Analyses
Autocratic dynamics and democratic backsliding: what GDELT lets us see
How GDELT event data helps study protest, repression and authoritarian adaptation, and what the measurement leaves out.
Working English version. The Spanish edition is the reference text. This translation is published for reach and will be revised before the final launch.
What GDELT lets us see
The GDELT project (Global Data on Events, Location, and Tone) turns news from press, radio and television into coded events: who did what to whom, where and when. That structure makes it possible to study protest and repression at a scale no research team could cover by hand.
The promise is attractive: millions of events, more than a hundred languages and constant updates. The risk is equally clear: an event is not an objective fact, but a signal mediated by censorship, communications infrastructure and source bias. This piece summarises what some of the most cited studies have found and what precautions the data require.
To understand how actors, actions and geolocation are coded, see the GDELT variable dictionary.
Data note. The previous version of this portal showed illustrative charts that were not linked to a reproducible download. This version keeps the argument and cites the studies behind the figures. The protest series will be added as a visualization with documented data and method.
The 2011 rupture
The literature on protest and authoritarianism agrees on a shift in the global dynamics of conflict after 2011. Cantoni, Yang, Yuchtman and Zhang document, with more than 1.2 million standardised protest events between 1980 and 2020 in 218 countries, a turning point associated with the spread of mobile telephony and social media.
The dominant interpretation is mechanical: technology lowers the cost of coordinating collective action and makes it easier to organise protest where traditional channels are controlled. Before 2011, consolidated democracies had a higher relative protest rate; afterwards, the rate in autocracies and weak democracies steadily surpassed that of mature democracies.
This does not mean autocracies become more fragile. It also forces them to innovate. The authoritarian response shifts towards surveillance, censorship and management of the information space.
Six studies, four cases
| Study | Case | Variables | Main contribution |
|---|---|---|---|
| Beraja, Kao, Yang and Yuchtman (2023), QJE | China, 344 prefectures | CAMEO 14 protests (2014-2020), AI contracts, patents | Protest stimulates surveillance procurement; public data subsidises private innovation. |
| Otlan, Kuzmina, Rumiantseva and Tertytchnaya (2023), Post-Soviet Affairs | Russia, Vremya news | Protest events, text corpus, topic models | State television reduces coverage of foreign protests during domestic mobilisation and frames Western ones as violence. |
| Lukito, Cui, Hu, Lee and Ozawa (2022), Media and Communication | Hong Kong, Indonesia, Myanmar | CAMEO 15, 17 and 18-20; ACLED; VAR models | Institutional propaganda temporally precedes physical coercion. |
| Cantoni, Yang, Yuchtman and Zhang (2023/2024), NBER / Annual Review of Economics | 218 countries | 1.2 million protest events (CAMEO 14) | Structural break in 2011: mobile internet penetration drives protest in autocracies. |
| Morales-Arilla (2024), working paper | Venezuela, 2019 blackouts | Geocoded protest, police coercion, power grid | Strategic rationing: electricity goes to loyalist strongholds and outages concentrate in opposition areas. |
| MLP DevLab / Donnelly et al. (2024/2025), OSF | 65 developing countries | Civic events and GKG thematic co-occurrences | Civic space closure concentrates between elections, after peaks in corruption news. |
The four cases that structured the older version of the portal remain useful as illustrations: China, Russia, Southeast Asia and Turkey. They are not a representative sample, but studies selected for their ability to show different mechanisms.
Mechanisms
The autocracy feedback loop
Beraja and his co-authors describe a feedback circuit. Protests detected by GDELT increase public procurement of surveillance technologies, especially facial recognition. That demand gives companies massive biometric data, subsidises their commercial R&D and, over time, reduces the capacity for protest.
The mechanism matters because it inverts the liberal intuition: repression does not only destroy social capacity; it also builds private technological capacity that is later reused in other markets.
Propaganda before coercion
Lukito and her co-authors combine CAMEO events with ACLED data and VAR models. Their finding is sequential: polarisation and dehumanisation campaigns precede the deployment of physical coercion. Propaganda is not an ornament of authoritarianism; it prepares the ground for repression to carry a lower reputational cost.
In Russia, Otlan and her co-authors document a complementary logic: during domestic mobilisations, state television reduces coverage of foreign protests and, when it covers Western mobilisations, frames them as chaos and violence. Information management is part of regime management.
Strategic rationing
Morales-Arilla studies the 2019 Venezuelan blackouts. Electricity was not distributed randomly: it was channelled to loyalist strongholds, while outages concentrated in opposition areas, alongside police repression. Infrastructure becomes an instrument of punishment and reward.
The limits of measurement
Working with GDELT in authoritarian contexts requires taking reporting bias seriously. Weidmann (2016) showed that events do not passively reflect conflict: they are signals mediated by censorship and source selection.
Three problems are especially relevant:
- A fall in protests may be an artefact. Fewer events do not necessarily mean less conflict; they may mean a more effective information lockdown.
- Urban and infrastructure bias. Algorithms prioritise outlets with a presence in capitals and connected areas, invisibilising the rural periphery.
- Series breaks. The 2014 GDELT update multiplied the number of monitored sources. Any longitudinal analysis crossing that year must treat the change as a structural break, not as a real increase in conflict.
The literature offers three methodological responses: normalise event frequency by the total volume of apolitical news for the country-year; triangulate GDELT with manually curated datasets such as ACLED, ICEWS or SCAD; and control for institutional covariates from projects such as V-Dem or Polity.
What to watch
If GDELT is used to study autocratization, the question should not be only how many protests appear, but what happens to coverage when protest challenges the regime. Censorship, source substitution and selective amplification are part of the phenomenon, not noise to be cleaned away.
The next piece in this series explains how democracy is measured with V-Dem and what a change of 0.1 in an index does, and does not, mean.
References
- Beraja, M., Kao, A., Yang, D. Y., & Yuchtman, N. (2023). AI-tocracy. The Quarterly Journal of Economics.
- Cantoni, D., Yang, D. Y., Yuchtman, N., & Zhang, Y. J. (2023/2024). Protests and the state. NBER Working Paper / Annual Review of Economics.
- Lukito, J., Cui, Z., Hu, A., Lee, T., & Ozawa, J. V. S. (2022). The state of the state: Propaganda and coercion. Media and Communication.
- Morales-Arilla, J. (2024). Strategic rationing and repression: Evidence from Venezuela. Working paper, Princeton University.
- Otlan, Y., Kuzmina, Y., Rumiantseva, A., & Tertytchnaya, K. (2023). Managing protest news in personalist regimes. Post-Soviet Affairs.
- MLP DevLab / Donnelly et al. (2024/2025). Civic space and inter-electoral repression. OSF / University of Pennsylvania.
- Weidmann, N. B. (2016). A closer look at reporting bias in conflict event data. American Journal of Political Science.