In What the Nose Knows, as well as here on FirstNerve, I’ve questioned the conventional wisdom about the dog’s sense of smell, namely that it is amazingly better than our own. Recent evidence suggests that the human nose, in terms of its sensitivity as an odor detector, is quite competitive with the canine nose. The dog’s ease in odor tracking may have more to do with differences in sniffing behavior, nostril design and the amount of brain devoted to analyzing olfactory information.
Still, the conventional narrative gets a boost with every report of dogs smelling termites, bedbugs, abnormal glucose levels in diabetics, bowel cancer, ovarian cancer, and so on. Just because a trained scent dog is handy in a pilot study (“does disease X have a smell?”) doesn’t prove that dogs alone are able to make this discrimination. And unless tight controls are built into the experimental design, we still have to rule out that subtle, unintended cues from the dog’s handler are giving us a false result.
This was brought home by a new study that examined the role of human handlers in the response of working scent dogs. Conducted by Lisa Lit and her colleagues at the University of California at Davis, the experiment looked at canine scent-tracking behavior without using any scent at all. It’s a trans-species examination of what social psychologists call “demand characteristics”, i.e., cues that study participants use to glean the aim of an experiment and behave accordingly.
In this case, 18 dog-plus-handler teams—all trained, certified, and experienced in the detection of drugs or explosives—search four rooms in a church for contraband. In view of the teams, an experimenter set down a metal box of gunpowder or marijuana samples. The samples, however, were never opened; this was a ruse to make the room searches believable.
Each dog/handler team searched the four rooms twice. An observer noted when and where the dog signaled an “alert.” Of course the correct response in every trial was “no alert,” as there were not drugs or explosives present. What were present, on some trials, were decoys: a red sheet of paper (for the handlers to notice) or a new tennis ball and a couple of Slim Jims (for the dog to notice), or both. Lit and her team were interested in how alerts by the scent dog (i.e., false alarms) were distributed across the experimental conditions.
The great majority of false alarms happened in rooms with a red sheet of paper; this included trials with and without a scent decoy. To Lit and her team,
this suggests that human influence on handler beliefs affects alerts to a greater degree than dog influence on handler beliefs.Lit et al., consider and dismiss the possibility that handlers were calling alerts in the absence of corresponding behavior from the dog. Instead, they suspect that the dogs were responding to subtle cues from the handler, who in turn was influenced by the apparent location cue of the red paper. This would be an instance of the Clever Hans Effect. (Dig up your Psych 1 notes, people.)
Even without formal training, dogs respond to human cues such as pointing, nodding, head turning and gazing. The mental pull of these cues is powerful, to the point that a companion dog looking for food will ignore a bowl full of food and head to an empty bowl if directed there by his owner. Highly trained scent-detection dogs don’t fall for obvious distractions; clearly, however, they are not immune from subtle, even unintended, influence from their handlers.
Scent-tracking dogs have proven useful in search and rescue missions, detection of contraband, and tracking of criminal suspects. Whether the results of so-called scent lineups conducted by police dogs should be admitted as courtroom evidence is another question altogether. Radley Balko wrote about a recent case in Texas that brought these issues to the forefront.
Research on the olfactory ability of dogs are becoming increasingly sophisticated. While some results, like Lit’s, raise important cautions, another new study adds to the positive side of the ledger. It involved a rigorous and well-designed assessment of the dog’s ability to distinguish identical human twins (aka monozyogtic or MZ twins) by smell. Previous studies (in 1955, 1988, 1990 and 2006) have been a mixed bag. It has been claimed that MZ twins can be distinguished, cannot be distinguished, cannot be distinguished when they share a similar diet, and can be distinguished by some dogs but only if the twins don’t live together.
The new study, by researchers in the Czech Republic, starts on a strong note by using a single breed of dog with similar levels of training, namely ten German Shepherds, each a trained and proven scent-detection animal employed by the Czech Republic Police. The key body scents were provided by pairs of kids living together: two sets of MZ twins and two sets of DZ twins. Their MZ/DZ status was confirmed by DNA testing. BO was collected according to Czech Police forensic protocols (cotton pads in glass jars) and presented to the dogs in a seven-jar “lineup.” The dog signaled a scent match by lying down next to the jar that smelled like the target his handler gave him at the start of each trial. Various combinations of target and lineup scents were used; on trials when the target scent was not in the lineup, the correct response for the dog was to not lie down.
Remarkably, every dog made the correct judgment in every trial (10 dogs, 60 trials each). It seems the mixed results of earlier studies were due to variable skill levels among the dogs tested. Still, not one single incorrect response in 600 trials? (Experimental data with zero statistical variance is, uh, a little unusual.) On the other hand, if it’s this easy for dogs, I bet that humans are able to smell the difference between MZ twins living together.
The studies discussed here are “Handler beliefs affect scent detection dog outcomes.” by Lisa Lit, Julie B. Schweitzer, & Anita M. Oberbauer, published in Animal Cognition 14:387-394, 2011, and “Dogs discriminate identical twins,” by Ludvik Pinc, Luděk Bartoš, Alice Reslová, & Radim Kotrba, published in PLoS One, 6(6):e20704, 2011.