Arab world turmoilRisk-analysis models could not predict Arab world upheavals
Political instability is influenced by everything from the weather to local economic conditions and infant mortality rates; these factors interact in complex ways, and data quality can be low; together, this makes for a daunting forecasting challenge; says one expert: “All of our models are bad, some are less bad than others”
Upheaval across the Arab world // Source: thealligatoronline.com
The apparent suddenness of the eruption of unrest in the Arab world raises a question: Is it possible to forecast unrest? Analysts say that the upheavals in the Arab world are the perfect opportunity to put the best models of human conflict to the test.
New Scientists reports that, unfortunately, only those with access to classified intelligence documents will find out the results.
It is a problem that has troubled the field of conflict modeling for years. Many military funding bodies, notably the U.S. Defense Advanced Research Projects Agency (DARPA), bankroll attempts to forecast revolutions, terrorist activity, and other conflicts — but results emerge into the public domain only piecemeal, if they come out at all.
Take the Political Instability Task Force, funded by the CIA and based at George Mason University in Fairfax, Virginia. Since it was formed in 1994, it has used historical data on conflicts, political structures, and economics to rate the stability of countries around the world. Details of the model are publicly available, but the forecasts that the task force hands to the CIA are not.
Monty Marshall, a member of the task force, says models tend to do well at simulating historical events, but fail when it comes to predicting future unrest. “I don’t know if any of these efforts have been successful, and if they have been we wouldn’t know about it because it’d be classified,” he told NS.
Wired ran an article on this topic two weeks ago. It quotes Mark Abdollahian of Sentia Group, which has built dozens of predictive models for government agencies, saying: “All of our models are bad, some are less bad than others.”
The poor performance is not surprising. Political instability is influenced by everything from the weather to local economic conditions and infant mortality rates. These factors interact in complex ways, and data quality can be low. Together, this makes for a daunting forecasting challenge.
Even a model that managed to capture that complexity would find it impossible to pin down the timing of events. Take the case of Rwanda. In the early 1990s, models of the country could have noted the tension between Hutu and Tutsi, the country’s two main ethnic groups. The model might have flagged Rwanda as an at-risk state. No model, however, could have predicted the April 1994 plane crash that killed president Juvénal Habyarimana, and tipped the country into civil war and genocide.
This, says NS, is the crux of it. “Better models, able to simulate the complex networks that govern social stability, may be waiting in the wings. They could give an indication of which societies are unstable at a given time, and could even provide hints as to why. But certain triggers will almost certainly continue to evade modelers.”
