We use data science to find predictors for court decisions
Why does the EVICT project use a data-driven methodology to study legal data?
It would be impossible to manually analyse the thousands of court judgments. Therefore, the EVICT project adopts a data-driven approach that is unique in the legal discipline.
The project aims to contribute to the development of legal research methodology by complementing traditional legal research techniques with data science techniques.
Using citation network analysis, we aim to conceptualise the right to housing as a network of connected international housing rights and to conduct the first systematic empirical exploration of the impact of this right in datasets of case law from both national supreme courts as lower level courts.
With the help of machine learning techniques, EVICT will identify predictors for court decisions and will explain how these predictors (may) mirror the international right to housing. Combining traditional legal methods with data science techniques could – if successful – be used in future research projects to analyse data in other areas of the law.
Does the EVICT project use traditional legal research techniques?
Yes, it does. The EVICT project combines doctrinal and normative research techniques with methods from data science. The key characteristics of doctrinal legal analysis are that the legal texts are manually collected, read, summarised, commented upon, and placed in the context of the overall legal system.
Moreover, we will conduct an empirical-normative analysis and study the various arguments in case law on, for example, the role of the right to housing, used by the parties and courts in case law. These arguments are related to, for example, theories on the public/private law divide, the hegemony of property rights, and the relationship between national sovereignty and international law.
How does the EVICT project use citation network analysis to study legal data?
EVICT will apply citation network analysis techniques to conceptualise the international right to housing and to determine the impact of this right on national case law.
Legal texts, such as case law, can be conceptualised as a network because of their referential character.
The EVICT project identifies whether and how
1) international courts and committees refer to other judgments/decisions; and
2) national courts refer to judgments of international and national courts, and to committees.
Legal texts such as judgments are seen as the nodes in the network, and references between the cases as the edges. The research team collects citation data and calculates relevant network statistics.
EVICT combines these data with the information obtained by doctrinal analysis, empirical-normative analysis and a Systematic Content Analysis to make sure that the network also contains contextual information on the judgements.
How does the EVICT project use machine learning to study legal data?
The EVICT project will use Machine Learning (ML) techniques to analyse case law of lower level courts. Building on earlier research of our team, we will build a system that predicts whether or not the court will decide that the persons should leave their home.
If the model predicts the results adequately, we will subsequently analyse which words, sentences or facts make the most impact, and thus identify what are predictors for the court’s decision.
For this task, EVICT will use supervised ML, and use ML algorithms such as Support Vector Machine and Neural Network approaches. Such an approach requires a substantial amount of data. Adopting this approach in EVICT is possible, because the EVICT project has developed a unique, manually annotated dataset consisting of over 1100 eviction judgments of Dutch district courts and courts of appeal.
The EVICT project also aims to explain, for example, the predictors identified by ML. Therefore, the data-driven approach is combined with an empirical-normative analysis. The data will provide insights into the use of normative arguments in case law, which are related to the theories discussed above. This method is comparative in nature: the analysis shows how solutions adopted elsewhere function and the weight attached to normative arguments.
EVICT project adopts a unique data-driven approach. We combine traditional legal methods with data science techniques, such as network analysis and machine learning, to find and explain predictors for court decisions.
Publications on legal research methods and data science
Medvedeva, M., Xu, X., Wieling, M., & Vols, M. (2020). JURI SAYS: An Automatic Judgement Prediction System for the European Court of Human Rights. In S. Villata, J. Harašta, & P. Křemen (Eds.), Legal Knowledge and Information Systems: JURIX 2020: The Thirty-third Annual Conference, Brno, Czech Republic, December 9–11, 2020 (pp. 277-280). IOS Press.
Medvedeva, M., Vols, M., & Wieling, M. (2020). Using machine learning to predict decisions of the European Court of Human Rights. Artificial Intelligence and Law, 28(2), 237-266. https://doi.org/10.1007/s10506-019-09255-y
Vols, M. (2020), Juridisch onderzoek, The Hague: Boom Juridisch 2020.
Bruijn, L. M., Vols, M., & Brouwer, J. G. (2018). Home closure as a weapon in the Dutch war on drugs: Does judicial review function as a safety net? International Journal of Drug Policy, 51, 137-147. https://doi.org/10.1016/j.drugpo.2017.08.003
Vols, M. & Jacobs, J.P.A.M (2017), Juristen als rekenmeesters: over de kwantitatieve analyse van jurisprudentie, in P.A.J. van den Berg & G. Molier (Eds.), In dienst van het recht, Den Haag: Boom Juridische uitgevers.
Vols, M., Tassenaar, P.G. & Jacobs, J.P.A.M. (2015), Dutch Courts and Housing Related Anti-social Behaviour. A first statistical analysis of legal protection against eviction, International Journal of Law in the Built Environment, 7, pp. 148-161.
Stay in touch!
Project reference: 949316
Programme type: ERC Starting Grant
Principal investigator: Michel Vols
Host Institution: University of Groningen, the Netherlands
Project duration: 60 months
This project has received funding from the European Union’s ERC Research Grant under grant agreement No 949316