Mapping the EU Regulatory State: Evolution and Design, 1993–2023

A key challenge in regulatory governance scholarship is the development of content-based measures to analyze regulation systematically. Existing approaches to measuring rules often focus on structural features such as word count and volume but fail to systematically capture the substantive dimensions of regulation—who is being regulated, the level of stringency, the prescribed or constrained actions, and the underlying rationale. In European Union (EU) studies, the lack of such measures limits the ability to examine regulatory designs across diverse policy areas, explain the drivers of regulatory choices, and assess how different configurations of regulatory elements influence transposition dynamics. This study introduces a novel computational approach that integrates Natural Language Processing (NLP) and Large Language Models (LLMs) to systematically extract and categorize key components of regulatory design, specifically actors, stringency, aims, objects, and actions. NLP techniques, including dependency parsing, are used to extract these elements, while LLMs simplify legal text, resolve anaphora, and identify relevant actors. The study then applies the proposed computational approach to the analysis of EU regulatory design across 21 policy areas from 1993 to 2023, covering all three major sources of EU regulation: Directives, Regulations, and Decisions. By doing so, it provides new empirical insights into the evolution of the EU as a regulatory state.