Measuring the Emotional Toll of Aversive Training Methods
The tension between dog trainers who use primarily reward-based methods (positive reinforcement) and those who rely more on aversive methods (positive punishment/negative reinforcement) is not new. Many trainers (myself included) believe that not only are reward-based methods more humane than methods that rely heavily upon aversives, but also that they lead to more effective learning outcomes, happier dogs and, following The Ben Franklin Effect, to happier trainers as well.
Beliefs aside, where does science fall on this issue? We have some evidence showing that relying primarily upon positive reinforcement when training dogs is effective, improves a dog’s confidence, and is associated with low levels of stress. In recent years, clicker training, a popular application of reward-based training, has also garnered study and has been shown to be at least as effective as other forms of reward-based training. (Much of this evidence is reviewed in earlier Science Dog blogs and in my book, Dog Smart).
Studies of Aversive Training Methods
We also need information regarding the effects of aversive training techniques. Specifically, how might the use of punishment and negative reinforcement affect a dog’s learning response and emotional state?
First, there is plenty of evidence that using aversive stimuli such as e-collars, collar corrections, physical punishment and verbal reprimands can effectively change behavior – this is precisely why these techniques were used in the early days of training and continue to persist. They work – and often work rapidly.
But, is there a cost? Yeah, there is. For the dogs.
A number of studies have compared the behavioral effects of aversive training methods to reward-based training methods in dogs. Here is a brief summary:
- 2004 Survey Study: Behavior problems in dogs were positively correlated with reliance upon correction-based training methods but not with reward-based training methods. Owners who used reward-based training reported overall better obedience in their dogs when compared with those who used primarily negative reinforcement and punishment (1).
- 2004 Observational Study: German Shepherds trained for guard work using an e-collar (shock collar) showed lowered body postures, high pitched yelping, avoidance behaviors, and redirected aggression – all signs of fear and pain. These signs were not seen in dogs who were not shocked during training. This study also showed that dogs trained with shock learned to associate the presence of their owner with impending shocks, even outside of the training context (2).
- 2011 Survey Study: Dogs owned by people who stated that they used a high proportion of punishment when training were significantly less playful and were less likely to enjoy interacting with visitors compared with dogs of owners who reported that they used primarily reward-based methods (3).
- 2014 Study of Two Training Schools: A comparison of dogs trained at a school that used reward-based methods to dogs trained at a school that used primarily correction-based methods found that dogs who were trained with negative reinforcement showed significantly more stress signs and were less likely to offer spontaneous eye contact with their owners than were dogs trained with reward-based methods (4; Reviewed in The Consequences of Consequences).
- 2014 E-collar Study: Pet dogs were trained to come when called either using an electronic (shock) collar or using reward-based training methods. All of the dogs successfully learned to come when called. However, dogs trained with e-collars were more tense, yawned and vocalized more frequently, and explored their environment less than dogs trained positive reinforcement. These behaviors were interpreted as signs of stress and anxiety in the dogs trained with an e-collar (5).
- Aggression and Aversive Methods: Several papers have reported increased prevalence and risk of aggression-related behavior problems in dogs trained with aversive methods, when compared with dogs trained using reward-based methods (6,7,8).
Most of the published studies have relied either upon observations of dogs during training sessions or upon owner-reported survey data. One question that has remained elusive is how might different training approaches impact a dog’s emotional state outside of the training setting and in the long-term? In other words, do the types of training methods that are used have a lasting influence on a dog’s behavior, emotional health, and owner relationship? Two recent studies, conducted by separate groups of researchers, provide some answers (9,10). Both used a relatively new test paradigm to measure the emotional state of dogs – the cognitive bias task.
The Cognitive Bias Task
Measuring an animal’s emotional state can be tricky. This is because emotions are an internal and private experience. Just as I cannot be directly privy to another person’s emotions or know what they are feeling “inside”, this is also true with our dogs. However, we do have a number of tools to help us.
Specifically, different emotional states are associated with distinctive sets of behaviors, body postures and vocalizations. The most obvious that come to mind are expressions of joy/happiness and those of fear/anxiety. Not to put too fine a point on this, but the same holds for our attempts to decipher the emotions of another person. When measured in controlled studies, typically via video analysis by double-blinded evaluators, body postures and behavior can provide a measure of a dog’s emotional state. Another objective measure is salivary cortisol, used to reflect a dog’s degree of stress.
In recent years, several new measures have gained acceptance and validation in the scientific community. One of these is the cognitive bias task, also called the judgement bias task. This test relies upon the assumption that an individual’s underlying emotional state will influence his or her behavior when faced with an ambiguous task. When used with dogs, the task involves the choice of approaching a bowl that may or may not contain food treats (11).
How Does it Work? The dog is first taught to discriminate between a rewarded versus an unrewarded location. One location is consistently rewarded by the presence of a yummy treat in the bowl. The second position is consistently not rewarded – the bowl is empty. Once the dog clearly anticipates receiving a tasty treat in one position, but not in the second position, bowls located in ambiguous positions (partway between the two locations) are tested. The time that dogs take to learn the task plus the time (latency) that it takes the dog approach and choose a bowl are recorded.
What Does this Mean? Results reflect a dog’s inclination to have either an optimistic (there might be a food treat in that bowl – I should go look!) or a more pessimistic (I bet that bowl is empty – and I may be reprimanded if I go there….) interpretation of seeing the bowl placed in an ambiguous location. Dogs who take longer to learn to anticipate a food reward in a consistent location and/or subsequently, to either fail to approach or take a long time to approach a bowl are scored as having a more pessimistic or more negative emotional state. In other words, these dogs anticipate the absence of a food reward rather than the possibility of its presence. They may also anticipate punishment should they make a mistake or approach the new location. While this test is not universally accepted and has limitations, there is evidence that it can be helpful in discerning dogs’ underlying emotional states in a variety of settings (12). Two recent papers report on its use in dogs who were trained using reward-based methods and those trained primarily with aversive methods.
The Seven Schools Study
The first paper studied 92 dogs, divided into three groups. Group Reward (42 dogs) was recruited from three dog training schools that used exclusively reward-based methods; Group Aversive (28 dogs) was enrolled in a training course at one of two schools that used primarily aversive methods; and Group Mixed (22 dogs) was enrolled at one of two schools that used a mix of reward-based and aversive methods (9). Measurements: The dogs were videotaped during training sessions on three occasions. Saliva samples were collected prior to and after training sessions for cortisol analysis. In addition, all of the dogs participated in the cognitive bias task, conducted at least one month following class completion.
- Behavior during class: During training, dogs who were trained with aversive techniques displayed significantly more stress-related behaviors, panted more frequently, and were more likely to be evaluated as being tense when compared with dogs trained with reward-based methods. Aversively-trained dogs also demonstrated significantly more avoidance behaviors such as turning away from the owner and showing a lowered body posture or crouching/lying down.
- Cortisol: Aversively-trained dogs also had significantly higher increases in salivary cortisol in response to a training session compared with dogs who were trained using reward-based methods.
- Cognitive Task: Dogs trained with aversive methods had longer latencies (took longer to search for food) and were scored as more “pessimistic” than dogs trained with reward-based methods (were less likely to anticipate a food reward in the bowl).
The authors conclude that these data, collected from a large number of dogs being trained at multiple schools, found a strong association between the use of aversive-based training methods and increased frequency of stress-related behaviors. In addition, dogs trained with aversive methods were less likely to anticipate a food reward when allowed to seek a location that may have provided food. Not only were aversively-trained dogs more hesitant to search for the reward, they also learned the task more slowly than did dogs trained with reward-based methods – even thought the task was trained with only a reward-based approach.
The In-Home Study
The second study was conducted by a group of researchers headed by Rachel Case of Dogs Trust in the UK (10). A group of 100 dogs and their owners were recruited for the study. The owners completed a phone interview and questionnaire to assess and verify the frequency that they used reward-based versus aversive-based training methods with their dog. Dogs were divided into two groups; 50 dogs who were regularly trained using two or more aversive methods and 50 dogs who were trained only with reward-based methods. Measurements: Each dog was trained and then tested in the Cognitive Bias Task described previously.
- Learning phase: Unlike the previous study, dogs who were trained using at least some aversives did not take longer to learn the task.
- Testing: However, when tested using ambiguous locations, dogs who were trained using aversive methods took longer to explore the potentially rewarding locations when compared with dogs who were trained with reward-based methods. Similar to the previous study, this suggests that dogs trained with aversive methods were more hesitant to explore and to make a choice (i.e. anticipated a lack of a reward rather than the presence of a reward).
The results of this study corroborate those of the study comparing dogs from different training schools. This is especially significant because the two studies used the same cognitive (judgement) bias task. The authors of this study conclude that dogs who are trained using at least some aversive methods demonstrate a more negative emotional state, even outside of the training context.
The Eeyore Effect
We now have data showing that dogs who are trained with aversive methods are more highly stressed during training and also express signs of less positive (happy) emotional states even outside of a training context. The first study found that dogs who were trained aversively were hesitant to make selections and also learned the task more slowly, even when being trained with completely reward-based methods. This is pretty heart-breaking. The second study found that while all of the dogs learned to anticipate food rewards in a certain location, the dogs who were trained aversively were more hesitant to explore and search for the food treats.
The terms optimism and pessimism are the age-old paradigm of glass half-full versus glass half-empty. A person (or dog) who is said to be optimistic is someone who generally expects good outcomes (consequences) while someone who is pessimistic generally anticipates poor or negative outcomes to prevail. For those who are familiar with the classic Winnie the Pooh stories, you will remember that Winnie, the stories’ protagonist, was a perpetual optimist, always seeing the bright side of things. Winnie’s sidekick, Eeyore, on the other hand, was the hapless donkey who, while certainly lovable, went through life convinced that all situations would end in doom.
Are dogs who are trained with aversive methods inclined to become the Eeyores of the dog world? Consider that a dog who is repeatedly punished for offering an incorrect behavior may subsequently develop an outlook, if you will, that is pessimistic or generally negative (or even depressed?) in nature. It makes sense of course – if aversive stimuli in the form of corrections, voice reprimands, physical coercion, are a regular part of a dog’s life, then it certainly should be no surprise that such a dog becomes more inclined to anticipate unpleasant outcomes in new situations than pleasant outcomes. And of course, the opposite may also be true. A dog who has learned that many newly offered behaviors (but not all) will be rewarded with a food treat, praise or a game, may become a dog who tends to view new situations with a more rose-tinted outlook.
The Emotional Toll of Aversive Training Methods
So, at this point in time, I think it is sufficient to say that we DO have enough science to tell us that: (a) dogs do better emotionally in the training setting when trained using reward-based methods; and (b) training dogs with aversive methods may damage them both emotionally and in their relationships with us, even outside of the training setting.
Help your dog to be a Winnie and not an Eeyore.
- Hiby EF, Rooney NJ and Bradshaw JWS. Dog training methods: Their use, effectiveness and interaction with behaviour and welfare. Animal Welfare 2004; 13:63-69.
- Schilder MBH and van der Borg JM. Training dogs with the help of the shock collar: short and long term behavioral effects. Applied Animal Behavior Science 2004; 85:319-334.
- Rooney NJ and Cowan S. Training methods and owner–dog interactions: Links with dog behaviour and learning ability. Applied Animal Behaviour Science 2011; 132:169-177.
- Deldalle S and Gaunet F. Effects of two training methods on stress-related behaviors of the dog (Canis familiaris) and on the dog-owner relationship. Journal of Veterinary Behavior: Clinical Applications and Research 2014; 9:58-65.
- Cooper JJ, et al. The welfare consequences and efficacy of training pet dogs with remote electronic training collars in comparison to reward-based training. PloS One 2014; 9:e102722.
- Arhant C, et al. Behavior of smaller and larger dogs: Effects of training methods, inconsistency of owner behaviour and level of engagement in activities with the dog. Applied Animal Behaviour Science 2010; 123:131-142.
- Casey RA, et al. Human directed aggression in domestic dogs (Canis familiaris): occurrence in different contexts and risk factors. Applied Animal Behaviour Science 2014; 152:52-63.
- Herron ME, Shofer FS and Reisner IR. Survey of the use and outcome of confrontational and non-confrontational training methods in client-owned dogs showing undesired behaviors. Applied Animal Behavioural Science 2009; 117:47-54.
- de Castro ACV, Fuchs D, Morello GM, Sastur S, de Sousa L, Olsson AS. Does training method matter? Evidence for the negative impact of aversive-based methods on companion dog welfare. PLos ONE, 2020: 15(12)e0225023.
- Casey RA, Naj-Oleari M, Campbell S, Mndl M, Blackwell EJ. Dogs are more pessimistic if their owners use two or more aversive training methods. Scientific Reports, 2021; 11:1-8.
- Burani C, Pelosi A, Valsecchi P. A promising novel judgement bias test to evaluate affective states in dogs (Canis familiaris). Animal Cognition 2022;doi.org/10.1007/s10071-021-01596-z
- Burani C, Barnard S, Wells D, Pelosi A, Valsecci P. Using judgement bias test in pet and shelter dogs (Canis familiaris): Methodolgocial and statistical caveats. PLos ONE 2022;15:e0241344.
3 thoughts on “The Eeyore Effect”
Love this article Linda – thanks for the excellent review of the status of dog training science. I’m just curious as to why the call out to clicker training – seems like that might confuse some folks. I’m a certified canine behavior consultant myself and I don’t use clickers for most of my training – I love them for tricks and shaping some behaviors.
But, from what I have seen, the science shows us that verbal reward markers are as effect as clicker training for dogs … but it feels like to me that by calling it out you are indicating a preference for clicker training? Maybe I’m reading it wrong. But I dislike the in-fighting in the positive reinforcement dog community over this and other things, so am concerned with anything that feels like it might fan that flame. Thank you. Jodi
Hi Jodi, Thanks for your note. Clicker training is mentioned because it is a popular and well-known application of +R training and also because there are now several published studies of its use. I am not sure why the single statement that I make about clickers would be confusing. Thanks however, for your thoughts! Linda Case
Pingback: A New Scoop for Citizen Scientists | The Science Dog