Pick a country. See how close its media environment is to historical pre-conflict patterns — and what's driving it. Validated against Rwanda, Myanmar, Ethiopia, and the 2022 Russian invasion of Ukraine.
PRISM adapts the Groseclose-Milyo media slant framework — originally designed to measure ideological positioning of U.S. outlets through citation analysis — and extends it to conflict environments. The core question: how polarized is a country's media ecosystem, and how closely does that polarization match patterns we've seen before major crises?
The model scores five dimensions: citation concentration (are outlets citing the same sources or splitting into echo chambers?), narrative convergence (how uniform is the framing across outlets?), information entropy (diversity of topics and viewpoints in circulation), source-outlet alignment (are media outlets tracking closer to state or non-state actors?), and escalation velocity (how fast is rhetoric intensifying?).
Each dimension is scored 0–100 and weighted. The composite score is compared against historical baselines from pre-conflict media environments (Rwanda 1993, Yugoslavia 1991, Ukraine 2013) to generate a pattern match. All weights are adjustable — you can change what the model prioritizes and see how the output shifts in real time.
Every number traces back to an indicator. Nothing is opaque. If you disagree with a weight or an assumption, change it and watch what happens to the score.
Model an intervention: what happens to the threat level if you add a new outlet to the ecosystem?
Define reference entities (think tanks, parties, factions, institutions) with known ideological positions. Position: -1 (left pole) to +1 (right pole).
Define media outlets to analyze. Audience is relative weight (e.g., millions of viewers/readers).
Enter citation counts: how many times each outlet cited each anchor entity during the analysis period.
Add anchors and outlets first to generate the citation matrix.
Paste CSV data below. Format: Outlet, Audience, Anchor1, Anchor2, ...
First row must be headers. Subsequent rows contain outlet data with citation counts per anchor.
Want the full picture? Network graphs, scenario simulator, GDELT live search, and custom data input.
Ad Fontes and AllSides answer: "How biased is this outlet?" This tool answers: "How dangerous is this information environment, and where is it heading?" The unit of analysis is the ecosystem, not the outlet. The output is predictive, not descriptive. And the framework is grounded in conflict studies, not media literacy.
Adapted from Groseclose & Milyo (2005). Each outlet is scored [-1, +1] based on its citation patterns of anchor entities with known positions, using IDF weighting to discount universally-cited sources.
Audience-weighted standard deviation of SQ scores, normalized to [0-100]. Measures whether the ecosystem is clustered (healthy) or bimodal (polarized).
Derived from Mueller & Rauh (2018, APSR) who found that the disappearance of shared topics (diplomacy, trade, judiciary) is more predictive of conflict onset than the presence of conflict topics. NCI measures pairwise cosine similarity of outlet citation profiles.
Shannon entropy of ecosystem-wide citation distribution. Based on the Yugoslavia pattern: Milosevic controlled 95% of Serbian media by 1989 (Fogg, W&M; MOM Serbia). Low entropy = state capture or monopoly narrative.
Derived from the Myanmar finding: 98% of social media users were on Facebook, and algorithmic reinforcement created sealed information bubbles (Yale Macmillan Center, 2024). XD measures the audience-weighted extremity of media consumption.
Jamison et al. (2023, SSRN) found that media sentiment spikes, not levels, are the strongest short-term conflict predictors (AUC 86-94%). This metric captures the rate and acceleration of polarization change.
Five-factor model with weights calibrated from case literature. The public showcase demonstrates the framework; full calibration parameters, pattern matching trajectories, and proprietary weighting schemas are available under commercial license.
Threat levels are aligned with Stanton's Ten Stages of Genocide and Benesch's Dangerous Speech Framework:
| Range | Level | Interpretation | Historical Pattern |
|---|---|---|---|
| 0-20 | Normal | Healthy media competition | Canada/Germany baseline |
| 20-40 | Watch | Polarization trends detected | Early divergence signals |
| 40-60 | Warning | Narrative fragmentation accelerating | Ethiopia 2019, Myanmar 2020 |
| 60-80 | Alert | Pre-conflict media patterns detected | Yugoslavia 1991, Myanmar 2021-Q1 |
| 80-100 | Critical | Media environment resembles onset conditions | Rwanda 1994, Yugoslavia 1992 |
The tool compares the current ecosystem's [Polarization, NCI, Entropy, Echo, Velocity] vector against reconstructed trajectories from five documented cases: Rwanda (Yanagizawa-Drott 2014, Roozen & Shulman 2014), Myanmar (Yale Macmillan 2024), Yugoslavia (Fogg, PBS Frontline), Ethiopia (Nigussie et al. 2024), and a stable democracy baseline. Pattern matching uses normalized Euclidean distance.
Foundational
Groseclose, T. & Milyo, J. (2005). A Measure of Media Bias. QJE, 120(4).
Mueller, H. & Rauh, C. (2018). Reading Between the Lines. APSR, 112(2).
Yanagizawa-Drott, D. (2014). Propaganda and Conflict. QJE, 129(4).
Benesch, S. (2014). Countering Dangerous Speech. USHMM Working Paper.
Stanton, G. (2016). Ten Stages of Genocide. Genocide Watch.
Hegre, H. et al. (2019). ViEWS: A Political Violence Early-Warning System. JPR, 56(2).
Case Studies
Roozen, K. & Shulman, S. (2014). Tuning in to the RTLM. Conflict & Communication.
Nigussie, D. et al. (2024). Echoes of Violence: Hate Speech in Ethiopian Media. SAGE.
Jamison, A. et al. (2023). Media Sentiment and Future Conflict Events. SSRN.
2025-2026 (Cutting Edge)
Petrova, M. & Tapsoba, A. (2025). Information and Conflict. Economic Policy, Oxford.
Piccardi, T. et al. (2025). Reranking Partisan Animosity Alters Polarization. Science.
EPJ Data Science (2025). Predicting Collective Violence Through Hostile Social Media Discourse.
STFT-VNNGP (2025). Forecasting Geopolitical Events with Sparse Temporal Fusion Transformer. arXiv.
VIEWS/PRIO (2025-2026). Violence & Impacts Early-Warning System monthly forecasts.
ACLED (2026). Conflict Index & 2026 Watchlist.
Enter your email to receive the full polarization assessment with PRISM score, instability classification, conflict-pattern comparison, and methodology documentation.
Commission a full media polarization assessment with custom outlet data, calibrated anchor entities, historical trend analysis, and conflict-pattern comparison. Available for media monitoring organizations, peacebuilding NGOs, government agencies, and research institutions.