ESRA 2027 sessions by theme

Back to Overview of Sessions

Session Details

The rapid expansion of non-traditional and newly emerging data sources has prompted National Statistical Institutes (NSIs) to rethink how official statistics are produced. Many NSIs now operate integrated statistical systems that combine information from multiple sources, including surveys, administrative registers, web-scraped data, and digital trace data. By leveraging the complementary strengths of these sources, such systems enable the production of official statistics that are more timely, granular, and comprehensive.

At the same time, integrating multiple data sources has fundamentally changed the challenges of statistical inference by requiring explicit evaluation of the quality of each contributing source. In particular, it has renewed attention to representation errors arising from nonresponse in probability (i.e., randomly selected) samples and selection bias in non-probability samples. It also provides opportunities to identify and model measurement error that would otherwise remain unobserved, for example through latent variable models such as factor analysis, latent class analysis, and hidden Markov models. Furthermore, data integration introduces additional sources of uncertainty, including linkage errors, which can exacerbate both representation and measurement errors. Together, these challenges underscore the need for new inferential frameworks capable of jointly accounting for multiple sources of error in integrated statistical systems.

In this session, we will discuss a variety of statistical methods handling the intricacies of data quality assessment in multi-source data. We will discuss how new frameworks can leverage the strengths of the different data sources and how to properly account for their limitations.