When is administrative data good enough to replace statistical information? A quality indicator based on census comparison

Abstract

Statistics Portugal is considering the use of administrative data in the 2021 Census. To face this challenge, the quality of the available administrative data is measured comparing administrative data with census microdata. The goal is to evaluate the risks of replacing part of the census collected information with information obtained from administrative sources. Record linkage methods were applied and fifteen variables from seven administrative datasets (namely Social Security or Students register) were selected based on the potential for replace census collected information. For each matched record pair, information from corresponding administrative variables is compared, producing an equality rate estimate. The results show very high equality rates when comparing information from each matched pair of records to both geographical and demographic variables (municipality of residence, sex, date of birth, legal marital status, country of birth, country of citizenship). When comparing socioeconomic variables, results are less homogeneous (nevertheless, data obtained by sources related with labour force, also got high correspondence rates for compared record pairs). Considering that some statistics might be obtained by other sources, some Census microdata (regarding economic and educational characteristics of the population) were compared with data from the national Labour Force Survey. These results converge to the general comparison results of this exercise. Finally, the results of the Post Enumeration Survey of the 2011 Census were used to verify the reliability of the comparison results.