Boalt Hall Law Professor and Visiting AEI Scholar John Yoo writes in a short piece on the AEI website that we should consider using data mining to pursue terrorists. He makes at least two errors: one historical and one statistical.

Discussing the recent vogue for making U.S. law more like Britain's, Yoo writes:

[I]ncreasing detention time or making warrants easier to come by merely extends an old-fashioned approach to catching terrorists. These tools require individualized suspicion and "probable cause"; police must have evidence of criminal activity in hand. Such methods did not prevent 9/11, and stopping terrorists, who may have no criminal record, requires something more.

It's hard to put aside that the vogue for making U.S. law more like Britain's would undo part of what the Revolutionary War was fought for. And Yoo's placement of the phrase "probable cause" in quotes — I hope that's not to suggest that the language of the Fourth Amendment is quaint.

But putting all that aside, Yoo's first error has to do with more-recent history. He argues that traditional investigative methods "did not prevent 9/11." But traditional investigative methods weren't applied to the problem. 

Operatives like Khalid al Midhar — an individual with jihadist connections known to the United States — entered the country, left in June 2000, and returned July 4, 2001 on a visa the United States gave him. As the 9/11 Commission pithily noted, "No one was looking for him." Traditional investigative methods can't be said to have failed when they weren't being used.

Yoo's second error is to believe that data mining can help locate terrorists. Data mining cannot be made useful in counterterrorism: The absence of terrorism patterns means that it is impossible to develop useful algorithms. The corresponding statistical likelihood of false positives would cause the results of a data mining operation to waste the time and energy of investigators while threatening civil liberties. 

Data mining does give a "lift" to marketers' attempts to find people with certain propensities and interests. But the "failure rate" (if the goal is to find new, willing customers) is typically above 95%. This is with hundreds of thousands, or even millions, of patterns to work with. Data mining also helps ferret out credit card fraud — again, using the thousands of instances of this crime that happen each year to develop useful algorithms.

Probability theory teaches that the percentage of false positives a test produces will rise dramatically as the incidence of the sought-after condition drops. If you're searching American society for left-handed people (8–15% of the population) a data mining operation might work pretty well.  If you're searching for the 10, 12, or two terrorists in the United States, an imperfect test will be useless, time-wasting, and thus harmful to the national security mission.

No, the Fourth Amendment is good policy as well as a part of the not-old-fashioned Constitution. It is better to focus investigations, not broaden them. The best way to find wrongdoing is to look where there is probable cause to believe something is afoot.