Arguments over racial profiling at the airport security line typically turn around the assumption that such screening, at least to some extent, works. The idea may be unsavory, but it sounds logical: If we target people with a higher probability of being terrorists — whether they have Saudi passports, beards or headscarves — we’d have a better chance of catching real terrorists in the process.
The question becomes one of morals. Is this the right thing to do? Does the societal benefit (catching more terrorists) outweigh the cost (compromising our ethics)?
William Press, a professor of computer science and integrative biology at the University of Texas at Austin, now realizes we don’t have to weigh this dilemma at all. Racial profiling, he has concluded, simply doesn’t work. Never mind how you feel about it. The math doesn’t add up.
Plucking out of line most of the vaguely Middle Eastern-looking men at the airport for heightened screening is no more effective at catching terrorists than randomly sampling everyone. It may even be less effective.
Press stumbled across this counterintuitive concept — sometimes the best way to find something is not to weight it by probability — in the unrelated context of computational biology. The parallels to airport security struck him when a friend mentioned he was constantly being pulled out of line at the airport.
“He’s not on any do-not-fly list, and it occurred to me it was exactly this phenomenon,” Press said. “Either explicitly or implicitly, there was some kind of profiling going on, and the same innocent individual was being screened over and over again. That draws resources away from the screening that would find the bad guy. I realized those were basically the same problems.”
Racial profiling, in other words, doesn’t work because it devotes heightened resources to innocent people — and then devotes those resources to them repeatedly even after they’ve been cleared as innocent the first time. The actual terrorists, meanwhile, may sneak through while Transportation Security Administration agents are focusing their limited attention on the wrong passengers.
Miller-McCune’s Washington correspondent Emily Badger follows the ideas informing, explaining and influencing government, from the local think tank circuit to academic research that shapes D.C. policy from afar.
Press tested the theory in a series of probability equations (the ambitious can check his math here and here).
“I was flabbergasted,” he said.
Sampling based on profiling is mathematically no more effective than uniform random sampling. The optimal equation, rather, turns out to be something called “square-root sampling,” a compromise between the other two methods.
“Crudely,” Press writes of his findings in the journal Significance, if certain people are “nine times as likely to be the terrorist, we pull out only three times as many of them for special checks. Surprisingly, and bizarrely, this turns out to be the most efficient way of catching the terrorist.”
This model minimizes the overemphasis on people like Press’ friend, creating the best trade-off between over- and under-screening profiled passengers.
“It’s just a little piece of math,” Press said, “that somehow has escaped the textbooks because you don’t need it very often.”
Square-root sampling, though, still represents a kind of profiling, and, Press adds, not one that could be realistically implemented at airports today. Square-root sampling only works if the profile probabilities are accurate in the first place — if we are able to say with mathematical certainty that some types of people are “nine times as likely to be the terrorist” compared to others. TSA agents in a crowded holiday terminal making snap judgments about facial hair would be far from this standard.
“The nice thing about uniform sampling is there’s nothing to be inaccurate about, you don’t need any data, it never can be worse than you expect,” Press said. “As soon as you use profile probabilities, if the profile probabilities are just wrong, then the strong profiling just does worse than the random sampling.”
Press would like to see policymakers get behind this conclusion, even if they can’t follow along with his algorithms.
“They don’t have to look at the details of the math,” he said. “I think that, when you come down to it, there’s an alternative that avoids the political minefield and is consistent with democratic values and is relatively easy to do.”
Randomly sample everyone.
