The key isse in Florida v. Harris, which the U.S. Supreme Court heard on Wednesday, is how a judge should determine whether a drug-sniffing dog is reliable enough to provide probable cause for a search. Should he simply accept police assurances that the animal has been properly "trained" or "certified," as the state of Florida and the Obama administration argue? Or should he, as the Florida Supreme Court said, ask for additional evidence, such as the nature of the dog's training and certification, its performance in tests, and its track record in the field? Police and prosecutors are particularly keen to exclude that last category of evidence, insisting, rather counterintuitively, that the record of a dog's hits and misses is irrelevant in evaluating whether its alert indicates a "fair probability" that drugs will be discovered. It is pretty clear why the people seeking to justify dog-triggered searches are afraid of these data: They show that dogs vary widely in their accuracy and often are more likely to be wrong than right.

On Monday I noted the 2011 Chicago Tribune analysis of data from suburban police departments that found vehicle searches justified by a dog's alert failed to turn up drugs or drug paraphernalia 56 percent of the time. A 2006 study by the New South Wales Ombudsman in Australia looked at more than 10,000 searches triggered by dog alerts and discovered that 74 percent of them found no illegal drugs. In other words, as the National Association of Criminal Defense Lawyers (NACDL) and the American Civil Liberties Union (ACLU) note in their Harris brief, "any given alert was almost three times more likely to be a false alert than an accurate one." More-recent data from New South Wales indicate an even higher error rate: 80 percent.

The NACDL and the ACLU also describe a 1984 operation in which Florida state police stopped about 1,330 vehicles at roadblocks and walked dogs around them. If one dog alerted, another was brought in, and vehicles were searched only if both dogs indicated the presence of illegal drugs. That happened 28 times, but those searches yielded just one drug arrest. "Despite the requirement that two dogs alert before a search," the NACDL/ACLU brief notes, "police found illegal narcotics sufficient to justify an arrest in only 4% of cars searched," meaning "the likelihood of a false alert was approximately 96%." With impressive chutzpah, Florida's lawyers claim "this example at most shows a 'false' alert rate under 2%." You see what they did there? Twenty-seven false alerts divided by 1,330 cars equals 2 percent. But the relevant question is not the percentage of vehicle stops that resulted in a false alert; it is the percentage of alerts (in this case, double alerts) that turned out to be false. For purposes of probable cause, we want to know the likelihood that a search triggered by a dog's alert will find drugs. In this case it was 4 percent—not, as Florida's tricky calculation implies, 98 percent. 

What is going on when a dog alerts and no drugs are found? Police and prosecutors want us to believe these are not really false alerts at all: Either the drugs were so cleverly hidden that the cops could not find them, or the dog detected otherwise imperceptible traces left by drugs that were recently in the vehicle or by contact with drug users' hands (which is what the prosecution claimed in Harris). But these explanations are completely unfalsifiable and amount to nothing but self-serving speculation. Furthermore, the "residual odor" excuse weakens rather than strengthens the case for probable cause, since the traces could have been left by someone other than the driver: a passenger, a previous owner, even (as the defendant's lawyers in Harris suggest) a cash-strapped addict who tried to open the vehicle's door while looking for stuff to steal. In addition to residual odors, dogs can be confused by the smells of food, other dogs, and legal chemicals they have been trained to associate with drugs. But perhaps the biggest source of error is the cues that dogs pick up from their handlers, whose expectations can influence the animals' behavior.

A 2011 study led by University of California at Davis neurologist Lisa Lit shows how powerful this effect can be. Lit and her colleagues had 18 handlers walk their police dogs through four rooms where they were told drug or explosive scents might be hidden but where in fact there were no target substances to be found. Each team went through each room twice, for a total of 144 sweeps, and generated 225 false alerts. The alerts were especially likely near markers that the handlers were told indicated the presence of scents—even more likely than at unmarked locations where the researchers had hidden Slim Jims and new tennis balls. "Human more than dog influences affected alert locations," Lit et al. concluded. "This confirms that handler beliefs affect outcomes of scent detection dog deployments."

In a new Huffington Post column, Radley Balko describes a couple of incidents that further illustrate this point:

A couple months ago, a U.S. Customs dog trained in drug detection somehow managed to find a package containing counterfeit passports. While dogs can be trained to detect drugs or explosives, they can't tell a fake passport from the real thing—no dog is that good. Rather, what likely happened is that her handler noticed something suspicious about the package. The dog picked up on her handler's body language, then alerted to please her handler.

Earlier this year, a HuffPost review of a police dog team with the Illinois state police found an instance in which the dog alerted to a trunk full of illegal (untaxed) cigarettes. Even the smartest of police dogs can't determine with a sniff whether a trunk-load of Marlboros was purchased in or out of state. (Not to mention that the dog wasn't trained to detect tobacco.) The dog likely was merely reflecting whatever suspicions its handler had about the driver.

A dog's sensitivity to its handler's cues (whether conscious or subconscious) means that a suspicious cop who wants to search a vehicle (as in Harris) or a home (as in Florida v. Jardines, the other canine case the Supreme Court heard on Wednesday) can essentially get permission from a dog, regardless of what it actually smells. As Balko observes, "drug dogs have become little more than a way of converting...hunches into probable cause." In the case of a home search, the cop will have to take the additional step of obtaining a warrant, but if courts are to treat an alert by a supposedly well-trained dog as the equivalent of probable cause, that is a mere formality. The upshot is that a police officer armed with an obedient dog can search wherever he wants.