A model based categorisation of the Italian municipalities based on non-response propensity in the 2011 Census

The Italian National Statistical Institute had certified the quality of the 15th Italian population and housing census thanks to a Post Enumeration Survey (PES) taken throughout the months immediately after the Census. The aim of the PES is to produce total estimates adjusted for under coverage and, for the first time, over coverage. The model underlying the under and over coverage estimation, takes into account the differences between individual probabilities of responding to the Census. For this aim a regression unit-level model was applied; in order to study the individual probability to be censused on the basis of which the Hard to Count Index (HTC) of Italian municipalities it was created. In the model were used variables derived from the PES questionnaire and additional area-level variables from other sources. HTC categorises the 8092 Italian municipalities in 3 different levels, partitioning the distribution of municipal non-response propensities, based on percentiles. This paper describes in detail the multilevel logistic regression model used to study non-response probability, the development of the HTC, the methods and the analysis carried out to evaluate the goodness of index, regarding the census coverage.