Estonian Case-Law Citation Network

Every citation made by an Estonian court, as a machine-readable graph.

Estonian courts publish their decisions, and the state publishes its legislation, but nobody publishes the link between them. This dataset is that link: which decision relies on which statutory provision, which decision cites which earlier case, and which decisions apply EU law. It was extracted from the full text of 564,176 published decisions spanning 1993 to 2026.

It contains no judgment text and no party names — only identifiers, case numbers, section numbers and dates.

Why this exists

Empirical legal research on Estonia has had no citation graph to work with. Riigi Teataja publishes decisions as documents; commercial legal databases build citation indexes but keep them proprietary. Questions that are routine in other jurisdictions — which provisions actually get litigated, how precedent propagates between instances, whether a rule is settled or contested — have not been answerable at scale here.

A few things the data shows immediately:

• Case law concentrates hard. 3.05 million citations resolve to just 197 acts, and to 9,684 distinct provisions.
• Procedure dominates substance. The single most-cited provision is Tsiviilkohtumenetluse seadustik § 162 (allocation of costs), cited by 87,324 decisions, ahead of Võlaõigusseadus § 113 (default interest) at 81,403.
• Most litigation is about money. Six of the ten most-cited provisions concern costs, interest or the performance of obligations.

Files

Full field descriptions are in datapackage.json (Frictionless Data Package, validates with frictionless validate datapackage.json).

Every identifier resolves to a public page:

• a decision → https://lahend.ee/lahend/{ruling_id}-{case_no_slug}
• an act → https://www.riigiteataja.ee/akt/{act_id}
• a provision → https://lahend.ee/praktika/{act_id}/{paragraph}
• an EU instrument → https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:{celex}

Getting started

import pandas as pd

cites = pd.read_csv("data/ruling_law_citations.csv.gz")
acts  = pd.read_csv("data/acts.csv.gz")

# The most-litigated provisions
top = (cites.dropna(subset=["act_id"])
            .groupby(["act_id", "paragraph"])["ruling_id"].nunique()
            .sort_values(ascending=False).head(20)
            .reset_index()
            .merge(acts[["act_id", "abbrevs"]], on="act_id"))
print(top)

# Precedent in-degree: which cases the courts keep coming back to
prec = pd.read_csv("data/ruling_citations.csv.gz")
print(prec["cited_case_root"].value_counts().head(20))

How it was built

Citations were extracted from decision full text by pattern matching, then resolved against the consolidated legislation corpus:

1. Provision references (TsMS § 162 lg 1, VÕS § 113 lõige 1) are parsed into an act abbreviation, a section number and, where present, a subsection.
2. Abbreviations are resolved through a curated mapping of 303 abbreviation forms to Riigi Teataja act group ids. This is deliberately not taken from Riigi Teataja's own lyhend field, which is populated for only a few hundred of 117,000 acts and contains errors — in that field KarS points at a 1990s ministry funding decree rather than the Penal Code.
3. Case references are matched to a case root, the part of a case number shared by every instance hearing the same case.

Limitations

Read these before drawing conclusions.

• 5.8% of provision citations are unresolved (act_id is null). These are abbreviations with no curated mapping, mostly rare or repealed instruments. The abbreviation as written is retained in act_abbrev_as_cited so they can still be analysed or resolved by others.
• Only decisions with available full text could be mined. 353,812 of 564,176 decisions have text; the rest appear as nodes with no outgoing edges. Coverage is much better for recent years, so citation counts are not comparable across decades without normalising by the number of decisions with text in each year.
• Cited case roots are not resolved to decision ids. A case root may match several decisions (one per instance) or none in this corpus.
• A citation is not an endorsement. The extraction records that a decision referred to a provision, not whether it applied, distinguished or rejected it.
• The corpus is what Riigi Teataja publishes. Decisions withheld from publication are absent, and Estonian courts do not publish everything.

Licence and attribution

Released under CC BY-SA 4.0. The underlying decisions and legislation are public documents of the Republic of Estonia; the citation extraction, resolution and structuring are the work of Nimistu MTÜ.

Citation

Archived on Zenodo with a DOI. The concept DOI below always resolves to the newest version; cite the version DOI if you need to pin an exact release.

BibTeX:

@dataset{nimistu_2026_case_law_citations,
  author    = {{Nimistu MTÜ}},
  title     = {Estonian Case-Law Citation Network},
  year      = {2026},
  publisher = {Zenodo},
  version   = {1.0.0},
  doi       = {10.5281/zenodo.21598030},
  url       = {https://doi.org/10.5281/zenodo.21598030}
}

About

Built and maintained by Nimistu MTÜ, a non-profit that publishes Estonian public data. The corpus behind this dataset is searchable for free, in full text, at lahend.ee — including case law by provision and consolidated legislation.

Corrections and questions: info@nimistu.ee