Forecasting mass atrocity and genocide

Ben Goldsmith’s Atrocity Forecasting Project has developed a tool for predicting and hopefully preventing mass atrocities around the world.

By Tim Groenendyk

Ben Goldsmith

"Our model also captures cases that are only just appearing in the news like Central African Republic and Myanmar." Associate Professor Ben Goldsmith

“The Project is not so much about understanding the causes of genocide as actually finding particular indicators that are the strongest forecasting tools for genocide,” said Associate Professor Ben Goldsmith and chief investigator at the Atrocity Forecasting Project.

“Our aim was, in particular, to create a forecasting model for genocide and politicide (the deliberate and systematic mass killing of a politically identified group) and something that could map atrocities more generally.

“It’s a quantitative tool that can be used to help predict and prevent mass atrocities.”

The Asia Pacific Centre for the Responsibility to Protect and AusAID funded the Project which also collaborates with University of New South Wales computer scientists and fellow chief investigator Professor Arcot Sowmya.

The forecasting model uses commercially available statistical and econometric applications but Goldsmith said that it’s the techniques they use to analyse the data that are more important than the software.

“Instead of just looking at variables that are generally static, we included variables like elections and changes in the political system or changes in the economy over a short period of time.

“The difference between our model and others is that it’s very sensitive to time varying factors.”

The tool analyses each country in two stages. The first stage of the model predicts political instability: civil war, a serious coup or a reversal of open governance.

“That estimate, for all countries in the world, is included in a model that then focuses specifically on the risk of genocide in that same period.”

The second stage estimates the likelihood of genocide or poiliticide over a certain period of time, such as in the next year or five years.

Goldsmith tested his model using a technique called ‘out-of-sample forecasting’ - which simulates actual future forecasting - to measure the accuracy of the results.

“For example, we’ll build the model so that it fits data up to the year 1990. Then we’ll forecast the year 1991, and of course the model hasn’t seen the data from that year.”

In a forthcoming paper in the Journal of Peace Research they compare their model against a commonly used tool developed by a CIA funded team. Not only were their results accurate but outperformed this model.

“We’re pretty confident that it’s among the best forecasting model of the sort.

“You don’t know what countries next year or the next five years are going to have instability so we argue you really need to include all countries in the world, or at least most countries that are at risk of that instability, in your forecast model, which is the real advantage of ours.”

The Atrocity Forecasting Project already lists Syria and Libya in their Top 15 Countries at Risk of the Onset of Politicide and Genocide, where mass killing of civilians because of their political or ethnic identity still occur.

“But our model also captures cases that are only just appearing in the news like Central African Republic and Myanmar.

“Myanmar, even though the political system is liberalising, we actually find that a risk factor for genocide in the country.

“It’s not one of the most powerful factors, but when you have a regime that begins to liberalise - and otherwise has risk factors - then that ‘opening up’ can also present the sort of instability in which the state might consider the option of mass killing.”