Detection Dogs: Are Dogs and Handlers Biased? (Part 2)
Dogs can detect incredibly faint odors. But what if their human handlers give them the wrong cues? In Part 2 of this series, the Dog Trainer explores further complexities in the work of detection dogs.
Page 2 of 2
Did an Artificial Environment Affect the Dogs’ Performance?
The Working Group listed a number of factors that they claim cause failures in an artificial environment like that of an experiment – “extreme expectations” and “pressures associated with blind testing,” for two. I guess those objections are plausible enough, but on the other hand I can’t help but notice that the Working Group is also arguing that the experimental environment wasn’t tightly enough controlled. And if we’re talking about the pressures of testing in an artificial environment, well, the more tightly controlled the experiment is, the more artificial the environment.
How well testing under artificial conditions replicates and predicts real-world results is a valid question to ask about Lit’s experiment as well as about any other. And, like any study, Lit’s has its flaws. But isn’t it a shame that the Working Group’s whole focus was on debunking it, and not on inviting detection dog trainers and handlers to ask themselves some questions about what they might be doing wrong?
Roger Abrantes’ Study
And now for the study by Roger Abrantes, Abrantes is well known in dog training circles; he has a doctorate in evolutionary biology and he writes and lectures on dog behavior and training. His study was published jointly by the for-profit educational firm with which he is affiliated, and the Portuguese Military Police. It wasn’t published in a peer-reviewed journal, which means that no independent experts have looked it over. And I’m no expert in study design, so I could easily have missed both strengths and weaknesses.
Okay, got that out of the way. Abrantes and his team wrote that they tried to make their testing situation as realistic as possible. Their procedure resembled the original study, except that in one of the five testing situations the scent of drugs or explosives was actually present. Also, Abrantes and his team told the dog handlers that they suspected scent was present. Remember that Lit’s team told the handlers that scent was present, even though it wasn’t. So in Abrantes’s experiment the handlers were less biased – at first blush, a more realistic scenario. The result? They didn’t find that handler expectations significantly affected the number of false alerts.
The Degree of Handler Bias Makes a Difference
This result makes sense, given that the handlers’ expectations of finding target scents were less strong. But in real life a handler’s level of suspicion that drugs or explosives are present must vary – sometimes minimal, sometimes extremely high. And it’s an unfortunate fact that sometimes high suspicion is based on prejudice rather than evidence. As I see it, the two studies combined suggest it’s important for handlers to be hyperaware of their own biases, especially where those are strong. An erroneous arrest is pretty high stakes. Incidentally, even in Abrantes’s low-bias study there were 59 false alerts.
Rewarding Correct Non-Alerts
Abrantes also points out that the dog who doesn’t alert when no drugs are present has performed just as perfectly as the dog who does alert when drugs are present. But the training of detection dogs usually focuses on rewarding them for correct alerts much more than on rewarding them for correct non-alerts. This means that dogs themselves may be biased – not about the suspect, but about what behavior their handler wants to see. By the way, we trainers often get on our clients’ case when you don’t notice and reward polite non-behaviors, such as not barking when the doorbell rings. But you’re in exalted company, because detection-dog trainers make exactly the same mistake.
Implications of Detection Dog Error
I promised I’d step off the Dog Trainer patch to talk about the implications of training mistakes made with detection dogs. In 2013, Leslie Shoebotham published an article in the Loyola Journal of Public Interest Law entitled “Off the Fourth Amendment Leash.” Shoebotham argues that law enforcement officials have a couple of incentives to use unreliable, or marginally reliable, detection dogs. (Before you get angry with me: Human beings, like all animals, respond to incentives. Law enforcement officers are just as human as anybody else, for good or ill.)
One important incentive is forfeiture statutes, laws that allow the police to seize cash and other assets found associated with a drug case. These assets are often used to beef up local police budgets. And sometimes an asset – cash, in particular – is exactly what a drug-sniffing dog alerts to. If the handler’s expectations have led to an erroneous alert, then we’re talking about a potentially innocent person not only being arrested but also losing their rightful property.
You can read Shoebotham’s article and judge the arguments for yourself – it’s well worth plowing through the legalese. And be glad that if you accidentally taught your dog to, say, bark for attention, the worst consequence is likely to be some unwelcome noise.
That’s plenty for this week, isn’t it? Stop by and see me on Facebook, or write to me at firstname.lastname@example.org. I welcome your comments and suggestions, and though I can’t reply individually, I may use them as the basis for future episodes. Thanks for reading!