Published online 2 February 2009 | Nature | doi:10.1038/news.2009.73


Racial profiling methods may be flawed

Screening by ethnicity could be allowing more criminals to slip through the net.

People going through a metal detectorTo search or not to search?Corbis

Singling out individuals for surveillance, investigation or screening on the basis of their race or nationality is probably useless for catching those engaged in or planning criminal acts. That is the conclusion of a computer scientist who has exposed the basic mathematical flaw in this approach to crime prevention.

At first glance, the approach seems logical, despite many people's moral objections. If all previous acts of politically motivated terrorism have been committed by a particular nationality, then doesn't it make sense to focus searches on those groups?

Not necessarily, says William Press of the University of Texas at Austin. Do the maths and you discover that a simple-minded application of these actuarial methods is worthless: all you end up doing is repeatedly picking out the same innocent people1.

"This is a fascinating and exciting finding," says Bernard Harcourt, a professor of law and political science the University of Chicago in Illinois. "It will trigger a much-needed debate."

Worse than chance?

It is hard to know how often actuarial selection is really used. Airport security policy, for example, is a closely guarded secret. The Transportation Security Administration, which oversees airport security in the United States, will admit only that random screening "is a component" of their procedures, says Press. So whether these methods take into account things that security personnel can see, such as behaviour — or apparent ethnicity — is unclear.

Some think that ethnicity is so clearly a key factor in the process of identifying potential criminals that it has become satirized as the new 'crime' of FWM, or Flying While Muslim — the latest variant of the US driving 'offence' called DWB: Driving While Black.

"There's a widespread belief that, either officially or unofficially, this sort of profiling is going on", says Press. "But it is hard to know what the truth is."

Press says that, if 'prior probabilities' of an individual being a potential offender are assigned accurately, and if screening always results in accurate detection (neither of which is guaranteed), then the chance of finding those with true criminal intent is best if the search is conducted from the top down: from the highest to the lowest probability.

But in most situations, that is not possible because you can't test everyone, you can only screen a sample. And if a person is searched and cleared in one instance, that has no bearing on what happens in a subsequent search.

One 'obvious' method is to make the chance of screening proportional to the chance of being an offender. But Press shows that the number of times this approach repeatedly picks out innocent individuals means that offenders are found no more efficiently than in a random search.

Thus, if actuarial screening is being done, says Press, the chances are that it's being done ineffectively.


This doesn't mean that the technique has no benefit. Press calculates that the best strategy picks out high-risk individuals with a bias proportional to the square root of their probability.


That might be feasible if reliable, quantitative actuarial data are gathered in advance and selection is made by computer. But Press feels that, "given how socially fraught the issue is already, and given how weakly profiling helps, we shouldn't do it at all".

These calculations add weight to other arguments against actuarial methods. Harcourt, for instance, says that these approaches may be counterproductive in crime prevention because people are likely to change their behaviour in response. They may foster resentments in targeted populations, for example, and that will bring more innocent people from those populations into contact with the criminal-justice system2. In fact, says Harcourt, these techniques "may actually increase crime, depending on how different populations respond to increased policing".

Given Press's result, Harcourt adds, "there is absolutely no reason left to profile." 

  • References

    1. Press, W. H. Proc. Natl Acad. Sci. USA 106, 1716-1719 (2009).
    2. Harcourt, B. E. Against Prediction: Profiling, Policing, and Punishing in an Actuarial Age (University of Chicago Press, 2006).
Commenting is now closed.