RDT War Database 1944–2025: Resource Competition and Armed Conflict Escalation

Read full text Summarise with VIRAT
Authors: Abdullah Oraj Huseyniklioglu
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The RDT War Database: an episode-level dataset of armed conflicts, 1944-2025. WITHDRAWAL NOTICE (18 September 2026): releases v12, v13, v14 and v15 are WITHDRAWN and must not be used for analysis. They repaired the file by trusting its English episode-name column and rewriting dates to match; that column is the broken one. Use RDT_WarDatabase_v18_20260919.csv. The withdrawn files stay in the deposit so the error can be checked. See WITHDRAWAL_NOTICE_v12_v15.txt. The original v11 file was never modified. This deposit contains 205 conflict episodes coded on resource linkage, conflict structure, duration and outcome. Resource linkage is recorded twice and independently: as a five-category dominant-resource typology (none, water/land, oil/gas, minerals, strategic/geopolitical), and as a stricter binary resource-driver sub-code (G-2) assigned only where a source documents behaviour directed at the resource itself. Additional codings cover war type, external involvement, power asymmetry, temporal pressure, dependency intensity and a unified escalation recode. Duration is recorded in days. A conflict is treated as right-censored only where no end year is recorded, that is, where it was still running at the 31 December 2025 collection cutoff; 20 of 205 episodes meet this condition. Sixty episodes previously carrying an "Ongoing" outcome label despite a recorded end year have been relabelled; the original label is preserved in the column result_v10_orijinal and the affected rows are flagged by sonuc_duzeltildi. Known defect in the English name column: the English episode name belongs to a different war from the one the rest of the row describes. A case-by-case reading of all 205 rows establishes 28 such rows, one row that cannot be tied to any single conflict, and one that duplicates another row under a contradictory name. In every adjudicated case the row's own Turkish category field, its recorded years and the UCDP/PRIO Armed Conflict Dataset agree against the English name. RDT_WarDatabase_v18_20260919.csv replaces those 28 names and flags the other two, and alters no date, duration, outcome or resource field, with one documented exception (record 186). v16 and v17 are intermediate releases kept for provenance. AD_KATEGORI_KARSILASTIRMA_20260918.csv holds the screen over all 205 rows. ONSET-WINDOW MEASUREMENT (20 September 2026): the resource labels have now been checked against dated evidence inside each episode's onset window (the calendar year before onset through the year after it), under three mechanism rules covering distributional claims, non-state resource finance, and observable resource-related events. 168 of the 176 core episodes are coded; the eight that are not are named in the accompanying manuscript. Of 136 positive legacy resource labels, 59 are corroborated and 77 return no qualifying passage under a bounded English-language search; three of 32 None labels are overturned. Agreement between the legacy label and the measured exposure is 0.52 against 0.42 expected by chance. The coding sheets, one row per episode with source, locator, supporting passage, decision, rationale and retrieval counts, are deposited as M1_M3_PILOT_KODLAMA_20260918.csv, PILOT2_KODLAMA_45_20260919.csv, PILOT3_KODLAMA_TUR3_20260919.csv and PILOT4_KODLAMA_TUR4_20260920.csv. A zero records the outcome of that bounded search and not the absence of the mechanism. All four coding rounds followed the same written protocol, and every positive was re-checked by the author against its cited source; this is not independent human coding, and no reliability coefficient is claimed for it. README_REVISION.txt is the version manifest and names the script behind each reported result. The rating matrix from a five-coder exercise on a random 50-episode subsample remains in this deposit, under reliability/, for provenance. No reliability coefficient is claimed from it. Earlier versions of this description reported Krippendorff's alpha figures from that matrix; those figures are withdrawn, because the exercise was not blinded and its independence cannot be established from the deposited materials. Candidate cases are drawn from Gantzel and Schwinghammer (2000), the Correlates of War inter-state and intra-state war data, and the UCDP/PRIO Armed Conflict Dataset, under a threshold of at least 25 battle-related deaths in a calendar year. Note on two single-letter codes that expand into Turkish: external involvement takes V for present and Y for absent (var / yok); power asymmetry takes H for high asymmetry and M for symmetric. Reading Y as "yes" inverts the external-involvement variable. Superseded file: RDT_WarDatabase_v2_7_FINAL.csv

Fields of Study & Subjects

Political Conflict and Governance Natural Resources and Economic Development Defense Military and Policy Studies

Publication details

Publication date
22 March 2026
Language
EN
Document type
Preprint