Scent dogs not quite as accurate as believed.

They may be reading their handler more than their nose. [Link]

But don't blame the dogs; their noses work fine. In fact, the Pentagon’s Defense Advanced Research Projects Agency recently conceded, after 12 years and millions of dollars of research, that the canine snout, fine-tuned by millions of years of evolution, is still far more sensitive and reliable than any technology man has been able to muster when it comes to detecting explosives in places such as Iraq and Afghanistan.

The problem is our confusion about when dogs are picking up a scent and when they are responding to cues from their handlers. The Economist's "Babbage" blog summarizes a recent study led by Lisa Lit, a neurologist (and former dog handler) at the University of California-Davis, that demonstrates the startling consequences of that confusion:

[Researchers] asked 18 professional dog handlers and their mutts to complete two sets of four brief searches. Thirteen of those who participated worked in drug detection, three in explosives detection, and two worked in both. The dogs had been trained to use one of two signals to indicate to their handlers that they had detected something. Some would bark, others would sit.

The experimental searches took places in the rooms of a church, and each team of dog and human had five minutes allocated to each of the eight searches. Before the searches, the handlers were informed that some of the search areas might contain up to three target scents, and also that in two cases those scents would be marked by pieces of red paper.

What the handlers were not told was that two of the targets contained decoy scents, in the form of unwrapped, hidden sausages, to encourage the dogs' interest in a false location. Moreover, none of the search areas contained the scents of either drugs or explosives. Any "detections" made by the teams thus had to be false. Recorders, who were blind to the study, noted where handlers indicated that their dogs had raised alerts.

The results? Dog/handler teams correctly completed a search with no alerts in just 21 of the 144 walk-throughs. The other 123 searches produced an astounding 225 alerts, every one of them false. Even more interesting, the search points designed to trick the handlers (marked by the red slips of paper) were about twice as likely to trigger false alerts as the search points designed to trick the dogs (by luring them with sausages). This phenomenon is known as the "Clever Hans effect," after a horse that won fame in the early 1900s by stomping out the answers to simply arithmetic questions with his hoof. Hans was indeed clever, but he couldn't do math. Instead he was reading subtle, unintentional cues from the audience and his trainer, who would tense up as Hans began to click his hoof, then relax once Hans hit the answer.