Data Science Projekt: Analysis of Legal Practice 4010-PDSa-30

The course is a project-based Data Science course in which participants carry out a real-world analytical project — from data acquisition and understanding, through exploration, assessment of data quality and limitations, formulation of a research question and operationalisation, to modelling, evaluation, inference, and communication of results. Classical frameworks for the Data Science process, such as CRISP-DM, serve as a point of reference, but the course extends them to include issues that are particularly important in empirical work with data: the data-generating and selection process, measurement validity, the scope of inference, and analytical reproducibility.
The empirical material is SAOS (the Polish Court Judgment Analysis System) — an open database containing several hundred thousand judgments of Polish courts and providing a public API. It is real-world, semi-structured, and methodologically demanding material, combining rich metadata on court judgments with the unstructured text of written justifications. The primary focus is on generalisable Data Science problems and methods rather than computational law. Judicial decisions serve as empirical material, while the necessary legal context is introduced only to the extent required for conducting the analyses correctly.
Particular attention is devoted to how warranted claims about a phenomenon can be derived from observational data. A large number of observations does not eliminate problems arising from the data-generating process, data incompleteness, or source selectivity. Students therefore learn to identify the population to which the data refer, recognise mechanisms of selection, and determine the scope of conclusions that can legitimately be drawn from the available data.
The course emphasises competencies that are crucial to the quality of real-world analytical projects but are easily overlooked in courses focused primarily on algorithms. These include rigorous exploratory data analysis, diagnosis of data quality and completeness, deliberate operationalisation of constructs, and measurement validation. Unstructured text is treated as a full-fledged source of data: students learn methods of text representation and automated feature extraction, together with the challenge of assessing the validity of variables created using NLP methods or language models.
An important element of the course is the distinction between three analytical objectives: description, prediction, and explanation. Students learn that the choice of model and evaluation criteria should follow from the type of claim they intend to make. In the part of the course devoted to inference, particular emphasis is placed on the limitations of observational data, potential confounding factors, and robustness analysis.
Students work in teams using Python or R. Reproducibility is treated as a default standard: projects should have a clear structure, document the acquisition and processing of data, and allow the main analyses, tables, and visualisations to be reproduced. The semester is structured around two main stages of work: a data and research design report, covering exploratory analysis, data diagnosis, the research question, operationalisation, and the analysis plan; and a final analytical project accompanied by a research paper. Peer feedback between teams forms part of the workshop process but is not treated as a separate graded deliverable. The final outcome is a team research paper prepared to a standard suitable for publication as a preprint or for further development towards submission to a peer-reviewed journal.

Session topics:
1. Introduction to computational law - Lecture by Dr Łukasz Górski; research problems and applications in computational law; the specific characteristics of court judgments as a data source.
2. Data Science as a research process; SAOS as empirical material
Structure of a Data Science project; characteristics of SAOS; research questions that can be addressed using these data; limitations of the source.
3. Data acquisition via API; data structure and preparation of the analytical dataset.
4. Exploratory data analysis (EDA) as the core of analytical work.
5. Data quality and completeness; the data-generating process and source selectivity.
6. Data report — team presentations and feedback.
7. From a problem to a research-tractable question; scope of inference.
8. Operationalisation and measurement validation; data integration from multiple sources where needed.
9. Text as data — text representation, feature extraction, and automated measurement.
10. Description, prediction, and explanation — from the research question to the analytical strategy.
11. Analysis plan and project consultations.
12. Modelling, evaluation, and a reproducible analytical pipeline.
13. Inference from observational data and robustness analysis.
14. From results to argument: structure and writing of a research paper; project consultations.
15. Final presentations and course wrap-up.

The order of topics and the level of detail in which they are covered may be subject to minor adjustments.