Math Backs Limited Profiling in Airport Screening

Too great a dependence on profiling passengers by ethnicity or nationality is an ineffective way to conduct airport screening to catch terrorists, according to a statistical model for examining rare events.

In fact, the model predicts that limited profiling would be more efficient, according to an article published online this week in The Proceedings of the National Academy of Sciences.

The model is not widely known, said William H. Press, a computational biologist and computer scientist at the University of Texas, Austin, who is the author of the report.

Called square root biased sampling, it is a way to identify significant events that can be recognized when they are noticed but are otherwise lost in a sea of data. He uses it to find specific snippets of DNA in vast seas of genomic information.

One day last summer while passing through an airport, “it hit me,” Dr. Press said. The method could be used to improve the search for terrorists.

At airports, passengers go through the primary screening checkpoint, and some are selected for secondary screening, which includes a baggage search and some kind of body search, like a pat-down.

If a profiling system is used to determine secondary screening, Dr. Press said, a passenger could be pulled aside for having the same last name as someone on a watch list.

Under such a system, Dr. Press asked, how often should someone with a high profile be selected for secondary screening compared with everyone else, and what would be the smallest number of secondary screenings necessary to find a terrorist?

Suppose some people are 100 times more likely than average to be a terrorist.

Should they always be selected for secondary screening?

Should they be pulled aside 100 times as often as people who do not fit any profile?

Or should they be screened as everyone else is, meaning that profiling would not be used?

Using basic calculus, Dr. Press ran the numbers.

The first option is the worst, he said. All efforts are devoted to screening people who are considered most suspect, even though the real terrorist may not match the highest profile. That method misses the terrorist.

The second option is bad because it screens innocent people far too often.

The third option is bad because an opportunity is missed to take a closer look at high-profile passengers.

The mathematical solution says the answer is between the second and third option, Dr. Press said.

To catch the terrorist in the above example, high-profile passengers should be screened 10 times — or the square root of 100 — as often as other people.

That is the best way to find the bell ringer, he said.

If the profile suggests certain people are 10 times more likely than average to be terrorists, they would be screened only three times, or the square root of 10, more than average.

At that level, Dr. Press said, it is fair to ask if you should be profiling at all.

“We have been told that strong profiling will somehow find and siphon off the worst offenders and we’ll be safe,” Dr. Press said. “It’s not true. The math does not support that.”