Paradata analysis of the 2010 Population Census: investigation of factors associated with nonsampling errors in the data collection stage

Abstract

The relevance of a population census for a national statistical system is undeniable for its thematic and territorial coverage. Nonetheless, the complexity and size of a census operation lead to challenges for ensuring timeliness and quality of the results. This paper presents potential factors associated with non sampling errors detected in the data collection stage based on the analysis of Brazilian 2010 Population Census microdata and paradata. Data obtained from the field work monitoring system, called paradata, is used to provide information about divergences observed between data collected by enumerators and supervisors, also it is used the census microdata. The latter carried out follow-up interviews in households selected by the supervision/monitoring plan. Human resources databases containing socio-demographic information of enumerators and supervisors is also brought to enhance the analysis. The statistical modeling utilized is generalized hierarchical models, in which the response variable is defined as the occurrence of a discrepancy (or divergence) between the information collected by enumerators and their supervisors. The results indicate that the different hierarchical levels investigated are relevant to decompose data variability and hence have to be considered in the analysis. However, respondents’ characteristics have markedly more influence on the chances of a divergence than those of enumerators’ and supervisors’. In addition, there is evidence that respondents who are male, illiterate (or with low educational level), older and living in households with indicators reflecting poor life conditions present higher odds in favor of the occurrence of divergences on data collected by enumerator and supervisor.