Methodology
Data sources, preparation steps, analytical methods, and limitations used for the Russia–Ukraine conflict analysis.
01
Data Sources
UCDP Georeferenced Event Dataset
The primary dataset for this analysis is the UCDP Georeferenced Event Dataset. Event-level records were used to analyze reported battle-related fatalities, civilian deaths, event frequency, dates, and geographic information.
Our World in Data – War & Peace
Our World in Data conflict resources are also included in the broader project data workflow. The Russia–Ukraine findings presented on this page are based primarily on the filtered UCDP event data used in the project pipeline.
02
Data Preparation
Conflict Filtering
The UCDP dataset was filtered using the conflict name "Russia - Ukraine" to isolate the target conflict. This produced 37,892 event records for the Russia–Ukraine analysis.
Date Standardization
Event dates were standardized to support consistent time-series analysis. The resulting monthly dataset contains 50 monthly records covering March 2014 through December 2025.
Fatality Variables
The UCDP best estimate was used as the primary event-level fatality measure. Civilian deaths were retained separately to support analysis of civilian share within the recorded fatalities.
Monthly Aggregation
Event-level records were aggregated by year and month. Monthly summaries contain event counts, total reported deaths, and reported civilian deaths for time-series analysis. The resulting dataset contains 372,593 reported battle-related deaths.
Pipeline Output
37,892 event records were filtered from the source dataset and transformed into a 50-month analytical series. The processed records contain 372,593 reported battle-related deaths and 25,985 recorded civilian deaths.
03
Analytical Approach
Exploratory Data Analysis
Pandas, Matplotlib, and Seaborn were used to inspect the processed data, validate the cleaned datasets, and identify relevant temporal and geographic patterns.
Monthly Time-Series Analysis
Monthly reported battle-related fatalities were analyzed to identify higher and lower reported-fatality periods. The analysis also calculates month-to-month changes and deaths per recorded event.
Annual Analysis
Monthly records were further grouped by calendar year to compare annual deaths, event counts, civilian deaths, civilian share, and deaths per recorded event. Partial years are explicitly identified before making annual comparisons.
Geographic Analysis
Event location information was retained from the original event-level dataset, allowing geographic distribution to be examined alongside the temporal analysis.
Visualization
Static visualizations and an interactive Tableau Public dashboard were created to communicate monthly, annual, and geographic patterns from the processed data.
04
Tools & Technologies
Python
Data processing
Pandas
Data analysis
Matplotlib / Seaborn
Visualization
Tableau Public
Interactive dashboard
05 · Limitations
Interpreting the Results
The analysis covers March 2014 through December 2025. The 2014 coverage is partial, beginning in March, so its annual total should not be directly compared with complete calendar years without accounting for the difference in coverage.
The analysis is based on publicly available conflict-event records and therefore reflects the coverage and definitions of the underlying datasets. The 37,892 event records represent recorded events in the filtered dataset, not an exhaustive count of every conflict incident.
Reported casualty figures may be incomplete or revised as source datasets are updated. Differences in reporting, definitions, and dataset methodology can also affect comparisons across time.
Geographic and temporal patterns should therefore be understood within the scope of the available data rather than as a definitive record of every fatality.
This project focuses on data processing, analysis, trends, and visualization. It does not make political interpretations or conclusions about the conflict.