Assessing the quality of data on international migration flows in Europe: the case of undercounting

Undercounting is a serious data quality issue that can lead to a directional bias in migration statistics. It may be caused by a lack of legal requirements for reporting migration events between and within countries, or difficulties in enforcing such requirements. The main sources of information on undercounting are the metadata accompanying official statistics and expert opinions. However, metadata related to undercounting are very limited. Similarly, expert opinions may be arbitrary, elicited from few experts who might not know all the details of the migration data shared by different countries, or who might not take into account changes in methodologies or definitions, or retrospective updates of the data after censuses. This paper aims to develop a methodological solution for the assessment of undercounting in international migration data, and has three main objectives. First, the paper provides an overview of the available metadata and expert opinions on undercounting in European migration flows. Second, the study proposes a new, data-driven, year-specific, and duration-of-stay-adjusted approach to classifying undercounting that enables scientists, researchers, policymakers, and other users to combine information from various sources. The proposed methodological solution relies on bilateral migration data provided by Eurostat and the UN, as well as migration data provided directly by some national statistical institutes (NSIs), to compare flows, in the same direction reported by a given country with high-quality data reported by another set of countries. The duration-of-stay correction coefficients are calculated by using an optimization model, or are taken from previous migration models. We construct metadata and expert opinion scores and combine them into a single classification of undercounting. Third, the final outcome is a dynamic classification of undercounting for 32 European countries (2002-2019) that is easily accessible and, flexible, and that allows for changes to the underlying assumption via an online Shiny application. Our findings suggest that the highest level of undercounting in migration data are observed in the new EU member states, particularly in Bulgaria, Latvia, and Romania. However, we also show that for certain periods, there have been notable levels of undercounting in many other European countries, including in those countries that are traditionally assumed to maintain reliable population statistics.