Tuesday, 20 May 2008

applying the broken windows policy to terrorism

Often the most fascinating part of learning about science is the conclusions scientists (and others) draw from the evidence. Aaron Clauset, who has spent three years at the Santa Fe Institute modeling statistics of terrorism, has found that size and frequency of attack follow a power law.
the statistical distribution fits severe events like 9/11 to the same curve as more common but less severe ones that kill a dozen or so people. the pattern suggests that such rare and large events are not outliers, as was previously thought, but are somehow interconnected with the smaller attacks. the authors claim that if an underlying connection exists, then taking measures to discourage small-scale attacks might also prevent severe ones.
Hat tip 3QD

As I've said before, power laws can emerge automatically from random connections between nodes; an explanation involving a power law is often a claim that no explanation is required--that the behavior is to be expected, like an equilibrium condition. It is divergence from power laws (in situations where we might expect them) that would require explanation. But of course the point of power laws is that they indicate a systematic regularity in phenomena that might otherwise be assumed to be unrelated--for example, large and small terrorist attacks. The final sentence in the above quote is the fascinating one: that not only is there a relationship between small and large attacks, it is a strong enough one that decreasing the number of small attacks should also decrease the number of large attacks.

A professor of mine once noted that there is a strong inverse correlation between the number of mules (per capita) and professorial pay within the region (that is, if one area has more mules, we would expect professors there to be paid less). Yet there is no reason to think that one causes the other--that is, killing mules will not increase professorial pay. In fact, a common cause is responsible for both: the more agricultural a state, the more mules there will be, and the lower professorial pay will be. The more urban a region, the fewer mules and the higher the pay.

It certainly seems right that large-scale terrorism should be more closely related to small-scale terrorism than are mules to payscales, but mere similarity is no guarantee.

The idea that large terrorist attacks can be prevented by dealing sharply with small terrorist attacks is an intriguing one, and it has strong similarities to a crime-reduction policy that has become popular in major US cities in recent years. The Broken Windows policy begins from the premise that most crimes are crimes of opportunity, and opportunities are more apparent in "bad neighborhoods"--those with broken windows. To reduce subway crime, then-Mayor Giuliani of New York instituted the policy of cleaning graffiti from trains on a daily basis. No train left the yard with graffiti on it. Indeed, the policy worked: graffiti on trains dropped dramatically. As did littering and turnstile jumping (although the latter probably had more to do with the complementary policy of daisy-chaining jumpers together and processing their tickets en masse with a roving paddy-wagon).

But in the larger scheme, Broken Windows has had some unfortunate side effects: many more people are incarcerated for small crimes, increasing the number of three-strikes convictions, straining the penal system, and inadvertently (one hopes) targeting the homeless and mentally ill (whose crimes are usually of the nuisance variety).

Could fighting Broken Windows Terrorism have similar unexpected consequences?


Aaron said...

I'm not much of a fan of the broken windows theory of crime, mainly because it seems too simple to correctly capture the underlying processes that encourage criminal behavior. In the domain of terrorism, the trick to changing the fundamental character of the severity of terrorist attacks, and thus to preventing more large-scale events, will be to figure out what the underlying cause is that explains the power-law form.

In other words, take your example with mules. A naive policy maker might look at the anti-correlation and decide to slaughter all the mules in his province, and thus, presumably, increase the professional pay in the area. The problem with this approach is that it doesn't understand the root cause of the correlation. The problem with the broken windows theory is that it assumes a particular causal relationship, when, in fact, this question isn't settled, either for criminal activity or for terrorism.

Zachary Miner said...

The additional problem with using broken windows policies is that one of the reasons it succeeded is not because stopping people committing little crimes stops *other* people from committing big crimes. Rather, two facts are at work here: 1) a small percentage of people commit most of the crime, 2) most people who commit big crimes also commit small crimes. So, if you arrest lots of people for small crimes, you're bound to incarcerate a bunch of those high-rate offenders who have also been doing most of the big crimes. Problem is, of course, not everyone who commits small crimes is a high-rate offender ... like, the homeless person whose "crime" is being mentally ill in a society which has no especially good way of helping the mentally ill.

It would be an interesting question, then, to see if terrorists who commit small acts also commit big acts, because if they don't then the drive of the policy would be in jeopardy. It wouldn't work if, for example, a canny terrorist organization decided to "save up" its members who had had no police contact in order to allow them to slip under the radar when it came time to commit a big act, rather than "train up" those same members through a series of smaller acts.

Also, a study of plane hijackings showed that terrorists reduced their hijacking attempts not because of increased penalties coming as a result their acts, but rather because of a greater likelihood that they will not succeed in carrying out the act at all. As in, increased use of metal detectors put people off of attempting airborne hijackings, because it meant that they might not get into the air at all. There are some methodological questions which go along with this study, but it was interesting anyway (Dugan, LaFree, and Piquero 2005).

Isaac said...

Aaron: You're right that the causal theory behind Broken Windows is by no means confirmed by its apparent effectiveness at reducing crime rates. Still, as a pragmatist I am less worried about the lack of understanding than with the risk of deleterious unintended consequences.

Zac: So Broken Windows is structurally similar to regressive taxation--it disproportionately targets a large population in order to ensure the capture of a sub-population.

I would bet that terrorism does follow the escalation pattern of most criminal behavior, and so Broken Windows actually should be effective--until terrorists adjust in one of the ways you suggest (or should I say "envision"?)

It should probably not surprise anyone that likelihood of punishment is much more significant predictor of behavior than severity of punishment. Thanks for the pointer to that study.

Zachary Miner said...


>I would bet that terrorism does follow the escalation pattern of most criminal behavior

I agree with you that some aspects of terrorist behavior will follow the escalation pattern. As in, people will start having initial feelings of anger, followed by increased feelings fueled by becoming involved with a group of some sort, etc. However, in this case, the issue is whether acts of terrorism will be detected, labeled as terrorism, and therefore bring the person committing the act under the new policy. This policy would work better against someone like the Unabomber - who attacked serially - rather than the stereotypical Islamic extremist whose first and final "terrorist" act is a suicide attack.

>It should probably not surprise anyone that likelihood of punishment is much more significant predictor of behavior than severity of punishment.

There's a lot of research in criminology about this, focusing on the interrelations among certainty, severity and celerity (swiftness) of punishment, under the larger umbrella of Rational Choice Theory. The only problem with this is that - in typical hilarious fashion - some researchers have found that criminals have NO IDEA how likely it is that they will be caught, or how high their punishment is likely to be. So, sure, if you objectively increase their assessment of the certainty of being caught, they will commit fewer crimes. But how do you do that, when it appears that their initial perceptions are faulty? Also, wouldn't it be more effective to just increase the perceived certainty of arrest, rather than the actual certainty? (Answer: Yes, it is ... but again, how do you do that?)