Can statistics help catch terrorists?

By Ruth Alexander
BBC's More or Less

In the wake of the alleged attempt to bomb a Detroit-bound plane, Britain could step up airport security with targeted passenger profiling. But how effective is it?

The issue of passenger profiling can be controversial

Standing in a long queue at airport security with your fractious family, you might think it is pretty obvious you do not look like a terrorist.

But is your journey through security about to get easier - and your flight safer?

The government says it is considering screening passengers to decide which should undergo more rigorous security checks. But the practice, known as passenger profiling, is controversial.

People behaving suspiciously or with an unusual travel pattern might find themselves under additional scrutiny but racial or religious factors could also soon form part of the criteria.

But, in weighing up the merits of targeted versus random checks, the government will have to do its maths, because this seemingly straightforward idea is anything but in reality - not just because of fears that passengers will be unfairly discriminated against on grounds of ethnicity or religion, but because it involves some counter-intuitive number work.

For a start, in relying on who your statistics tell you is potentially high risk, your statistics could become unreliable, according to David Spiegelhalter, professor of the public understanding of risk at Cambridge University.

FIND OUT MORE More or Less is on BBC Radio 4 on Fridays at 1330 GMT, and repeated Sunday 2000 GMT Or catch up on the iPlayer

"You're combating an intelligent adversary who could change their strategy in response to what you do," he says.

To make sure your system is not predictable, any profiling strategy would have to be complemented by additional random checks across the whole passenger population.

"Game theory [an area of maths used, among other things, to predict and understand behaviour] shows that in response to an intelligent adversary, it often can be optimal to adopt a level of randomisation.

"There must be an element where your computer says this individual must be examined carefully, whether it's a five-year-old child or a 90-year-old, so any opponent knows they might be examined no matter how little they seem to fit the risk profile."

Unlikely bomber

Hard to imagine a five-year-old or a 90-year-old becoming radicalised.

But if people like that were never checked, a terrorist could see an opportunity.

Umar Farouk Abdulmutallab is accused of trying to bomb an airliner

There is a famous example of an apparently unlikely bomber - a young pregnant Irish woman called Anne-Marie Murphy, who was planning to fly from London to Israel in 1986, recounts Philip Baum, editor of Aviation Security International.

Ms Murphy's boyfriend - who, unknown to her, was a Jordanian militant - had planted explosives on her, hoping she would get on to the flight without arousing suspicion, Baum says.

But in this instance, Baum notes, the plot was stopped because of the intense passenger profiling carried out by Israeli flight security. Ms Murphy did not seem like a typical traveller for that flight.

To make sure innocent people are not used in this way, passengers should be selected for extra checks using a graded scale of risk, according to Prof Spieglhalter.

A computer scientist from the University of Texas, William Press, believes he has done the maths to come up with a good way of doing exactly that.

Additional checks

Suppose you have two people - passenger X and passenger Y - and your brilliant profiling suggests Mr X is 100 times more likely than Mr Y to be a terrorist.

Your stats tell you to pick out Mr X for additional security checks 100 times more often than Mr Y.

But William Press says that would be a mistake.

Security resources are finite, and the threat is varied

He says if you take the square root of their risk it narrows the gap between them - so you check Mr X only 10 times more often than Mr Y.

That means you do not expend all your resources checking only one tiny group of people every time they turn up at an airport - and hardly ever checking people who pose a lower risk, but still a risk.

Basically, you widen your net.

In reality, of course, your intelligence is unlikely to be as brilliant as that example suggests.

Getting good and useful data for profiling is difficult.

But, even if you supposed you had an almost perfectly accurate profiling system, there is another problem - terrorists are very rare and that leads to some counter-intuitive statistics.

If someone is stopped by security for extra checks because they seem to pose a potential risk, they are almost certainly innocent, Prof Spiegelhalter suggests.

"There are more than a 100 million people flying out of UK airports every year and you might be trying to foil one or two terrorist plots," he says.

"Even if you had a profiling system that was 99.99% accurate, there would be 1 in 10,000 errors.

"If you consider that around 100 million people go through UK airports each year, that would mean 10,000 innocent people would trigger the system."

"I would hope that in any training of staff, this is really rammed home.

"If you're not going to cause enormous resentment about this, it's got to be very delicately handled - maths tells you that."

So although statistics could help catch terrorists, to some degree, they do not make it easy, and they probably will not get you through airport security any quicker.