Tuesday, November 20, 2012

Detecting Early Stage Lung Cancer May Become a Clinical Function of Sniffer Dogs

Lung cancer is one of the most deadly cancers, and one that all too often is not detected early enough for treatment to be effective. Although progress is being made, as discussed in the papers below, standard x-rays often fail to detect lung cancer early enough for any therapy to be successful. The fact that dogs are fairly good at detecting a significant proportion of lung cancers thus makes their use in a clinical setting a real possibility. As one of the studies discussed below concludes, canine cancer detection is “virtually on the verge of respectability.” Whether or when dogs will be used to screen us for lung cancer remains uncertain, but recent research indicates that the odds are going up in favor of this canine skill having practical application in the future.

I had speculated, as had Mary Elizabeth Thurston even earlier (1996), that remoteness might drive the use of dogs in cancer detection into clinical reality. The argument was made that dogs could become a tool of Doctors without Borders or doctors who fly to remote Arctic villages. Now it appears that the possibility of early detection by canines of cancers that are particularly difficult to find at asymptomatic stages may be the factor that brings dogs into actual diagnostic settings. This blog describes several articles, but the pictures were supplied by Dr. Tadeusz Jezierski of the Polish Academy of Sciences.

Dogs Distinguish Lung Cancer Patients from Healthy and COPD Patients

Breath Sample Tube
A team in Stuttgart, Germany (Ehmann et al., 2012), used four family dogs (two German shepherds, one Australian shepherd, and one Labrador retriever), two males and two females, all between 2.5 and 3 years old. All were trained to indicate to breath samples of patients with lung cancer by lying on the floor in front of the test tube and resting the muzzle against the tube.Training involved the dogs distinguishing breath samples of healthy volunteers from those of patients with lung cancer. No training was performed to distinguish COPD samples.

The researchers collected 220 breath samples from patients with COPD (50) or lung cancer (60), as well as from healthy individuals (110). Dogs were presented with five probes position in five separate retainers on the floor, rubber caps removed. The position of the probe with lung cancer was determined randomly and was blind to the participants except those who set up the probes.

The dogs did not perform identically, but each performed consistently. The accuracy of a dog’s indication did not favor advanced tumor stages. “The overall sensitivity was 71% and the specificity was 93%....” That is, the dogs identified 71% of the breath samples of those with cancer, and incorrectly found cancer in only 7% of the individuals who did not have cancer. The table, adapted from one included in the study, illustrates how specificity and sensitivity are calculated.

Breath sample without cancer
Breath sample with confirmed LC
Dogs indicating presence of LC
Dogs indicting absence of LC
372/400 = 93%


71/100 = 71%

The researchers concluded that there is “a stable marker (or scent pattern) that is strongly associated with LC [lung cancer] and independent from COPD, but can be reliably discriminated from tobacco smoke, food odors and (potential) drug metabolites.” They argue that sniffer dogs as cancer detectors are “virtually on the verge of respectability.”

What the dogs are identifying, according to the Stuttgart group, remains speculative. They note that 3,481 volatile organic compounds have been described in human breath, mostly in amounts of particles per trillion. Which ones increase because of lung cancer is only beginning to be described. Recent studies have succeeded in distinguishing healthy from cancerous breath by electronic nose technology. In these recent results, tumor stage did not influence the outcome in any of the studies, “implying that exhaled breath profiling has the potential to evolve as a screening test for LC—once specific markers have been identified.” They add that “it is currently difficult to predict when a clinically applicable diagnostic device for breath analysis will be available.”

As to specific chemicals the researchers noted that “Metropolol, Verapamil and Tiotropiumbromide were consistently distributed between LC and COPD patients, but not found in healthy volunteers. Marcumar, Clopidogrel and Ezetimib were present exclusively in COPD patients.”

As to COPD, the researchers note:

“COPD often precedes and accompanies LC in smoking patients…. COPD is characterized by typical lung function deterioration, chronic systemic and local airway inflammation and structural changes in lung parenchyma. It has been shown that the level of exhaled biomarkers is altered in patients with COPD compared to healthy control subjects…. Moreover, since the development of LC is much more frequent in COPD patients than in healthy controls, attention needs to be focused on the subtle differences in exhaled biomarker profiles between LC and COPD.”

Study Compares Dogs and Machines

As noted by Buszewski et al. (2012b), a sweetish, acetone-like breath odor indicates a person has diabetes, while the odor of rotten eggs, caused by organic sulfide and thiol compounds, suggests liver disease. Chemical analysis requires being able to deal with very low concentrations of chemicals exhaled so certain procedures must be used before analysis. The most common method of enriching volatile organic compounds involves solid-phase microextraction followed by analysis by gas chromatography (GC) or GC-mass spectrometry (GC-MS).

Taking Breath Sample
Exhaled air consists of alveolar air, which has been in contact with blood inside the alveoli, and ambient air, retained in the mouth, nose, pharynx, trachea, and bronchi. Alveolar air is strongest as the end of expiration of breath.

A chemical sensory approach, gas chromatography and mass spectrometry (GC-MS) for cancer screening was, in a second paper by the Polish team, Buszewski et al. (2012a), compared with using dogs to detect cancer. The GC-MS approach allows for specificity in the chemicals—volatile organic compounds, or VOCs—that are to be detected, whereas dogs are something of a “black box technology” with a yes/no response, where it is unknown what individual odor or combination of odors dogs are responding to. More specifically, the experiment used an Agilent 7890A gas chromatograph coupled with a Tru-TOF spectrometer. Male German shepherds, 20-22 months old, that had successfully completed a three-phase training in scent lineups were also used. Breath samples were collected from 44 healthy volunteers and 29 patients with lung cancer.

For the cancer scent lineups, one breath odor sample from a patient with lung cancer was placed in a line-up with four samples from healthy volunteers. Dogs had been taught to indicate the lung cancer sample by sitting down in front of it. Positions of odor samples in the line-up were randomly changed for every trial. Handlers accompanying dogs did not know the placement of the cancer sample in the line-up.

In the GC-MS testing, all compounds detect in breath samples were compared with ambient air samples, and only compounds with concentrations at least 10% higher than those in ambient air were reported. A number of compounds were higher, at a statistical level, in the breath of patients with cancer. These were butanal, 2-butanone, ethyl acetate, ethyl benzene, 2-pentanone, 1-propanol, and 2-propanol.

Breath Sample Prepared for Sniffing
As to the dogs, the researchers found:

“The dogs indicated correctly the pattern of breath samples from lung cancer patients with detection sensitivity and specificity of 82.2% and 82.4%, respectively. False positive indications toward healthy controls amounted to 17.8% of trials. The differences between dogs’ indications of cancer samples vs. controls were highly significant….” Thus, Buszewski et al. found lower specificity but higher sensitivity than the Stuttgart group described above.

Analysis of the data suggested that dogs may be particularly effective at detecting ethyl acetate and 2-pentanone. Chemical analysis of breath samples indicates that for ethyl acetate, breath exhaled by healthy persons has 1.12 to 8.22 parts per billion, while for cancer patients the range is 3.98 to 22.89 parts per billion. For 2-pentanone, the healthy range is 1.80 to 4.11 parts per billion, but 3.25 to 8.77 parts per billion for cancer patients. Nevertheless, that dogs are signaling to levels of these two compounds is not established. The researchers conclude that the odor signature that dogs use for discrimination “may be related to some specific qualitative or quantitative olfactory impressions produced by the mixture of VOCs.”

Where is This Going?

In an editorial that appeared in the same issue of the European Respiratory Journal that included the Stuttgart study, McCulloch et al. (2012) consider where such canine studies may be leading. They conclude that “the high accuracy of canine scent detection of lung cancer suggests dogs might, in the future, make some modest contribution to successes in lung cancer screening and detection.”

They note that lung cancer remains the leading cause of cancer death in the U.S., but that early detection remains a challenge. There are promising developments, including “the ongoing National Lung Screening Trail in the USA showing that lung cancer mortality can be reduced by up to 20% with low-dose spiral computed tomography (CT) screening compared to chest radiography…. At first diagnosis, over 75% of patients have advanced stage disease.”

Dr. Jezierski with Court, a Labrador Cancer Sniffer
McCulloch and his co-authors accept that “critics may turn up their nose at the mention of using sniffer dogs,” but they argue that high-quality papers have shown promising results. "The development of an ‘electronic nose’ for cancer detection has been underway for several decades; however, dogs still appear to be ahead in the race and seem to have sniffed their way to the front of the line.” The researchers invite us to imagine a future in “dogs could be used as a noninvasive preliminary diagnostic "screening tool or be used to help reduce false positives and false negatives of existing imaging technologies.”

Training Cancer Sniffers Will Change if Specific VOCs Are Identified

One could divide canine scent detection into two broad categories, those where dogs are trained to identify specific chemicals or groups of chemicals, and those where the dogs recognize and perhaps follow a scent which is sufficiently complicated that it is uncertain what chemicals or combination of chemicals are being recognized by the dogs. Thus, a dog trained to recognize accelerants—an arson dog—will be trained to identify certain combustible substances that may cause fires. Drug and explosive detection dogs are trained to recognize the odor of certain drugs or explosives, though these items may consist of groups of chemicals and it may sometimes be uncertain which one or which combination is of interest to the dogs. Then there are dogs, like cadaver dogs and human tracking dogs, that follow a trail or pursue a scent that may contain hundreds if not thousands of chemicals, and no pretense can be made at identifying exactly what chemicals, or ratio of separate chemicals, are being recognized by the dogs.

Cancer sniffers are, at the moment, in the second category, though the research by the Polish group may indicate that there will come a point where the volatile organic compounds that dogs recognize in the breath of cancer victims can be specified. This raises the possibility that cancer sniffing dogs may ultimately be trained on specific chemicals present in the breath of people with cancer and not present in those free of cancer, which may increase the reliability of the dogs in identifying cancer (and not indicating to the breath samples of people without cancer). Training procedures are simultaneously being refined, as described by Walczak et al. (2012).

The Black Box vs. the Smell-o-Matic

It is also possible that, even if dogs find a clinical function in cancer detection, the existence of this function may be transient since, once cancer VOCs are sufficiently identified, there will come a point when technology will provide accurate methods of measuring those chemicals. When that technology is available in the medical marketplace at a reasonable price, the dogs will no longer be necessary. If the portability of the technology remains a problem, then dogs might continue to be useful in remote areas, as Mary Elizabeth Thurston suggested almost two decades ago.

The black box effect of dogs is also part of why a technological solution may eventually be preferable for certain forensic functions that dogs perform. The presence of drugs inside of a house or a car, suggested by a dog’s indication to an odor it has been trained to detect, does not establish the amount of narcotics present, or even if they are present since the dogs may be detecting a residual odor. In oral argument before the Supreme Court in a case arising from use of a drug-sniffing dog in Florida, Florida v. Jardines, Justice Elena Kagan asked, if the police had a “smell-o-matic machine” that “alerted to the exact same things that a dog alerts to,” whether this could be brought to the front door of a suspected marijuana grow house the same way counsel was arguing could be done with a police dog. There is ongoing research on such technology, some of it overlapping the medical research discussed above, but there will likely be significant differences from a dog if the smell-o-matic is ever used in the field. First, it is likely to give a precise read on each chemical it is capable of detecting. It will not be a black box—an all or nothing indication—such as is given by a dog. Thresholds could be established based on detected levels that would more precisely reflect the probability of drugs being found within certain specified distances, humidity, wind level, air pressure, walls in between, etc., which would make the smell-o-matic more reliable than the dog.

The smell-o-matic might, for instance, be able to determine the amount of cocaine on a $20 bill, and whether its presence there is high enough on the surface to suggest recent contamination, and to distinguish such a bill from one that is innocently tainted from contact days or longer before. Although there is evidence that dogs do not alert unless the contamination is fairly recent, as discussed in a prior blog, a well-publicized study performed by the Miami Herald found that $20 bills supplied by Janet Reno, Jeb Bush and other prominent citizens had significant amounts of cocaine. As discussed in Police and Military Dogs, some courts remain reluctant to assign much significance to a drug dog’s alert in a currency forfeiture action.

The smell-o-matic would presumably be capable of detecting ambient levels of drugs in the atmosphere of an area, which if sufficiently concentrated might explain some residual odor indications by dogs. A recent paper determined that, in the air of Italian cities, cocaine and cannabinoids “are almost ubiquitous,” varying in concentration from a few pictograms to nanograms per cubic meter. The research found that the levels were greater in parts of Italy where drug consumption is thought by law enforcement to be higher than elsewhere. Thus, Cecinato et al. (2012) found that “cocaine ranged from 0.26 ng/m3 (Turin) and 0.21 ng/m3 (Naples) to 0.05 ng/m3 (Verona) and 0.02 ng/m3 (Palermo).” Curiously, part of the purpose of this research was to determine correlations between ambient cocaine levels and tumors, though the researchers state that much more testing is needed before they can reach conclusions in this area.

There is one other major difference between a dog and a smell-o-matic, though presumably this would seldom be a factor in the medical setting. That is, the handler of the dog in the medical setting operates without any expectation that the target odor smelled by the dog indicates cancer or not. Everything is double-blinded to the nth degree. Although this can be done in scent lineups conducted in forensic laboratories, in police field work, particularly at potential crime scenes, an officer uses his senses and experience to anticipate what criminal activity his investigation may uncover. Such expectations, when held by a canine handler, may allow for the possibility of intentionally making the dog alert (see the blog on The Good Wife and Breakfast in Collinsville) or unintentionally cueing the dog to alert. This problem could be eliminated with the smell-o-matic, which, once it is developed, may assure a measurably high reliability. The portability of the dog, as well as its psychological impact on criminals, may still argue for its use even if a small enough smell-o-matic is developed, however. In sum, the smell-o-matic may be less intrusive, because less likely to produce fruitless searches, than a dog. Admittedly, It will take some time for such a forensics tool to be developed and, given budgetary constraints, even more time to come into widespread use.

Gromit, Cancer Sniffer
Thanks to L.E. Papet for providing source materials. Thanks to Tadeusz Jezierski, a contributor to Police and Military Dogs, for sharing a slide presentation of recent research which includes the pictures used here. Also, thanks to the professor and his team for having the wisdom (or at least the sense of humor) to name one of the cancer sniffers Gromit, after Wallace's brilliant and long-suffering companion.

  1. Buszewski, B., Ligor, T., Jezierski, T., Wenda-Piesik, A., Walczak, M., and Rudnicka, J. (2012a). Identification of Volatile Lung Cancer Markers by Gas Chromatography-Mass Spectrometry: Comparison with Discrimination by Canines. Analytical and Bioanalytical Chemistry, 404(1), 141-6.
  2. Buszewski, B., Rudnicka, J., Ligor, T., Walczak, M., Jezierski, T., and Amann, A. (2012b). Analytical and Unconventional Methods of Cancer Detection Using Odor. Trends in Analytical Chemistry, 38, 1-12.
  3. Cecinato, A., Balducci, C., Romagnoli, P., and Perilli, M. (2012). Airborn Psychotropic Substances in Eight Italian Big Cities: Burdens and Behaviors. Environmental Pollution, 171, 140-147.
  4. Ehmann, R., Boedeker, E., Friedrich, U., Sagert, J., Dippon, J., Friedel, G., and Walles, T. (2012). Canine Scent Detection in the Diagnosis of Lung Cancer: Revisiting a Puzzling Phenomenon. European Respiratory Journal, 39, 669-76.
  5. Ensminger, J. (2010). The Cancer Sniffers. In Service and Therapy Dogs in American Society, Chapter 6. Charles C. Thomas, Springfield (summarizing pre-2010 research).
  6. Florida v. Jardines, Docket No. 11-564, U.S. Supreme Court, transcript of oral argument held October 31, 2012.
  7. McCulloch M., Jezierski T., Broffman M., Hubbard A., et al. (2006). Diagnostic Accuracy of Canine Scent Detection in Early- and Late-Stage Lung and Breast Cancers. Integrative Cancer Therapies 5(1), 1-10.
  8. McCulloch, M., Turner, K., and Broffman, M. (2012). Lung Cancer Detection by Canine Scent: Will There Be a Lab in the Lab? European Respiratory Journal, 39, 511-512.
  9. Thurston, M.E. (1996). Lost History of the Canine Race. Andrews and McMeell, Kansas City.
  10. Walczak, M., Jezierski, T., Gorecka-Bruzda, A., Sobczyniska, M., and Ensminger, J. (September 2012). Impact of Individual Training Parameters and Manner of Taking Breath Odor Samples on the Reliability of Canines as Cancer Screeners. Journal of Veterinary Behavior, 7, 283-294.

Wednesday, November 7, 2012

Livestock Guard Dog Exclusion from Dog Bite Liability Complicates Cyclist’s Lawsuit

Livestock guarding dogs are raised with sheep or other livestock, and become so attached to the herd that they protect the herd and its members, sometimes to the death when a predator, such as a mountain lion, is sufficiently dangerous.  They are raised with the herd animals they will protect, and are encouraged to prefer the company of those animals to that of people.  Unlike herding dogs, they are not servants of the shepherd, but rather of the sheep.  Playing with other dogs or people is discouraged, since the idea is for the dog to prefer being with the sheep as a member of the herd.  Obedience training is also discouraged because this increases the bond with the master.  They are generally large breeds capable of taking on predators, including Great Pyranees, Komondors, Tibetan Mastiffs, and others.  It has been argued that war dogs began as sheep guarding dogs. 

Great Pyrenees Puppy Being Raised to Guard Sheep (USDA photo)
Dogs brought up to guard livestock are often shy of people, and can be aggressive towards a person who comes near the herd.  Sometimes this leads to tragedy.

A Dog Attack in Colorado

On July 9, 2008, Renee Legro was in a bicycle race sponsored by the Vail Recreation District, which holds a number of such races every year.  While on a public road in the White River National Forest she was attacked by two predator control dogs owned by Samuel and Cheri Robinson.  The Robinsons had a permit, issued by the U.S. Forest Service, to graze sheep in the area that included the site of the attack.  The Vail Recreation District also had a permit to use the road where Legro was riding.  Legro sustained serious injuries and sued the Robinsons. 

Legro filed a claim under Colorado’s dog bite statute, which provides:

“A person … who suffers serious bodily injury or death from being bitten by a dog while lawfully on public or private property shall be entitled to bring a civil action to recover economic damages against the dog owner regardless of the viciousness or dangerous propensities of the dog or the dog owner's knowledge or lack of knowledge of the dog's viciousness or dangerous propensities.” Colorado Revised Statutes (CRS) 13-21-124

The Robinsons filed a motion for summary judgment, arguing that Legro’s claims were barred by Colorado’s Premises Liability Act, which provides in part:

“In any civil action brought against a landowner by a person who alleges injury occurring while on the real property of another and by reason of the condition of such property, or activities conducted or circumstances existing on such property, the landowner shall be liable only as provided in subsection (3) of this section.”  CRS 13-21-115(2)

Subsection (3) of the Premises Liability Act restricts a landowner’s liability generally to a failure to exercise reasonable care to protect against or warn about dangers of which the landowner actually knew or should have known.

Predator Control Dog Exclusion

The Robinsons also argued that they could not be held liable even under the dog bite statute, CRS 13-21-124, because of a specific exception which provides:

“A dog owner shall not be liable to a person who suffers bodily injury, serious bodily injury, or death from being bitten by the dog: … (f) While the dog is working as a hunting dog, herding dog, farm or ranch dog, or predator control dog on the property of or under the control of the dog’s owner.”

The trial court granted the Robinsons’ motion for summary judgment, in large part because of the predator control dog exclusion of the dog bite statute.  Legro appealed. 

Appellate Decision

There was no factual dispute that the Robinsons’ dogs were predator control dogs.  The appellate court noted that the predator control dog exclusion applies to property owned by the dog owner, but also to property under the dog owner’s control, even if the property is not owned by the dog owner.  The court said that control must “mean, at a minimum, sufficient control over the property such that a dog owner has the right to exclude persons from the property.” 

The trial court had concluded that the Robinsons had a lease on the property where the attack occurred, but the appellate court said that “the record shows that the Robinsons operated under a nonexclusive permit to graze sheep.”  More specifically:

“[T]he attack occurred in the White River National Forest on land where both the Robinsons and Renee Legro were engaged in lawful activities, each pursuant to a U.S. Forest Service permit. We conclude that having a permit to graze sheep where the attack occurred does not establish the Robinsons' control of the property within the meaning of the exclusion.”

The area where the Robinsons grazed sheep was open to hikers, bikers, horseback riders, off-road vehicle users, and bicyclists.  The Robinsons could not even exclude a person from walking among or going near the flock of sheep. 

The Robinsons countered that their use of the property came with certain reporting responsibilities to the U.S. Forest Service, including reporting wild animals, such as wolves, and they had to pay taxes on their interest in the property.  Nevertheless, they could not exclude anyone from it. 

The court reversed the trial court’s grant of summary judgment as to the dog bite claim, and remanded for further proceedings. The case may now proceed to trial, though settlement is quite possible given the seriousness of the injuries. 

Insurance Company Offers to Settle

In a rare instance where settlement discussions are made public, the court noted that the Robinsons’ insurance adjuster had agreed to settle Legro’s claims for $643,500, but Legro had insisted on $1 million.  This came to light because when the trial court granted the Robinsons’ summary judgment motion, the offer was arguably still open.  Legro attempted to accept it by sending an email.  The appellate court found, however, that no agreement had actually been reached and the offer had effectively been revoked after the trial court’s decision. 

Criminal Conviction Leads to Fine and Community Service

A story carried by the High Country News says that in the same week that Legro was bitten, the dogs attacked at least one other biker.  The dogs were Great Pyrenees that the Robinsons euthanized after the incidents.

“Sam Robinson was convicted of unlawful ownership of a dangerous dog under a Colorado law previously applied only to urban pets. In October, he was sentenced to 200 hours of community service and a $500 fine plus restitution."

Ariana Brocius, the reporter covering the story for High Country News, continues: 

“At the same time, bears, mountain lions and coyotes are thriving, and with the loss of some traditional predator control options ranchers are relying more heavily on guard dogs for protection, says Bonnie Brown, executive director of the Colorado Woolgrowers Association…. The dogs are the most effective predator defense left, say ranchers and rangeland specialists, because they live with the herd, bark and chase off mostly nocturnal predators. In the past year without their dogs, the Robinsons lost 26 percent of their sheep, compared with 7 percent in an average year. And last summer's incident is probably not the last, warns fourth-generation sheep rancher Anthony Theos, president of the Colorado Woolgrowers Association. If Robinson's case sets a precedent, he says, ‘This could be the beginning of the end for us.’"

Great Pyrenees and Akbash Dog Guarding a Flock (USDA photo)
Brocius quotes Pettee and Eagle County Animal Services Director Natalie Duck as saying she receives a few complaints each summer, but notes that "the danger is there, particularly for mountain bikers, who, unlike hikers or ATV riders, appear suddenly and noiselessly and thus might seem particularly threatening to dogs, says Brown.”


Bikers and others should be warned that having races across privately-owned areas, even if not fenced in, may not so easily avoid the predator control dog exclusion.  Those setting up races like the one Legro participated in should make the effort to learn about any predator control dogs that might be guarding herds near which the race will be run.  Similarly, livestock owners that graze on public lands should be aware that overly aggressive livestock guarding dogs may become a civil, and as here, even a criminal liability. 

It is arguable that the U.S. Forest Service bears some responsibility, though probably not liability, here.  By allowing competing and sometimes conflicting uses of wilderness areas, the agency allows for the possibility of this kind of injury.  As the only party with full knowledge of the circumstances, it could have notified the Robinsons in advance that there might be cyclists coming near their herd, possibly avoiding the tragedy that resulted. 

It is a shame when such things happen.  Livestock guarding has been a function of dogs since near the beginning of domestication, and is mentioned in Homer (where dogs of a swineherd are only deterred from attacking Odysseus by his quick adoption of a submissive posture).  Although livestock-guarding dogs on large ranches may work best if not familiarized to people, those working near population centers or places with public access are sometimes best taught to accept the presence of people near the herd.  For an excellent account of how such dogs are trained, see Jay Lorenz and Lorna Coppinger, Raising and Training a Livestock-guarding Dog, Oregon State University Extension Service (May 2002) (with a helpful list of additional sources).

Legro v. Robinson, 2012 WL 5266059 (Colo.App. 2012)

Thanks to L.E. Papet, Hilary Phillips, and Leigh Anne Novak for suggestions and corrections.

Thursday, November 1, 2012

Do Dogs Display Guilt after Eating Forbidden Hot Dogs? It Depends How the Data Is Massaged

An advertisement in a Budapest newspaper sought participants in a study about pet dogs.  The ad specified that the only owners who need reply were those whose dog (1) was older than eight months, (2) had lived with the owner for at least six months, (3) could remain calm if left alone in an unfamiliar room for three minutes, and (4) had not participated in food-reward studies. Also, owners had to say that if they prohibited their dog from eating food in a room, they would be unsure whether the command would be obeyed if they left the room. 

Generally, the experiment involved recording the greeting behavior of the dog after a brief separation from the owner, followed by the owner telling the dog not to eat some hot dog pieces on a plate and leaving the room.   The owner returned to the room, but could not see the plate with the hot dogs behind a barrier, and had to assess whether the dog’s behavior indicated that it had eaten or not eaten the food. 

Most owners (40 of 54) correctly stated what their dog had done while they were out of the room. Nevertheless, the researchers eliminated a number of the responses on various grounds, after which there was no longer statistical support sufficient to conclude that owners could actually recognize guilty behavior in their dogs.   

Dog Behavior

The research was conducted by Julie Hecht, Adam Miklosi, and Marta Gacsi of the Department of Ethology of Eötvös Loránd University in Budapest.  All three authors are among the most renowned dog behavior people now working, and all have been cited in prior blogs, as well as in both of my dog books. 

Their discussion begins with the behavioral aspects of domestication, where they note:

“Among scientists, the emotional experience of social, non-human animals remains a contentious topic.  While primary emotions, like anger and fear, serve an evolutionary function to adapt and respond to social and environmental demands, the precursors for the experience of secondary emotions, like embarrassment, pride and guilt, are not widely attributed to non-human animals.”

Dog owners, however, believe that dogs experience guilt, and they believe this more of dogs than of any other companion animals.  One study (Morris et al., 2008) concluded that 74% of dog owners claim to recognize guilt in their dogs; the next-highest attribution, at 36%, is found in horse owners.

The scientists describe guilt as “a self-conscious, evaluative emotion associated with a self-perceived violation against an established rule to which one has agreed to adhere.”  Guilt is useful in reinforcing social relationships and behaviors, and is associated with actions such as constricting posture, moving the head down and averting gaze, which have been described in non-human primates.  Wolves sometimes demonstrate such behaviors.  The scientists cite Alexandra Horowitz’s conclusion that “when dogs display these behaviors in particular interspecific social contexts, owners interpret the behavior to signify knowledge of transgression and therefore indicative of guilt.”  Horowitz, however, found that the “guilty look” was, in fact, “a response to owner scolding, not an appreciation of a misdeed.”  (Horowitz, 2009)

Objectives of the Study

Hecht and her colleagues wanted to explore two questions:
  • Is there a difference in dog presentation of ABs [associated behaviors of guilt] when greeting their owners between dogs that disobeyed a social rule in owners’ absences and dogs that did not disobey?
  • By relying solely on dog greeting behavior, can owners determine, upon returning to a room, whether their dog transgressed?
The scientists found 64 pet dogs whose owners filled out a questionnaire.  The questionnaires revealed that owners mentioned 11 associated behaviors of guilt (“guilt behaviors”), which were:
  •  lowering body 
  •  tail down 
  •  moving away from owner 
  •  freezing 
  •  turning 
  •  head away from owner 
  •  lowering head and lack of jumping 
  •  putting ears back 
  •  averting eyes 
  •  approaching low and slow 
  •  lack of barking
Video recordings indicated that jumping was not revealed during scolding.

The questionnaires revealed that 91.8% of owners claimed that their dogs sometimes showed guilt behavior, and that this behavior indicated that the dog knew it had done something wrong. Of these, 59.2% acknowledged that a dog’s guilty behavior caused them to scold the dog less.  This is consistent with previous studies concluding that attributing human-like emotions to non-human animals affects how humans interact with these animals.  (Anyone who remembers childhood knows that acting guilty often reduced the punishment.)

Dogs were of many breeds, slightly more females than males, though many more women dog owners participated than men, 47 against 6.  This curiously high imbalance should have been explained.  

Phases of the Experiment

(1) In the first phase of the experiment, the dog, the owner, and the experimenter entered the test room.  The dog was off-leash and free to explore. The room contained a table that was low enough to make it easy for the dog to see anything on it. The experimenter and owner left the room for three minutes. The owner re-entered the test room but, instead of interacting with the dog, the owner stood at the door and described the dog’s behavior to the experimenter, who remained in the hallway but could hear the owner through the closed door.  The owner then put the leash on the dog and they both left the room. 

(2) In the first food phase, the owner, the experimenter, and the dog entered the test room and the dog was taken off leash.  The owner held a plate with two pieces of hot dogs.  The owner and experimenter pretended to eat from the plate for thirty seconds to “create the desired social context—that the food was for humans and not dogs”—then the owner put the plate on the table without saying anything to the dog.  If the dog ate any of the hot dog pieces in the next thirty seconds, the owner scolded the dog for up to ten seconds while the experimenter turned her back (to give the owner privacy, which it was hoped would make the owner act natural while scolding).   

(Two questions we have at this point: When did the owner know that he or she would be able to scold for the dog eating, which had not been prohibited?  If this was known before the dog ate, there might already have been a degree of prohibition from the owner’s body language.  Also, if a dog is regularly punished for eating human food, would not the “social context” of the experimenter and the owner pretending to eat from the plate effectively create a command?)

Regardless of whether the dog had eaten or not, the owner next picked up the plate and the experimenter and the owner walked around the edge of the room.  If the dog had eaten, the experimenter replaced the hotdog from her waist pouch, and the owner and experimenter again pretended to eat from the plate. Then the owner put the food back on the table, but this time specifically warned the dog not to eat it.  If the dog ate the food anyway, the owner scolded it. (So some dogs had been scolded twice by this point.)

(3) In the next food phase, the owner put the food on the table and told the dog not to eat it, then both the owner and experimenter left the room.  The dog was left alone for three minutes.  The owner then re-entered the room but could not see the table because of a small barrier that prevented a view from near the door.  The owner had been instructed to decide, based on the dog’s greeting behavior, whether it ate the food during the three-minute absence by saying “yes,” “no,” or “not sure.”  The experimenter asked the owner (from the hallway), “Why do you think this?” to get as much detail as possible regarding the dog’s behavior.  If the dog had eaten, the owner could scold.  If not, the owner could praise. 

(4) In a final phase, the opening non-food phase to watch the initial greeting behavior of the owner and the dog was repeated. 

What the Researchers Saw

The researchers examined whether dogs that had eaten differed from dogs that had not in the part of the experiment which occurred as owners entered the room after having previously warned the dogs not to eat the hot dog pieces.  The researchers who scored the dogs’ behaviors at this point were themselves unaware of whether the dogs had or had not eaten the hot dog pieces, as were the owners at the moment the greeting occurred.  The researchers could detect no difference between eaters and non-eaters in their display of guilt behaviors. 

Nevertheless, some differences in greeting behavior were detected by the researchers.  Dogs generally showed fewer guilt behaviors when greeting owners than when being scolded, which seems consistent with Horowitz’s observations.  Dogs that did eat when their owners were out of the room showed more guilt behaviors in greeting after the owner returned than they had in the first phase when separated from their owners with no food involved. Thus, these dogs showed more guilt behaviors after violating a command than they did in the baseline established by the first greeting in the experiment.    

The researchers also detected that dog which did not eat in their owners’ absence showed fewer guilt behaviors in the final phase (4) than they had when greeting their owners after (3). Thus, dogs that many people would call “good dogs” because they did not eat, nevertheless showed more guilt behaviors when they had the opportunity to be bad than they did when they did not have this opportunity at the end of the experiment.  This might suggest these good dogs were fearful of being punished for what they did not do, or just better trained.   There was no discussion of the level of training previously received by the dogs in the experiment. In either case, guilt behaviors as observed by the researchers were not randomly distributed throughout the greetings in the experiment.   

Owners’ Responses

When asked if a dog had eaten in the owner’s absence, 4 owners were unsure but 58 said “yes” or “no.”  Of those 58, 40 were correct and 14 incorrect. Thus, overall, owners were successful in determining whether dogs had or had not violated a command not to eat. 

Nevertheless, the researchers did not think that this confirmed that owners could correctly assess whether a dog had eaten or not because some owners perhaps had a bias based on a dog’s prior behavior in the experimental context.  They first eliminated results of owners who seemed to have a bias towards thinking one way or the other based on their mentioning the dog’s history inside the experiment.  The results after this adjustment were still better than chance.  If, however, the researchers looked only at owners whose dogs ate while the owner and the experimenter were in the room but before the prohibition on eating the hot dog pieces had been explicitly stated, but did not eat after the prohibition was stated and the owner and experimenter were still in the room, the result was no better than chance.  This selection of the data excludes dogs that did not eat because they were unsure if there was a command in place, or accepted that there was a command from the pretend eating of the owner and the experimenter, or from body language of the owner if the owner knew that he or she was going to scold if the dog ate, regardless of the lack of explicit command.

Whether selecting a specific slice of the data proves that dog owners cannot really tell guilt in dogs is not clear to us, and the scientists do not in the end claim that it is.  After all, as noted before, the researchers did find, in their analysis of the videotapes, that dogs that ate when their owners were out of the room showed more guilt behaviors in greeting after the owner returned than they had in the first phase when separated from their owners with no food involved.   The researchers acknowledge that their experimental design left some possibilities open, and further testing will be necessary to say more.   

Design Issues

As dog people, we are concerned about several aspects of the experiment.  It is not stated when the dogs last ate.  Were they allowed normal food intake?  Was there a period in which they were not allowed to eat? Contrary to popular belief, not all dogs like hot dogs.  If a dog ate a hot dog piece when the owner and experimenter were in the room, it might not have been interested in trying again.  Does this make its response more valid, as determining the owner’s ability to sense guilt, than if it disliked hot dogs enough not to eat during any of the opportunities? 

The interpretation of a dog’s actions by its owner might also be subject to more detailed analysis.  What guilt behaviors did each individual owner specify?  Could those specified by an individual owner have indicated the owner recognized some behavior associated with guilt, while the dog’s overall behavior, on the 11 guilt behaviors as assessed by the researchers from the videotapes, was more neutral? 

Additionally, it must be noted that some owners, in our experience, list other behaviors as inducing guilt including jumping on furniture, urinating, or scratching the floor, some or all of which may have existed in the experiment during the time the experimenters and owners were out of the room, as this was not discussed in the report. Arguably, physiological changes should also be recorded. 

Perhaps Only Some Dogs Show Guilt

One of us (JE) grew up with two Cocker spaniels, Blackie and Sandy.  My mother always insisted that she could tell when Sandy had done something he was not supposed to do because he would display guilt.  Blackie, in her opinion, did not feel guilty about anything.  As a child, I always assumed that she was right about this, and I believe that Blackie was punished more because of the feeling that he was recalcitrant. 

Guilt seldom had anything to do with being left alone in the house because dogs of Pullman, Washington, in that era were more often outside than inside.  Our dogs were not in the house except in the worst weather and had an insulated doghouse that had been built by the same contractor who built our house.  Guilt was usually about digging something up in my mother’s flower garden or eating something from the outdoor flagstone fireplace grill.  Blackie, in particular, had a curious love of carrots and could dig up two or three a day, which annoyed my father when it began.  Soon he was not punished for this, however, as once my father accepted Blackie's love of carrots, several rows of the vegetable were planted for his use alone.  
Perhaps Blackie had a bit of the wolf left in him.  He was the one who learned how to get away by (1) wrapping a leash around a tree and pulling back and forth until the leash broke, (2) climbing over the 6-foot high steel hog fence that my father strung so the dogs would have a large yard, and (3) digging under this fence after my father strung barbed wire across the top to prevent Blacky from climbing over it.  After that my father gave up and any time a bitch was in heat in Pullman we’d have to wait for the irate owner to call and tell us where Blackie was hanging out.  Sometimes Blackie was kept in the house for a few days if the owner of the female was sufficiently angry. 

Sandy never disobeyed the rules and never went off in search of females that might want some attention. My mother and I cried for days after he was hit by a car when I was ten years old.  Blackie died a natural death at 16, a year after we moved to Fresno. 

Defining Guilt in Dogs

Whether my mother was right about Sandy showing guilt I cannot say. Hecht and her colleagues observe that “an ambiguous social situation generated by repeated scoldings and greetings—not uncommon for experiments investigating guilty behavior—could affect the behavioral displays in question in a complex way.”  Perhaps this acknowledges that there are more things in dog heaven and earth than meet the scientist’s eye. How “repeated” must scoldings be to affect the results of an experiment like this?  I suspect Sandy and Blackie were only punished two or three times a year. 

Guilt is a somewhat loaded concept, and in many uses includes an element of remorse.  Such is not established, and not intended, in this research.  The researchers describe guilt as “a self-conscious, evaluative emotion associated with a self-perceived violation against an established rule to which one has agreed to adhere.”  Perhaps this could be rephrased to say that guilt is a self-conscious evaluative emotion associated with an understood violation of a rule to which an animal has been trained or expected to adhere.  The dog does not feel remorse, but rather anticipates disapproval or punishment based on not adhering to a command.  Some of the results arguably support this.  The study at least establishes that perception of guilt behaviors is more than a simple anthropomorphic projection we put onto the behavior of our dogs. (Whether remorse in humans is partly an internalization of punishments once received I leave to the psychoanalysts.)


Thinking that dogs have human thoughts and emotions that they may not in fact have, i.e., anthropomorphism, as argued by James Serpell (2003) and others, probably existed very early in the human-canine relationship, and may be fundamental to the success of domestication.  It explains something about why many of us see dogs as members of our families and societies.  Homer’s account of a hunting hound recognizing his master in Book 17 of The Odyssey contains elements of anthropomorphism.  The dog, Argus, has not seen his master for two decades but is not strong enough to crawl to him.  Odysseus cannot pet the dog for fear that Penelope's suitors will recognize him and kill him.  He has to hide a tear, but perhaps tries to speak to the dog indirectly by asking the swineherd, who is serving as his guide, why such a finely built dog is left to lie on a dung heap. 

As Homer knew, there is room for guilt on the human side of a relationship between a man and a dog. To this day, I feel guilt on remembering how I ignored Blackie when I was a teenager, trying against hope to fit into the California culture that was new and frightening to a maladjusted misfit, paying no attention to my long faithful dog as he prepared to die.  I failed him, but the memory of that failure has become part of my conscience, and I hope has made me a better dog owner as an adult.  To put it another way, Blackie is still part of me.  That is one of the things our pets do for us. 

This blog was written by John Ensminger and L.E. Papet. 

Hecht, J., Miklosi, A., and Gacsi, M. (2012). Behavioral Assessment and Owner Perceptions of Behaviors Associated with Guilt in Dogs.  Applied Animal Behavior Science, 139, 134-142.

Additional References:
  1. Horowitz, A. (2009). Disambiguating the Guilty Look: Salient Prompts to a Familiar Dog Behaviour. Behavioral Process 81, 447-452.
  2. Keltner, D. (1995). Signs of Appeasement: Evidence for the Distinct Displays of Embarrassment, Amusement, and Shame. Journal of Personality and Social Psychology, 68, 441-454.
  3. Morris, P.H., Doe, C. and Godsell, E. (2008). Secondary Emotions in Non-Primate Species?  Behavioural Reports and Subjective Claims by Animals Owners. Cognition and Emotion, 22(1), 3-20.  
  4. Serpell, J. A. (2003). Anthropomorphism and Anthropomorphic Selection—Beyond the “Cute
    Response.” Society and Animals, 11(1), 83–100.