Économie et Statistique n° 483-484-485The overhauled Census: progress in methodology and contribution to knowledge
2011 census and European regulations: imputation procedure specific to three variables
The changes introduced to the questionnaire for the annual census surveys from 2011 gave rise to a considerable amount of complex methodological work, such as that associated with the massive imputation of three variables in the 2011 census: previous place of residence indicator, period in which construction of the building was completed and activity within the meaning of the ILO. For the first two, the context is that of a change in the information collected from 2011 onwards. Each time, two methods are proposed. The first is a fairly simple approach using a model based on the assumption of uniform distribution. The second is more complicated and uses a calibration per municipality or grouping of municipalities and takes data collected in annual surveys using the old classification, in order to find an average for estimated structures with data collected in annual surveys using the new classification. The imputation of ILO activity (three modalities: occupied active / unemployed / inactive) in metropolitan France and the overseas départements (DOMs) is related to the construction of a completely new variable from individual auxiliary variables correlated to the activity and available both in the census and the Labour Force Survey. To do this, despite a difference in collection modes, which created a heterogeneity that could be corrected in part, we used a “spontaneous” declaration of activity variable, which was available in both sources. This provides a fairly satisfactory link with the ILO activity. First, we calculated the probabilities of moving from a spontaneous activity modality to an ILO activity modality. These probabilities were then adjusted to obtain the average for numbers in the different modalities of ILO activity given in the Labour Force Survey in each region/sex cross. In particular, ILO unemployment rates by region/ sex must be subjected to minimal disruption after imputation.