Credit: STATEC

Luxembourg's statistical institute STATEC has published the preliminary results of the project "Preferences through Twitter", which provides insight into how people's feelings have changed during the COVID-19 pandemic.

Led by STATEC Research, "Preferences through Twitter" provides a sentiment analysis of tweets reveals to reveal how people in Luxembourg fared during the COVID-19 crisis. As part of this project, STATEC Research collaborated with researchers from the University of Johannesburg (UJ) and Auckland University of Technology (AUT), with funding from the National Research Fund (Fonds National de la Recherche - FNR), South Africa's University Research Co. (URC) and AUT, to compute a real-time measure of wellbeing - the Gross National Happiness (GNH) - in Luxembourg and in a selected group of European countries during the pandemic. 

Well-being data are usually collected via large-scale surveys that take time to administer and are thus only available after some delay. The Eurobarometer indicated that life satisfaction, a valid and reliable measure of well-being, decreased by 8% between Autumn 2019 and Summer 2020. According to STATEC, survey data alone can not explain this decrease.

Consequently, researchers turned to Twitter data. Every day, a large number of people around the world share their opinions and feelings via tweets. In Luxembourg, people share approximately 500 tweets per day, which is nearly 3 500 tweets per week. This wealth of short texts provides a real-time source of information that can be transformed into usable data using sentiment analysis (an automated process that uses natural language processing to determine the feelings and attitudes of a written text’s author). 

The GNH in Luxembourg increased from the end of June to mid-July, perhaps because COVID-19 and the lockdown seemed to be in the past. However, as people became aware of a resurgence in new cases, their well-being declined suddenly and continued to do so. In fact, the changes in GNH are consistent with the changes in the number of new positive cases, albeit in the opposite direction. There was a particularly sharp decline in GNH paired with a large increase of new positive cases in the month of October 2020.

The project also found that people in Luxembourg generally complied with containment measures, as shown by the close relationship between the changes in people’s distancing behaviour and the changes in containment measures. The correlation between the two measures was 80%. For the sake of comparison, the correlation was 71% in France, 68% in both Belgium and the United Kingdom, 55% in Germany and 64% in Spain. The correlation in Luxembourg was second only to Italy (86%) among the sample countries. 

Another advantage of sentiment analysis applied to Twitter data is that it allows researchers to compute GNH for any country in the world (with sufficient Twitter users). A comparison of Luxembourg with six other European countries that were severely affected by COVID-19 last spring (Italy, Spain, UK, France, Belgium and Germany) revealed that, since July 2020, the Grand Duchy has experienced greater changes than most of the other EU countries considered, as well as a greater decline in well-being: by the end of October 2020, Luxembourg had the lowest GNH score. Well-being declined in each country as we moved into autumn, but the changes in Italy, Germany, the UK and Belgium were not as severe as those in Luxembourg, France and Spain.

Whilst it is still too early to determine what caused these changes, the findings suggest that the country-specific mix of containment policies, contagion rates and economic performance may contribute to answering these questions.

At present, the research team is focussed on producing the GNH score for each day since 1 January 2020 for Luxembourg and the other six European countries. Additionally, the team is working to widen the number of variables derived from Twitter to monitor the changes in economic insecurity, trust in others and institutions and feelings towards immigrants. Above all, the team is evaluating whether this data can be considered as reliable and accurate. 

The researchers have gathered two pieces of evidence which support the validity of the GNH as a measure of well-being. First, the average GNH score correlates strongly with average life satisfaction as measured by the Eurobarometer. Second, GNH correlates meaningfully with a measure of negative emotions over time. 

These figures will be extended to cover the whole of 2020, enriched with additional explanatory variables and further tested for validity and reliability. Such information is expected to supplement the evidence based on survey data to provide invaluable insight for the general public, research community and policy makers.