Vladimir Putin’s invasion of Ukraine sent prices of food and fuel soaring. Around the world, people are suffering as a result. Millions risk starvation. Yet many governments, having borrowed heavily during the pandemic and with interest rates rising, appear unwilling or unable to cushion the blow. What does this portend for political violence and unrest? Can statistical modelling help us understand what to expect – and where?
Shortly after the articles in which the project was featured were published we were asked to share our methods and approach with the United Nations, which we did, in what Sondre Ulvund Solstad (the data journalist) was instructed to call “a high-level discussion at UNHQ”. The discussion centred on how the UN, which had been inspired by the journalist work, could use these methods to better understand and predict unrest, and how they could set up their own system to do so. Sondre also gave subsequent talks in response to requests and wide interest from academia and industry. Beyond this enthusiastic response, the main article featuring this project reached a wide audience by our standards (and was featured as a “fly” on the cover of our world-wide print edition). The project was also the basis for an editorial article (or “leader” – link 2 below), calling for and specifying policies to help more food and money reach those who need it.
The data journalism project consisted of three parts. First, original data collection and number crunching to contextualize the problem effectively in our journalism. Second, the construction of a vast dataset, with information including factors such as daily brent crude prices, inflation, unemployment, economic forecasts, government debt levels and revenues, countries’ demographic profiles, expected food prices and imports, and a corresponding set of data on unrest in different countries around the world based on data from ACLED. Third, careful modelling work. This involved attempting to understand precisely what we can – and cannot – learn from such data, and then, having established it was possible, using a machine learning method (gradient boosted trees) to build a model to estimate expected increases or decreases in unrest in countries around the world in the coming year.
Context about the project:
Just one thing: We were unfortunately was unable to share source and data used in the project due to data rights issues. Wish we could! If anyone wants more detail on the statistical/machine learning methods employed, do let me know (firstname.lastname@example.org).
What can other journalists learn from this project?
We hope other journalists are emboldened to attempt similar projects themselves – this type of work is usually done by industry or governments, who have their own agendas.
Beyond that, the hardest part of this project was figuring out what we could and could not learn from the data. To be concrete: in a situation with a lot of unrest data, it might for instance be tempting to try to predict where there will be most or least unrest. We concluded that was not possible, because the data from ACLED was not comparable across countries (put a bit crudely: an anti-government protest in Germany is not the same as one in Eritrea). However, within countries, data was comparable over time, enabling us to project increases or decreases in %.