Professor Michael Spagat, Royal Holloway, University of London
Chris - Here on The Naked Scientists this week, we are talking about the science of explosions and one group of people who often use them are insurgents, in other words, terrorists. We’re hearing about this on almost a daily basis in Afghanistan and also at times in Iraq. Now, it turns out that by studying the pattern of these events, we might be able to learn something and maybe even predict when they're going to happen. And Professor Michael Spagat who is at the Royal Holloway is working on this very thing. Hello, Michael.
Michael - Hello. How are you?
Chris - Very well, thank you. Welcome to The Naked Scientists. Do tell us a little bit about this work which you call The Ecology of Insurgency.
Michael - Okay. The reason we call it The Ecology of Insurgency is because we think of the conflict process as being a struggle over some kind of limited resource. The limited resource might be various things. So for example, it might be space or it might be something more like media attention. So, if you imagine that there are a soup of insurgent units out there or guerrilla cells and then there are also anti-insurgent units that are seeking them out, sort of shuffling around in space, and randomly bumping into each other. Sometimes insurgent groups might bump into one another and they might coalesce into a stronger group. Sometimes they might feel under pressure and fragment into a bunch of smaller weaker groups or insurgents might encounter anti-insurgent forces in which case, there could be a bloody clash and there could be casualties.
At the same time, insurgent groups might be planning attacks which we think of as a way of trying to call attention to themselves, for example, through getting media attention. We think of this attack planning process very much in the same way that people plan the routes that they take, driving home from work for example, where they're trying to predict what other people are doing, they're trying to predict which roads are going to be empty and that they’ll be able to drive down relatively easily. In the same way, insurgents are trying to predict what other insurgents are trying to do, struggling for this media attention. What they would want to do would be to launch attacks when other groups aren’t launching attacks. But much in the manner that we get traffic jams, people might be wrong in their predictions and they might hope that a certain road will be clear because the last three days, it’s been clear and then it might turn out not to be clear.
Chris - So you're saying, if we understand the pattern of these processes, then we’ll be in a position to predict them?
Michael - Yes. I think we can at least get a leg up in trying to predict them. I don't think we can ever predict them with complete accuracy. But yes, you can see certain patterns of, let’s say, quiet day, quiet day, heavy day. What happens on the fourth day after you've had a pattern like that for the last three days.
Chris - These people aren’t daft though and now, you've published this paper which you published before Christmas in Nature, wasn’t it? Are they potentially going to read your paper and think, “Well now, we’ll change our tactics.” And then you’ll have to invent a new model.
Michael - I don't think that such a thing would happen. In the same way that seeing that traffic patterns tend to be bursty or that ,somehow, you can have a particular route that is quiet for a long time, and then busy. Even if drivers have all that knowledge, I think it’s not going to undo the fundamental processes of second-guessing one another that leads to the pattern in the first place.
Chris - Is it still intriguing to think that in the same way as you see these fluxes in the way traffic jams build up in different routes around cities that you're seeing this same pattern with insurgency? How should people who are having to deal with insurgency change their practice in order to either minimise the impact of insurgents or to get rid of them?
Michael - Well, there are two things we find, one is this bursty pattern so that you have alternation of quiet periods and then suddenly, lots of attacks very quickly. And then the second thing we find is when you look at the size distribution of the events. That is, how many events do you have in which one person is killed, how many with two, how many with three, et cetera, up to attacks where hundreds of people are killed. What we find is that you get a lot more of the large events than you might have thought without actually looking at the data and the modellings. We’re finding on the one hand, there’s a bursty pattern where you suddenly get lots of events. And then also, large events are relatively common. And what that means is, in terms of emergency planning, that you've got to be ready with a lot of spare capacity because the system can be overloaded quite easily when you suddenly get lots of attacks and some of them are large.
Physicist Neil Johnson, University of Miami, has an interesting mathematical analysis of insurgent and terrorist activity reported in Discover magazine by Andrew Curry (http://discovermagazine.com/2010/jul-aug/07-the-mathematics-of-terrorism). Data from Columbia, Iraq, and Peru, plotting the number of casualties and frequency of attacks, are distributed as a power curve rather than a bell curve: many attacks with few casualties progressing to very few attacks with large numbers of victims. A power curve is also characteristic of stock market behavior and subatomic particles -- in other words, the activity fits a pattern that does not require behavioral information about the particular individuals or groups involved. His analysis posits that one crucial element is the presence of media: attention is more the point than body count. So if the character of the insurgents themselves is influential at all, it is the desire to be the only one making an attack on any particular day. But since there are a lot of decentralized groups who are not reporting to a single command (as would happen in a regular army) there is a clustering phenomenon. This analysis leads more to expectation than precise prediction: one small attack is worth going on alert for more of the same, while really large-scale attacks like 9/11, unlikely to occur in rapid sequence, just become more likely as time passes. The particularly unsettling aspect to the last observation is that, if the model really does describe the overall pattern of attacks as more mechanical than behavioral, predicting large-scale attacks becomes like predicting earthquakes: there is an inevitability about them that is independent of efforts to prevent them. We don't expect to prevent a massive earthquake in California; we focus on trying to survive it as well as possible. ChinaDoll, Wed, 27th Apr 2011