of the population census estimates (since the population census rely on all five rotation groups), in some regions there are significant fluctuations in the population series calculated from a single ACS, which has hitherto hindered the further use of this source.

However, the use of data from a single ACS offers several advantages, including the uniqueness of the observation period and increased timeliness in the event of changes in the questionnaire. For example, a modification of the housing sheet questionnaire allowing for a better identification of the links between persons residing in the dwelling took place in 2018. The results of this new household-family analysis led to the publication of a first study in early 2020 from the EAR, as opposed to 2023 if it had been necessary to wait for the results of the census.

Until the ACS 2018, the weights for these surveys were obtained from the sample weight and a calibration on a limited number of variables. Beginning with the 2019 EAR, a new methodology for calibrating the ACS is being implemented. This methodology aims to improve the accuracy of data estimated from a single ACS. It is based, among other things, on the use of auxiliary variables from the "Fichier démographique des logements et des individus" (Fidéli), which is compiled from tax sources.

The new weighting makes it possible to smooth fluctuations in the population and housing number series due to imbalances in rotation groups. This reflects in particular the gains in precision made possible by the enrichment of the auxiliary variables used in the adjustment. Like the weighting used up to ACS 2018, this new weighting does not aim at correcting the validity date of estimates in municipalities with 10,000 or more inhabitants, which remains linked to the reference date of the sampling frame used for the selection of the sample of addresses to be surveyed (July of the year preceding the ACS).