Abstract
This article examines, using automated text analyses, the EU politicisation in the media of six Eurozone countries (Belgium, Germany, Greece, Ireland, Portugal and Spain), between 2002 and 2017. By contrasting creditor and debtor countries, the article analyses how the Eurozone crisis affected the politicisation of the EU and its institutions using a unique dataset of 165,341 articles from 12 newspapers. The results show that the Eurozone crisis increased the politicisation of the EU, particularly in the countries that were at the forefront of the Eurozone bailouts. Importantly, the crisis contributed as well to a more multifaceted news coverage of the European Union, namely with a greater emphasis given to supranational institutions vis-à-vis intergovernmental ones. Yet, this supranational coverage was associated with the increasingly negative tone of articles. To that extent, this study shows that greater mention of EU institutions may not necessarily contribute to a Europeanisation of public debates.
Supplemental data for this article can be accessed online at: https://doi.org/10.1080/01402382.2021.1910778 .
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 See Table 1, in the Online appendix, for a list of the elections and newspapers analysed for each country.
2 Only news items or opinion articles from the main sections of the newspapers were collected. Thus, pieces published in entertainment, culture, sports sections, etc., were excluded.
3 Due to the use of OCR, some misspelling errors were occasionally produced. These errors were investigated by counting the percentage of unrecognized words in the articles, using a well-document/popular and free open-source spell-checking algorithm named UNSPELL for the automatic spellchecking. http://hunspell.github.io/. This process was implemented on a per language basis. Table 2 in the Online appendix presents descriptive statistics of the percentage of unrecognized words in each article’s title and body. As we can see, the results were very low with a median of 5% of unrecognized words per article. The distributions of these percentages are identical for all newspapers and years. Note that an unrecognized word is not necessarily a misspelled word produced by the ORC extraction process. It can very well be that the word does not appear in the dictionary (e.g. abbreviations) or that it was already misspelled in the original document. In the end, a small number of articles, with more than 35% of unrecognized words in their body, were dropped from the dataset.
4 Table 3, in the Online appendix has the list of terms used.
5 See Figure 9 in the Online appendix for a comparison of the salience results using more exclusive rules.
6 Valence shifters are words in the text that alter or intensify the meaning of polarized words. They include negators (e.g. ‘This movie is good’ vs ‘This movie is not good’) and amplifiers (e.g. ‘This movie is good’ vs ‘This movie is very good’).
7 Popular dictionaries include general purpose ones like the Bing Lexicon (Hu and Liu Citation2004), the NRC Emotion Lexicon (Mohammed and Turney Citation2010), or the AFINN dictionary (Nielsen Citation2011). More case-specific dictionaries for political science include the LIWC lexicon (Tausczik and Pennebaker Citation2010) and the LSD (by lexicoder, tested in Young and Soroka Citation2012).
8 We do not claim that sentiment analysis based solely on word-by-word lexicon lookups or even including valence shifters is state-of-the-art. We also know that this approach cannot really compete with dedicated sentiment analysis techniques from the machine learning literature. However, for this particular use case of detecting broader latent patterns of sentiment in a large selection of texts, our goal is to build a reliable tool.
9 The results do not noticeably change using exclusively the score of the article’s body (see Figure 7 in the Online appendix).
10 We summarize in Figure 8 (in the Online appendix) additional validation steps implemented for our measures of EU salience and contestation.
11 Table 6 (Online appendix) has the values for Salience, Contestation, Percentage of Neutral Articles and Tone for each election.
12 See Table 5 in the Online appendix.
13 See Figure 10, in the Online appendix.
Additional information
Funding
Funding
Notes on contributors
Tiago Silva
Tiago Silva is a post-doctoral researcher for the ERC MAPLE project at the Institute of Social Sciences, University of Lisbon. He holds a PhD in Political and Social Sciences from the European University Institute. His main research interests are political communication, online campaigning and European integration. [[email protected]]
Yani Kartalis
Yani Kartalis is a PhD candidate in Comparative Politics at the Institute of Social Sciences, University of Lisbon. He earned his MRes in Political Science at the Pompeu Fabra University in Barcelona. His research interests lie in the fields of political parties, political sociology and psychology, and automated data collection approaches for the social sciences. [[email protected]]
Marina Costa Lobo
Marina Costa Lobo (D.Phil, Oxford, Habilitation, Lisbon 2011) is Principal Researcher at the Institute of Social Sciences, University of Lisbon and vice-director of Instituto de Políticas Públicas (IPP). Currently, she is PI of the ERC Consolidator Project ‘MAPLE’, which researches the politicisation of Europe before and after the Eurozone crisis. She is the Director of the Portuguese Election Study and a member of the CSES Planning Committee. Her research interests include the role of leaders in electoral behaviour, economic voting and the consequences of EU politicisation. In 2021, she guest edited, together with Michael Lewis-Beck, a Special Issue in Electoral Studies on the topic of EU contestation and its effects on political behaviour. [[email protected]]