New Model Predicts Electoral Violence

For the study, the researchers built and trained a set of machine-learning models to forecast the likelihood of electoral violence on a global scale. Using a comprehensive set of data sources, with features including economic indicators, records of historical violence, political instability, and digital vulnerability, they then predicted the risk of electoral violence on a scale from no violence to severe violence. The study uses a broad definition of electoral violence, including intimidation and harassment as forms of violence.

We hope this model can help inform the work to protect and promote democracy and electoral integrity around the globe

The model was then evaluated on all elections between 2014 and 2023, using only data available before the elections. In this evaluation, the model showed an impressive predictive performance, with over 80% of predictions being correct. The paper also includes case studies evaluating how the final model, trained on data up until 2023, performed in predicting this year’s elections in Venezuela and India.

The final case study sets out the model’s prediction for the upcoming US election, where the model predicted a relatively high likelihood of electoral violence, with an approximate 51% probability of limited violence and a 1% probability of severe violence.

– We hope that the worst election-related violence in this US election cycle is behind us and that this model can help inform the work to protect and promote democracy and electoral integrity around the globe, says project leader David Randahl.

This paper contributes to political violence prediction by providing a medium-term data-driven forecasting tool for electoral violence and ties into the department’s tradition of forecasting systems.

The research was led by David Randahl, and the team also included Maxine Leis, Tim Gåsste, Hanne Fjelde and Håvard Hegre based at our department, as well as Staffan I. Lindberg from the V-Dem Institute at the University of Gothenburg and Steven Wilson from the Department of Politics at Brandeis University. The study was a collaboration with the Kofi Annan Foundation.

The paper is available as a pre-print through the VDEM working paper series website and technical report is available through the website of the Kofi Annan Foundation.

Stay tuned for an episode on the topic for the department's podcast Researching Peace!