Abstract
A body of law is an example of a dynamic corpus of text documents that are jointly maintained by a group of editors who compete and collaborate in complex constellations. Our goal is to develop predictive models for this process, thereby shedding light on the competitive dynamics of parliamentarians who make laws. For this purpose, we curated a dataset of 450,000 legislative edits introduced by European parliamentarians over the last ten years. An edit modifies the status quo of a law, and could be in competition with another edit if it modifies the same part of that law. We propose a model for predicting the success of such edits, in the face of both the inertia of the status quo and the competition between overlapping edits. The parameters of this model can be interpreted in terms of the influence of parliamentarians and of the controversy of laws.
Introduction
The legislative process is often perceived as a "black box" of complex negotiations, lobbying, and overlapping interests. While open-government initiatives have increased data transparency, robust quantitative tools to analyze the competitive dynamics of law-making remain scarce. When multiple Members of the European Parliament (MEPs) propose conflicting changes to the same sentence of a bill, what determines the outcome? Is it the political weight of the author, the inherent controversy of the law, or the procedural role of the individual leading the debate?
This research introduces the "War of Words" (WoW) framework to model these dynamics, treating the legislative process as a game played on a text. By quantifying the "influence" of politicians versus the "inertia" of controversial laws, we can better understand the mechanics of democratic decision-making.
The War of Words Dataset
To support this modeling, we curated and released the War of Words dataset, a massive collection of legislative edits from the European Parliament.
- Scope: 10 years of data (2009–2019) covering two legislature periods (EP7 and EP8).
- Scale: 449,493 individual edits proposed by 1,214 MEPs across 1,889 dossiers.
- Granularity: Unlike previous datasets focusing on broad amendments, this work breaks changes down into word-level insertions and deletions. It identifies over 96,000 specific conflicts where edits overlap, creating a unique map of legislative friction.
Modeling Legislative Competition
The WoW model employs two primary components to analyze the data:
1. The Edit Graph
Proposed changes are represented as nodes in a graph. If two edits modify the same part of a law, they are connected by an edge, representing a conflict. This structure allows for the identification of "cliques" of mutually exclusive proposals. Technical regulations often show isolated nodes, while high-stakes laws result in dense "hairballs" of conflicting edits.
2. The WoW Model
We adapted the Bradley-Terry model (common in sports rankings) and the Rasch model (used in psychometrics) to calculate the probability of an edit's acceptance based on three latent parameters:
- Skill ($s_u$): The latent influence or "strength" of an MEP.
- Inertia ($d_i$): A law's resistance to change, serving as a proxy for controversy.
- Rapporteur Advantage ($r$): A specific boost for the MEP appointed to lead the report on a dossier.
Key Insights and Results
The WoW model significantly outperforms random baselines, achieving a cross-entropy loss of 0.726 (EP8) compared to 0.942 for a random predictor. Several key findings emerged:
- The Power of the Status Quo: Most edits do not just compete with other MEPs; they face the "inertia" of the original text. The status quo is a formidable opponent that often prevails when MEPs are divided.
- Rapporteur Influence: The data confirms a measurable "home-field" advantage for rapporteurs ($r \approx 1.18$ to $1.31$), likely due to their control over the narrative and final report drafting. This advantage appeared to increase in the 8th legislature (2014-2019).
- Inertia and Lobbying: High inertia parameters correlate strongly with dossiers that faced intense public protest or corporate lobbying, such as "Copyright in the Digital Single Market."
Conclusion
By quantifying the latent variables of the legislative process, the WoW model provides a new lens through which to view democratic governance. While currently text-agnostic, future iterations will incorporate Natural Language Processing (NLP) to understand the content of edits and model how influence varies across different thematic committees. This work paves the way for more transparent and predictable legislative analysis.
Summary updated