The origins of EU legislation: agenda-setting, intra-institutional decision-making or interinstitutional negotiations?

Abstract

This article traces the origins of European legislation during the legislative policy-making process. It identifies three phases where parts of the text of legislative acts can be developed: (1) agenda-setting; (2) intra-institutional decision-making and (3) interinstitutional negotiations, depending on whether the content of the legislation originates respectively in the Commission proposal, the co-legislators’ positions or trilogue negotiations. Using a newly developed text-mining technique which computes in which phase each word of a legislative act originally appears, the article examines the relative importance of each phase and explores how it is affected by interinstitutional conflict. Applying this method to 219 legislative acts adopted between 2012 and 2018, it finds that most EU legislation originates in the agenda-setting phase, and that the new content developed during trilogue negotiations is limited. However, the importance of the agenda-setting phase decreases in cases with high levels of interinstitutional conflict.

Keywords:

• European Union
• interinstitutional negotiations
• ordinary legislative procedure
• text-mining
• trilogues

Acknowledgements

The authors thank the participants of panel ‘The EU Institutions and Legislative Processes’ at the 2019 EUSA Conference and the anonymous referees of this article for their valuable comments. This research has also benefitted from the input provided by the Statistical Methodology and Computing Service (SMCS) and the Louvain Institute of Data Analysis and Modeling in economics and statistics (LIDAM) at UCLouvain.

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Although the precise composition of the delegations varies across policy fields, the Council is usually represented in trilogues by the member state holding the rotating presidency; the EP by a negotiating team composed of the rapporteur, the shadow rapporteurs and the committee chair; and the Commission by high-ranking officials and/or the Commissioner in charge.

2 Our text-mining approach is based on existing techniques, relying on counting the modification of words between different versions of a text. These techniques have proved to have ‘real-world meaning’ (e.g. Grimmer and Stewart 2013; Hollibaugh 2019). This particularly holds true for legislative texts (Cross and Hermansson 2017; Wallace 2012).

3 Council positions can take different forms. To identify the COREPER position adopted before the first trilogue meeting, we consulted various Council documents, such as the COREPER analysis of the final compromise, which gives a brief overview of the negotiations and mentions the reference or date of adoption of the Council position for the first trilogue. In the few cases where there is no reference to a COREPER position, we use the last general approach adopted by the Council before the first trilogue.

4 We use the entire text of the legislative act, including both the recitals and the actual articles. We consider recitals as fully part of the legislative act, because they contain normative prescriptions, aiming thus at transmitting a message (den Heijer et al. 2019) and being used for the interpretation of the articles (Humphreys et al. 2015). Moreover, recitals are not considered trivial by the co-legislators, as they can be subject to hard negotiations.

5 Importantly, while DocuToads is also based on text-mining, it differs from the method we use for measuring the dependent variable. There are two main differences. First, our method and DocuToads do not compare the same texts. Whereas the measurement of the dependent variable relies on comparisons between all the positions of the institutions and the final text of the legislation, the measurement of the independent variable is based on pairwise comparisons of the initial positions of the three institutions and does not use the final version of the legislative act. Second, whereas our method only calculates the number of words that are added in each phase using multiple versions of the text, the DocuToads algorithm calculates the number of additions, deletions and modifications between only two texts. As a result, the measurement of the dependent and the independent variable are not linked and we measure different phenomena.

6 We also conducted dirichlet regressions as a robustness check (Maier 2014). Dirichlet regressions take into account the compositional nature of our dependent variables (that is, the fact that for all cases the sum of the three proportions equal a constant (here 1)) but their results are harder to interpret and must therefore be complemented with graphical analysis of the fitted values (Douma and Weedon 2019). Overall the results of the dirichlet regression confirm those of the fractional logit regressions. The two sets of independent variables have an effect on the proportion of words that appears during the two first phases. The only difference is that the type of act seems to have more effect than the policy area. The results of the dirichlet regressions as well as the graphical analysis are available in the Online Appendix.

Additional information

Notes on contributors

Thomas Laloux

Thomas Laloux is a FRS-FNRS research fellow and PhD candidate at the University of Louvain (Louvain-la-Neuve), Belgium. His main research interest lies in EU legislative decision-making and focuses more particularly on inter-institutional negotiations. His work has appeared in Journal of European Public Policy, Journal of Common Market Studies, European Political Science and Politique Européenne. [[email protected]]

Tom Delreux

Tom Delreux is Professor of Political Science and EU Politics at the University of Louvain (Louvain-la-Neuve). His research interests include the EU’s external relations, EU institutions, and EU and international environmental politics. [[email protected]]