Watch the recording

Session title: “Emerging best practices for predictive model development and use”

Time and place: 19 May 2021, Virtual

Event: NYU CIC Conflict Early Warning/Early Action workshop


  • Angelica Lindqvist-McGowan (moderator)
  • Håvard Hegre (panelist)
  • Paola Vesco (panelist)
  • Nils Metternich (panelist)
  • Youssef Chaitani (panelist)
  • Joaquin Salido Marcos (panelist)

Session overview:

As the images of the violent uprisings in Colombia and the rocket attacks in Gaza hit the news, the development of early warning systems to help predict and mitigate violence seems more pressing than ever. Recent research in the forecasting of conflict has benefitted greatly from the availability of refined data, new theoretical insight, as well as progress in computational techniques, paving way for new innovations in data-driven conflict forecasting. This panel presents a comprehensive overview of the most recent of these advancements, as well as their potential use, relevance and applications in the policy sector. Among the most promising innovations, the panelists discuss three systemic approaches to improving existing methods to forecast violence, and conclude with valuable insights from the user perspective.

Opening the panel is Dag Hammarskjöld Professor Håvard Hegre in presenting ViEWS – a Violence Early Warning System that produces monthly forecasts of political violence for Africa at both the country and the sub-national level. Based at the Department of Peace and Conflict Research at Uppsala University in Sweden, the system relies on fully open-access data, a transparent setup, a solid evaluation process, and an ensemble methodology that overall promote accuracy and performance.

The introduction is followed by a discussion by Dr. Paola Vesco on how the promotion of broad collaborations across diverse scientific teams has been shown to encourage considerable improvements in our collective ability to forecast violence. As clear from a recent forecasting competition organized by ViEWS, a diversified ‘crowd’ of predictive models is able to perform better in predicting conflict than any individual model in isolation.

Next, Dr. Nils Metternich will discuss how models that rely on actor layers and disaggregated information about rebel groups and their interrelations can improve the accuracy of the forecasts and provide much more detailed, accurate and relevant results than standard approaches based solely on location-based data. 

Last, Dr. Youssef Chaitani and Mr. Joaquin Salido Marcos will offer a concluding discussion on how early warning can be used to improve policy and programmatic prioritisation; how prediction models based on solid, transparent and accurate methods like those discussed in this panel can help decision-makers, international organizations, and other stakeholders to identify areas at risk, allocate resources where they are most needed, and boost increased efficiency in preventative strategies.