This article today at CNN reminded me of something I’ve been interested in for a while now — the reliability of sniffer dog use in criminal investigations and law enforcement.

There are two major uses of dogs in this context that are highly problematic: use of dogs for “scent lineups” to identify criminal suspects and use of narcotics detecting dogs. (In most other contexts, I’m a-okay pretty much across the board with dog tracking– I don’t have any problems with using dogs for detecting explosives or for border searches.) While drug scenting dogs are a much more complicated legal problem, relating to the constitutional requirement of probable cause for Fourth Amendment purposes, the use of dogs for scent lineups has a much greater empirical problem, and their use in the courtroom has not been methodologically sound [PDF].

The requirements for a sniffer dog to be certified are not particularly rigorous, and in Texas and Florida especially, it looks as if a lot of innocent people have been put behind bars as a result of unscrupulous practices by dog trainers. A major problem in the use of scent lineups is that, while dogs actually are pretty good at picking out the “correct” scent from a list of possibilities put before them, when provided with a scent line up with no correct matches, the dogs will all too often default to a “closest match” and produce a false positive.

For the use of narcotics dogs, the science isn’t much more comforting. Although you generally won’t get a false conviction because of the use of a drug sniffer dog (either the police find the drugs on you or they don’t), their use amounts to a real infringement on Fourth Amendment protections. Under United States v. Place, brief detention by law enforcement in a public area for the purpose of carrying out a “dog sniff” is not a search for Fourth Amendment purposes. In Illinois v. Caballes, Justice Souter dissented (from which the title of this post is also taken), noting that:

At the heart of both Place and the Court’s opinion today is the proposition that sniffs by a trained dog are sui generis because a reaction by the dog in going alert is a response to nothing but the presence of contraband. . . . Hence, the argument goes, because the sniff can only reveal the presence of items devoid of any legal use, the “sniff does not implicate legitimate privacy interests” and is not to be treated as a search. . . .

Pretty much every circuit — except possibly my home circuit, the 11th, which has not “squarely addressed” the question — treats it as probable cause to undergo a search when a sniffer dog gives a positive hit. The only evidentiary standards are that the handler must testify to the dog being “trained” and “reliable”. (The only exception to this rule that a not-particularly-thorough search turned up: United States v. Trayer, 898 F.2d 805, 808 (D.C. Cir. 1990)(dog giving positive hit alert at sleeper car on train is not in itself probable cause)).

And most worryingly, some circuits only look at the training of the dog, not the handler. This is pretty faulty reasoning, as the trainer is half the equation — they’re the ones who have to interpret what the dog is indicating. See United States v. Outlaw, 134 F.Supp.2d 701, 813 (5th Cir. 2003) (“an alert is simply an interpretation of a change in the dog’s behavior by a human handler”). In addition to bad trainers who aren’t used to the dogs and don’t know when they’re giving an alert, there’s also the possibility of very closely trained dog/handler combos being susceptible to the Clever Hans effect.

Dogs just aren’t perfect. For example, tired dogs do significantly worse than rested dogs — dogs can’t sweat so they have to pant to dissipate heat, and unfortunately you can’t exactly pant and sniff at the same time — so an overworked and overheated dog is going to produce less reliable results. However, even assuming all dogs are as infallible as Rin Tin Tin, a positive alert from a narcotics dog just doesn’t seem to meet the threshold for probable cause.

Using Bayes’ theorem, and granting dubiously high estimates of dog accuracy (98%) at detecting when drugs are present, and assuming .5% of the population possesses drugs on them at any given time, then then the odds of a dog alert correctly indicating the presence of drugs is only 1 in 5.

“If the dog sniffs 10,000 people, 50 (10,000 x .005) will possess drugs. Out of these, the dog will correctly alert to 49 (50 x .98). Of the remaining 9950 people that do not possess drugs, the dog will falsely alert to 2% of this group, resulting in 199 (9950 x .02) false detections.

Out of this population of 10,000, the dog has positively alerted to 248 people, 49 of which are correct detections and 199 are false alerts. Thus, the probability that an individual actually possesses cocaine based on this dog is 49 out of 248, a detection rate of less than 20%.” Robert Bird, An Examination of the Training and Reliability of the Narcotics Detection Dog, 85 Ky. L.J. 405 (1996).

That doesn’t sound like probable cause to me.

-Susan