Dog Days of Drug Detection

Drug dog and trainer
By Tyler Jones

Dogs trained to detect the presence of controlled substances are a ubiquitous and highly-trusted tool of law enforcement. At first glance, this trust seems well placed. We know that dogs possess an incredibly powerful olfactory sense that is broad enough to perceive a wide variety of substances, yet sensitive enough to detect even quantities too small for formal lab testing. Further, these natural abilities can be honed through stringent training programs, such as the U.S. Customs and Border Protection Agency’s Canine Program, which touts a 95% detection rate in lab settings. This evidence of accuracy and reliability is so trusted that courts have repeatedly ruled alerts by dogs to be per se probable cause for search and seizure.

This confidence, however, is misplaced. Despite an impressive collection of lab-based evidence in its favor, field studies and independent empirical experiments have repeatedly challenged the idea that an alert by a drug dog is a reliable indicator of illicit drugs. A two-year census of drug dog usage in New South Wales, which analyzed over 10,000 alerts and subsequent searches by police, showed that law enforcement was only able to find illicit drugs 26% of the time. This means that three out of four alerts by drug dogs didn’t produce any evidence of drugs. Similarly, in the United States, an 11th Circuit Court case that examined the use of drug dogs at random vehicular checkpoints found that, out of twenty-eight alerts by dogs, only one yielded illicit drugs. This is an accuracy rate of only 4%.

In addition to these field studies, laboratory studies conducted by independent organizations have brought the reliability of drug dogs into question. One of the shortcomings of many training organizations’ tests is that the handlers aren’t “blind” (i.e., the handlers know where the drugs are). This mirrors real-world conditions, in which the handler is nearly always aware of the circumstances that generated need for a drug dog. The handler’s observations and biases have the ability to unconsciously affect the process of detection. Many organizations don’t take this into account when testing drug dogs. To quantify this potential error, researchers from UC Davis created a deceptive experiment. They had dogs search four locations for drugs and told the handlers that locations with drugs were marked with a red mark. The catch? There were no drugs in any of the locations. The results were breathtaking. Out of 144 searches performed, 123 had at least one alert for drugs (that’s an 85% false positive rate!). Further, the dogs were twice as likely to positively identify areas that were marked as containing drugs. Clearly, handlers have a huge impact on their dogs’ decisions to alert, yet there are currently no requirements to blind handlers when they use their dogs to test for the presence of illicit drugs.

The inaccuracy of drug dogs has real world implications. Life-disrupting events, such as cash seizures, car impounds, searches, and arrests, are all predicated on the supposed efficacy of drug dogs. Yet courts and policy makers have repeatedly dropped the ball when evaluating it. In the most recent Supreme Court case to examine drug dog reliability (Florida v. Harris), the Court explicitly acknowledged practical evidence that drug dogs lack reliability, yet discounted it in favor of purely lab-based data from dog certification agencies. The Court’s reasoning ignored law enforcement’s incentive to use inaccurate dogs and examined the wrong form of accuracy (sensitivity instead of positive predictive value). The Justices’ decision demonstrates the danger of blindly relying on lab results instead of critically examining methods and statistics.

Drug dogs, like DNA testing, have long been perceived as accurate and innocuous law enforcement tools. But, as with DNA testing, the axiom of drug-dog accuracy is crumbling under the weight of clinical and field tests. Courts, themselves extensions of the law enforcement system, are naturally disposed to discount this mounting evidence until it becomes overwhelming. They should not.

92 queries in 0.724 seconds