Replication Data for: Identifying Delegation and Constraints in Legislative Texts: A Computational Method Applied to the European Union. European Union Politics

The article introduces a computational method for identifying delegating and constraining provisions in European Union laws. It is applied to more than six hundred thousand legal sentences, extracted from laws that have been adopted between 1958 and 2019. The dataset includes 1) Classification of the sentences of the 9,319 secondary measures (dataset_disaggregated.dta), 2) Number of provisions assigned to the thirteen classes in 9,319 secondary measures (directives and regulations), and several ratio measures (dataset.dta) 3) Classification of 3,000 sentences of a) the rule-based application, b) the authors, and c) the two annotators (annotations.dta) 4) Annotations of the authors and the two coders (intercoder_reliability.dta) and 5) Ratio measures from the manually annotated datasets of Franchino (2007) and Migliorati (2021) (manual_coded_MF.dta). The dataset also includes the files replicating the results in the Article, Tables A5 and A6, the comparison with manually annotated datasets, the intercoder reliability test, and Figures A1 and A2 are in the supplemental material folder. For the files replicating the machine learning analyses reported in Tables A2, A3, and A7 to A10, please see the Data Availability Statement and the Supplementary material of the article on the European Union Politics website. For the code and materials used to preprocess the EurLex corpus, train the spaCy NER models, and run the transformer fine-tuning experiments, please visit the corresponding GitHub repository: https://github.com/giovapagano/eu_delegation_constraints. The fine-tuned transformer models introduced in the paper are publicly available on the Hugging Face Model Hub: https://huggingface.co/giovannip