European identity over time

*Note that data from East Germany only becomes available in 1990. Data visualised for all of Germany prior to that is exclusively based on West Germany.

The data visualised here represents our novel measure of European identity, utilizing a Bayesian latent trait model. It combines various survey sources from 34 countries over a period of 41 years. 

This approach addresses the issue of inconsistency and variation in measures of European identification, as well as potential biases such as differences in question phrasing and translation. It allows an entirely new longitudinal perspective on European identity. 

The analyses made possible by this new perspective have the potential to be a significant step forward in our understanding of important macro mechanisms such as the relationship between European institutions and collective identities. It lays the groundwork for future research to examine the causality of the institution-identity link.

Data sources:
Eurobarometer, European Social Survey, European Value Study, World Values Study, International Social Survey Program, the EUI-YouGov survey, and IntUne data.

Technical Details

In the development of a measure for European identity, we utilize a Bayesian latent trait model as described by Chris Claassen (2019), subsequently applied in his 2020 study. This model facilitates the estimation of country-year public opinion indices from fragmented cross-national public opinion data. It employs a beta-binomial specification, which has been identified by Claassen as more effective than simpler binomial models for estimating uncertainty in diverse contexts and across a spectrum of opinions. Additionally, the model incorporates item slopes (factor loadings) and item-by-country bias parameters to improve the accuracy of both point estimates and uncertainty measurements.

The process for developing this measure involves a detailed review and selection of survey items from cross-national datasets, ensuring that the data spans multiple years. Due to historical data limitations, particularly with early Eurobarometer questions that require respondents to rank their attachment by geographical region which showed poor performance and potential bias in model priors, these and similar variables collected before 1982 are excluded. The primary data source used is the Eurobarometer; other sources include the European Social Survey (ESS), European Value Study (EVS), World Values Study (WVS), International Social Survey Program (ISSP), the EUI-YouGov survey, and IntUne data.

Variables are selected based on their conceptual alignment with the notions of European identity and the technical capacity of the model. Specifically, we focus on variables that gauge the cognitive and/or affective dimensions of individuals' self-identification with Europe and the EU. Early surveys that only addressed utilitarian support for the EU, such as the USIA surveys, are excluded.

Ultimately, the chosen variables are tested to determine their contribution to the model's latent space. This testing confirms the exclusion of certain questions, such as questions that require respondents to rank their attachment by geographical region, which do not correlate well with the latent space of the Bayesian latent trait model, thus validating their exclusion from the analysis.