Statistical framework for fully register based population counts

Abstract

The increasing availability of registers or administrative archives has been a strong push towards moving from traditional censuses to combined censuses or even completely register based censuses. In this context, a statistical framework needs to be designed in order to delineate all the statistical issues of the new estimation process. To this aim, a population frame needs to be defined for both surveying and estimation phases. Sampling surveys should be designed for quality assessment and for improving the quality of the register based estimation process. Drawing on similar experiences, a formalisation of the population size estimation process fully based on administrative data is presented. An application to Italian estimation process is reported.

Machine learning estimation of the resident population

“…At the centre of official statistics is the description of the population of a country or region and its characteristics, at different points in time. The National Statistics providers build these statistics using (i) administrative data, including as a main source the population registers [1], or (ii) more complex estimation methods when the latter are not available [2]. The present paper concerns the former case, which is prevalent in a growing number of countries such as all the Nordic countries as well as the Netherlands, Austria, Israel, Japan and the Baltic countries.…”

“…We mention here methods based on index-theory [3], defining individual scores as functions of multiple individual characteristics [4] or a cumulative link model of ordinal response used by Statistics Iceland for the Census in 2011. 1 We have already presented in several instances the approach adopted by Statistics Iceland in recent years [5][6][7]. It relies on formulating the over-counting as a binary classification problem (individuals are present in/absent from the country), which therefore has multiple solutions but also an optimum one which can be systematically defined, depending on the structure of the data and on the goal of the analysis.…”