War–Peace Index
A daily measure of whether global news coverage of the Russia–Ukraine conflict leans toward escalation or de-escalation, and how much of the news it takes up. Built from the GDELT news archive and updated every night.
Daily, 1 January 2018 to the latest complete day · 3,228 observations
Used in Expectations of War and Peace: Evidence from Ukraine's GDP-Warrant (2026).
The latest reading loads with the data.
The series
Does it move the right way?
On days with an unambiguous story, the index should point the obvious direction: sharply war-leaning on the day of the invasion, toward peace on days of talks. The lettered dates are marked on Figure 1.
| Date | Event | Peace | War | Direction | Signed intensity (‰) |
|---|
Three measures from three counts
Each day GDELT reports how many monitored articles match a peace query, how many match a war query, and how many it monitored in total. The index splits that into how much of the news is about the conflict and which way it leans; salience times direction equals signed intensity exactly.
Salience
(peace + war) / total
Share of all monitored news that is conflict coverage. It peaked at 19.1% on 27 February 2022.
Direction
(peace − war) / (peace + war)
Tone of conflict coverage, from −1 (all war) to +1 (all peace). Left blank on days with fewer than 50 matched articles.
Signed intensity
(peace − war) / total
Net peace-minus-war articles per 1,000 monitored. The main series in the paper; its daily change is the news shock.
How it is built
Every query requires both a Ukraine term and a Russia term, which keeps out stories that mention only one side, such as the downing of Flight 752 in Iran on 10 January 2020.
(Ukraine OR Kyiv) (Russia OR Russian) (war terms)
Peace terms: ceasefire, “peace talks”, “peace deal”, truce, armistice, negotiations.
War terms: invasion, offensive, shelling, airstrike, bombardment, mobilization, escalation.
The dictionaries are narrow on purpose. Words like “summit”, “diplomacy”, “missile” and “assault” turn up in routine coverage that signals no change in the conflict, so they are left out.
- Source. GDELT DOC 2.0 API, raw volume timeline, unsmoothed. Article counts and the total-monitored normaliser are read by series label.
- Calendar. Every day from 1 January 2018 is a row. GDELT returned nothing on 23 days; those rows are blank, and so is the next day's change, so no multi-day move is reported as a one-day shock.
- Double counting. An article matching both dictionaries counts in both. This raises salience slightly and cancels out of signed intensity.
- Thin days. 10 days had fewer than 50 matched articles. Direction is hidden on those days; signed intensity is kept, since the total-news denominator already pulls them toward zero.
- Updates. A scheduled job adds the previous day each night and publishes only complete days.
Download and cite
The full daily series, one row per calendar day, under CC BY 4.0: free to use with attribution. Underlying article counts come from the GDELT Project.
Download CSVdata through
Suggested citation
| Column | Definition |
|---|---|
peace, war | Matched article counts |
total_news | GDELT total monitored articles |
salience | (peace + war) / total_news |
direction | (peace − war) / (peace + war) |
signed_intensity | (peace − war) / total_news |
d_signed_intensity | One-day change; blank after a missing day |
low_volume | 1 if peace + war < 50 |