Reading wars from the news: media coverage in zones of armed conflict, 1979–2026

Russia's long arc, Israel's October 2023 shock, and what the news-media signal reveals about conflict zones.

Armed conflicts have a particular signature in news coverage: many events, mostly negative, often disputed in their framing. The same database that gives me a real-time signal of democratic backsliding (GDELT, 1979–present) also lets me look at how the coverage of conflict-affected countries evolves over years and across specific war events. This post focuses on four such cases: Russia (full 1979–2026 history), Israel, Iraq, and Turkey.

A caveat upfront. Ukraine and Iran — the two countries that the current news cycle would most naturally include in this analysis — are not in my current per-country dataset. They require additional BigQuery extractions that I have not yet run. So this post is about what is visible in the data I have, not a complete picture of the current geopolitical moment.

Russia: a forty-seven-year arc

The most analytically interesting feature of the Russia data is what happens to the tone signal when you correct for the GDELT v1→v2 measurement break in 2015. The raw mean tone shows a dramatic "cliff" around 2013–2015 that looks like it could be the Crimea annexation. In fact most of that cliff is the artifact: the database changed how it computed tone, and the level dropped six points globally. Once you z-score Russia's tone against the world average per year, the artifact disappears, and what you see is a steady decline beginning around 2008 (the Georgia war), not a 2014 cliff.

The Goldstein scale (which captures conflict versus cooperation in event types, and which is not affected by the v1→v2 change) tells a complementary and equally striking story. Russia was substantially more cooperative-toned than the world average from 1979 through about 2007. From around 2008, it crossed below the world line and has stayed there since. The Goldstein slope for Russia is −2.46 per century (p<.001) — among the steepest declines in the dataset. By 2025, Russia's Goldstein is near zero (i.e., events covered are roughly half conflictual, half cooperative on average) against a world mean of approximately 0.6 (cooperative on net).

Each major military operation registers as a discontinuity: the first Chechen war (1994), the second (1999), the Georgia war (2008), Crimea (2014), and the full-scale Ukraine invasion (2022). The 2022 invasion is the clearest single shock in the entire timeline. At monthly resolution, Russia's z-scored tone drops sharply in February 2022 and has not recovered through 2026.

Israel: the October 2023 shift

The Israel case is methodologically the most clean. Before October 2023, Israel's monthly z-scored tone bounces around the world average with substantial volatility — there is no clear pre-event trend. After October 2023, the distribution collapses into a deeply negative band and stays there. The before/after split is so clean that you almost don't need statistics: the kernel density of the two periods barely overlap.

This is what news-side monitoring of an active war zone looks like. The October 7 Hamas attack and the subsequent Israeli military response in Gaza produced a sustained shift in how Israel is framed in global news coverage, lasting at least the 30+ months for which I have data.

Iraq and Turkey

Iraq sits where you would expect — near the bottom-left of the global field, in what I call the "authoritarian-floor attractor" basin. The country has been there throughout the 2015–2026 window. The Goldstein signal is consistently negative.

Turkey is the moving case. Turkey's β is rising significantly (it is one of the eight canary countries from the global analysis), but the country is moving toward more positive tone and more cooperative coverage, not less. This is initially counterintuitive — Turkey is widely classified as a deteriorating democracy by survey-based indices. The data suggests that the framing of events in Turkish coverage has normalized over the past decade, even as institutional democracy has declined. Several possible explanations: media co-option, reduction in conflictual reporting, or genuine post-2016 stabilization. Distinguishing among them requires more than β alone.

What is missing

Ukraine and Iran are the most important gaps. Both would substantially strengthen this analysis. For Ukraine, we would have a direct read of the country's media trajectory before, during, and after the 2022 invasion — the analog of what Russia's z-tone shows. For Iran, we would have the Middle East counterpart to Israel: the country's media coverage through the period of escalating regional tension. I plan to extract both in the next data pull.

What can be done now: anyone with BigQuery access can run the per-country queries I documented in the GitHub repo and add Ukraine, Iran, or any other country to this kind of analysis. The methodology generalizes.

Source code and queries: github.com/alejandrofeged/conflict-framing.

---

Alejandro Feged-Rivadeneira is a researcher at Tecnológico de Monterrey working on computational social science, political economy of media, and human ecology.