Racial profiling inefficient as security tool, study suggests

Relying strongly on ethnic or nationality profiling when trying to spot criminals is no more efficient than random sampling, according to calculations made by a U.S. researcher.

Relying strongly on ethnic or nationality profiling when trying to spot criminals is no more efficient than random sampling, according to calculations made by a U.S. researcher.

But using profiling a little bit is more efficient than not using it at all in theoretical cases where people of certain ethnicities or nationalities are more likely to commit certain crimes, argues William H. Press in a study published Monday in the Proceedings of the National Academy of Sciences.

The problem is that relying heavily on profiling wastes a lot of resources on repeated screening of the same innocent individuals from the group with the high probability of committing the crime. At the same time. it offers too much protection for people who might commit the crime but don't fit the profile, said Press.

Profiling is a particularly bad method in cases where the difference in the probability of committing the crime is small between people who fit the profile and those who don't, he added.

Press, a professor with the computer science and biological sciences departments at the University of Texas in Austin who also works at Los Alamos National Laboratory, said he was surprised by the results.

Intuitively, he thought it would be more efficient to concentrate on the people most likely to commit the crimes. The fact that it didn't raised some ethical questions, Press said.

"In my mind, that should let us reopen this whole question of whether profiling is a good thing at all," he told, noting that only weak profiling is calculated to be efficient.

"If you're going to use it that weakly, then why don't we look again at what the social costs [of profiling] are?"

Press's study examined security screening based on the idea that a certain ethnic group or nationality has been shown in the past to commit a certain type of crime at a higher rate than average.

Assuming that this "prior probability" reflects the actual probability that someone will commit that type of crime, Press mathematically tested three screening methods to calculate how much screening would need to be done on average to catch a criminal using each of the following methods:

  • Random sampling.
  • Strong profiling, where a certain group is screened in proportion to the "prior probability" of someone from that group committing the crime. For example, if someone from the group is 100 times more likely to commit the crime, members of that group would be screened at 100 times the rate of people not of that group.
  • Square-root biased sampling, which is part way between strong profiling and random sampling. In this type of sampling, people are screened in proportion to the square root of their "prior probability." For example, if someone from the group is 100 times more likely than average to commit the crime, then they would be screened 10 times more than average.

Press found that given the assumptions, square-root biased sampling was the most efficient method.

His calculations assumed that there would be no prior record of who had already been screened, so that some people might end up being screened more than once.

"If you can keep records of who you screen and if you screen them and you decide ... then that's more efficient," he said. "But I argue that's not always practical, especially in a democratic government."