Photo by Doug Page

The short answer is no.

William H. Press, a University of Texas computer scientist began by asking, if you can generate accurate statistical profiles of wrong-doers, such as terrorists, how you should use these profiles in a resource-limited process to select individuals for secondary screening. For example, should you expend your resources on just the highest-scoring profiles, as the TSA does now?

“Mathematically, the answer turns out to be no, for two reasons,” Press said. One, that strategy tends to re-screen the same innocent people over and over again at checkpoints. And two, it provides sanctuary for terrorists who happen to have lower-scoring profiles.

“Strong profiling is generally no more effective at catching a potential terrorist than uniform random sampling of the whole traveling population,” he said.

In place of the institutionalized inefficiency practiced by TSA screening protocols, Press proposes what he calls square root biased sampling.

It works like this. Say that someone from a profiled group (Group P) is 16 times more likely to be a terrorist than someone from the average group (Group A). Using square root bias, someone from Group P should be screened only four times more often than people from Group A, four being the square root of 16. This reduces the number of Group P people subjected to repeated screenings, but it still screens people in Group P more than those from Group A.

Press told Homeland1.com that square root biased sampling is the golden mean between excess use of profiling and no profiling at all. Ideally, where the number of individuals selected for secondary screening is limited by fixed resources, square root biased sampling puts every terrorist on average maximally at risk of being selected and possibly caught soonest.

Of course, the main issue, not addressed in the paper, is whether it is possible to get statistically meaningful profiles in the first place, Press said.

But, assuming that it’s possible, Press foresees a computer program that could instantly calculate the risk factor for each individual who presents at a checkpoint. The program could be based on known factors, from ticketing information and elsewhere, plus factors selected on a touch screen by TSA personnel. To implement square root biased sampling, the computer would also instantly toss the dice, as it were, and signal either Pass or Select for secondary screening.

“The practical effect would be that screenings are distributed over a broader segment of the traveling population, but that high-profile individuals would still be screened somewhat more than average, the mathematically optimal strategy, given the assumptions,” Press said.

The Press paper is available free on the Proceedings of the National Academy of Sciences Web site at www.pnas.org/content/106/6/1716.full.pdf+html.