The essential task of the modern dog trainer

        Teach the dog how to behave naturally

                In an unnatural environment

        Master this

The rest is easy

I am the proud father of two. A daughter whose age at the time of this writing is measured in weeks, and a son who is roughly two and a half.

Like most toddlers, my son experiments with many different behaviors, often to see what he can get away with. He is experimenting with the world to learn what is within the boundaries of acceptable behavior.

Earlier this week when I was dropping him off at daycare, one of the caretakers in his room informed me that over the past couple weeks he has been occasionally hitting other kids.  They said that when they try to intervene, he often just laughs as though it’s a game, and continues acting up. Naturally, at this point I asked what they do to discipline him. Their response was that they tell him to stop, and they try to redirect him into another activity. When I asked if they thought this approach was effective, the answer was “no.”

Here’s the deal. I wasn’t terribly shocked when they told me that my son had attempted to hit another kid. Other kids have done the same to him. Little kids do things like this. In fact, he had on occasion tried to hit his mother and me at home.  However, because of the way we address the behavior at home, it quickly comes to an end, and he hasn’t repeated it for a long time.

The difference is consistent and effective discipline.  I don’t hit my child. I don’t raise my voice. In fact, I don’t even get frustrated (training dogs has taught me to have tremendous patience.) When my son tries to hit one of us, I calmly tell him that if he does it again, he is going to get a “time out.” If he chooses to try it again, I calmly pick him up, carry him into his room, and put him in his crib. No second warnings or further discussions at that point. At his age, the “time out” is relatively short, lasting only  about 2 minutes. When I go to get him, I ask if he is done hitting, he says “yes,” and then we return to our normal activities. 

At his daycare, they have a policy against putting kids in time out. I get it, you have to draw a clear line in the sand when it comes to disciplining other people’s children. But, in this case, time out works. The reason it works for my son is because he hates it. It makes him upset, he often cries, and calls for us to come and get him. In other words, this type of discipline is effective at reducing the target behavior precisely because it is an aversive and stressful experience for him and he does not want to repeat it. Redirection does not inhibit behavior, meaningful consequences do.

In the world of animal behavior, we would say that the consequence of “time out” is a form of “negative punishment.” In other words, I am punishing (making less) a behavior by taking something away (in this case playtime, freedom, attention, etc.)

In order for negative punishment to be effective, there are two main criteria that must be met:

(1)  The experience must be sufficiently unpleasant to motivate a change in behavior; and

(2) The animal must be able to connect the consequence of his/her action to the target (problem) behavior.

In other words, the animal must be able to assume that the behavior we are attempting to reduce is what causes the punishment, and the animal must be able to deduce from this the necessary change in behavior required to avoid the punishment in the future.

In the world of animal behavior, and dog training specifically, there is an underlying assumption held by many that negative punishment is more ethical than positive punishment (i.e., applying an aversive, rather than removing a privilege, reward, or opportunity for reward).  This assumption, however, is worth questioning.

Two immediate questions should come to mind when evaluating the ethics of any particular approach to training:

1.     What is the motivational or emotional value of the consequence (is it an attractive or aversive event, and to what degree)?
2.     How easy is it for the animal to connect the consequence to their behavior?

It is widely agreed upon that for a consequence to have an effect on the future occurrence of a behavior, the consequence needs to have a “motivational representation.” In other words, we must ask whether the consequence is aversive (something the animal would want to avoid), or is it attractive (something the animal would seek out)? The lynchpin of most arguments for more “humane” dog training, training which only involves reward-based techniques (Positive Reinforcement and Negative Punishment), is that humane dog training should not involve aversive consequences or stress. Most trainers who operate under this philosophy will admit that they use punishment, but will make the distinction that they are using negative punishment (removal of reward) versus positive punishment (application of an aversive), and that the former is inherently more humane than the latter.

It is this basic belief – that negative punishment is inherently and always more humane than positive punishment – that I seek to bring into question here.

In fact, if we look to the scientific literature, there is mounting evidence that this assumption is false. To begin, consider the work of  Schalke, Salgirli, Bōhm, and Harbarth (2008). In a pair of parallel studies, the researchers tested the cortisol stress levels, as well as the overall effectiveness, of three different consequences: (1) the Prong Collar; (2) the Electronic Collar; and (3) a Quitting Signal on 42 police dogs.

The “Quitting Signal” is a verbal cue that tells the dog that its behavior will not be rewarded (negative punishment). The dogs were trained with the Quitting Signal ahead of time using food and toys for a period of 4 months¹. 

In this study, the dogs were asked to heel, and then a “helper” appeared on the field attempting to provoke the dog. If the dog made a mistake (failed to heel), the appropriate correction was administered, either a prong collar correction, electronic collar correction, or they were given the quitting signal (non-reward). The researchers then tested the level of stress each correction produced in the dog.

The researchers found that the quitting signal produced the highest levels of stress (measured by cortisol), and the electronic collar produced the lowest. This is the exact opposite of what the popular reward-based philosophy of animal training would predict.

Before we go further, it is worth noting that there are several key factors that contributed to these results that also make it difficult to generalize these findings to the average family pet. The most significant is the fact that the dogs used were all Belgian Malinois, a breed which possesses far stronger appetitive motivation that the average dog. The more an animal wants a reward, the more stressful it is when the reward isn’t delivered, and these dogs really want the reward.  In fact Schalke noted: “Frustration is a high stressor for Malinois,” suggesting that the authors would agree  that negative punishment can produce high levels of stress in an animal, and in this case, higher levels of stress than with positive punishment.

However, it might also be argued that although the quitting signal produced higher levels of stress, the stress could be characterized as eustress (often considered “healthy’ stress) rather than distress, with the implication being that the former is healthier and more humane. However, these again are assumptions, and we should be careful about accepting such an appealing proposition as dogma.

There is in fact plenty of research that suggests we should be more careful with our assumptions.  Dickenson & Dearing (1979) were able to demonstrate in a transreinforcer-blocking² experiment that the omission of food (reward) has the same motivational representation to an animal as an electrical stimulation (aversive). To summarize the findings, the researchers were able to demonstrate that a signal, which a rat has learned to associate with receiving “No-food” can block new learning about a signal which means “A shock is coming.”

In the Dickenson & Dearing (1979) experiment, rats were taught that a tone means “no food.” Then the tone was given at the same time as a buzzer, both preceding a shock. The surprising result was that the rat’s learning about the buzzer/shock relationship was blocked by the tone/no food relationship. The tone/no food relationship was already represented in the rat’s mind as “tone means something bad/unpleasant is going to happen,” and since a shock is also unpleasant, the buzzer did not give the rats any new or helpful information from which to learn. What this all means is that we have reason to believe both “no food” (negative punishment) and “shock” (positive punishment) have the same motivational/emotional representation for the animal.

For those who like “sciency” jargon, Dickenson states in his book Contemporary Animal Learning Theory:

“Omission of an attractive stimulus activates the same motivational representation of an explicit aversive stimulus (such) as shock”

and

“The omission of expected food and the presentation of shock excited a common representation”

There are some conclusions that we may take from the two studies I mention above.  First, If we take the findings from Dickenson & Dearing, then it follows that negative punishment is represented in the animal’s mind as an aversive experience, in much the same way as positive punishment.  Moreover, the work of Schalke et. al. suggests that, at least in some circumstances, negative punishment can produce more stress, and thus be a more aversive experience, than positive punishment, such as an electronic collar.

Now that we have addressed the first question (i.e. What is the motivational and emotional value of the consequence?), we turn to the second question:  How easy is it for the animal to connect the consequence to their behavior?

We might start by again looking at the Schalke studies.  The authors sought to test the relative stress induced by electronic collars, prong collars, and quitting signal (negative punishment). However, not only did the authors report on the stress produced by each, but they also reported on the effectiveness of each at eliminating the target behavior.

Interestingly, although the quitting signal produced higher stress than the electronic collar, it actually had the lowest learning effect of the three tools. In fact, whereas both the electronic collar and the prong collar had a significant learning effect on 38 out of the 42 dogs, and 32 out of the 42 dogs respectively, the quitting signal had only a learning effect on 3 of the dogs, (38 of the dogs did not quit the behavior at all after the signal).  So in this case, negative punishment was both the least effective, as well as the most stressful, of the three techniques used (It should be noted that the electronic collar was both most effective, and least stressful).

These two studies are a just a few examples of similar findings in the literature, but it is certainly enough to add shades of grey to the black and white assumption that negative punishment is inherently more humane than positive punishment.

More recently, in 2015 in an article titled 51 Shades of Grey: Misuse, Misunderstanding and Misinformation of the Concepts of “Dominance” and “Punishment,”   Simon Gadbois of the Canid Behavior Research Laboratory at Dalhousie University noted:

“I have seen border collies go nuts if they are told only what they do right, and are ignored when making a wrong choice (for example, in a matching-to-sample task). In fact, ignoring wrong responses becomes very aversive, without really telling the dog what to avoid doing.”

While some may argue that the use of a non-reward marker (or quitting signal or conditioned negative punisher) does, in fact, tell the dog when a mistake is made, as we have seen in the Schalke studies, in some cases it is a very ineffective means of doing so. Between the lines in the Gadbois statement above is that this inefficiency can lead to very aversive experiences for the animal.

Back to the situation with my son. When addressing children, we have a significant advantage, which ensures that the behavior and consequence are connected: Language. The reason that negative punishment is far more useful when disciplining our children than when disciplining our dogs is that we can explain the connection to them. Note in the example above using my son, I told him “If you hit me again, you will get a time out.” Then, prior to removing him from his crib, I ask him to reaffirm his understanding: “are you done hitting?” “Yes, Dad.”

With dogs and other animals we do not have the luxury of semantic language. With animals, consequences are best understood when they begin and end with the behavior itself. If I withhold food from a dog as a consequence, the feeling of frustration, desire, or hunger may persist after the target behavior has ceased. However, with an electronic collar, I have the ability to both start and stop the consequence with surgical precision, with accuracy within one-fortieth of a second.

Again, I must reiterate, lest I be accused of arguing an absurdity,  that I am not in any way proposing that negative punishment is always more aversive and/or less effective than positive punishment. What I am stating is that if we are going to use science and empirical evidence as our compass in assessing the ethics of various training procedures, then we must be prepared to acknowledge several things:

1. That for punishment to be effective at eliminating a target behavior, the punishment itself needs to be aversive enough to compete with the motivation which is eliciting the behavior itself.
2. Negative Punishment (removal of reward) is effective because it creates an aversive experience.
3. At least in some cases, Negative Punishment is more aversive/stressful than Positive Punishment (application of an aversive, i.e. electronic collar).
4. At least in some cases, Negative Punishment is significantly less effective than Positive Punishment.
5. In some cases, Negative Punishment can be both more aversive and less effective at the same time (as seen in the Schalke studies).

This last acknowledgment is significant. How can we make generalized claims that positive punishment (i.e., prong collars, electronic collars, etc.) are inherently and objectively unethical, in favor of either ignoring incorrect behavior, or using a non-reward marker?

In the world of companion dog training, we are often faced with situations where dogs are exhibiting behaviors that are dangerous to themselves or others. In most cases these behaviors need to be controlled or eliminated quickly and effectively to both mitigate the risk to the dog or others, or to prevent the dog from being surrendered or euthanized. In such cases, there are times where a more aversive consequence can be justified if it produces a significantly more expedient result, thus maximizing safety for all involved. However, if a particular consequence is both more aversive, and less effective in a given context, it can almost never be justified.

We owe it to ourselves as professionals, the clients we serve, and most of all the dogs for whom we are stewards to take into consideration all the evidence of both effectiveness and aversiveness when evaluating possible training procedures.  There is no value in deceiving ourselves with false scientific claims and emotion-laden arguments. Furthermore, failing to take into account the effectiveness of training procedures, particularly concerning dangerous behavior, or other behaviors that place the dog’s future at risk is not only unethical to the dog, but to any other dog or person who may be hurt physically, emotionally, or otherwise as a result of the lack of effective treatment.

The point of all this is that black and white generalizations and cult-like dogma do not lend themselves well to the honest study of animal behavior, nor to productive discussions of animal welfare.

Ultimately, as to the question of which is more effective, humane, or ethical, positive or negative punishment? The only honest answer is “It depends…

–Tyler Muto

Notes

1. Below is the process which was used for training the Quitting Signal in the Schalke studies: 1st step:

The first aim of the quitting signal training was to condition the feeling of frustration with any vocable which was previously insignificant to the dog. To this end, the following program was carried out:

• 1)  The handler with many treats in one hand made fists with both hands. The handler held his/her hands in a certain position so that the dog was able to see both of them.
• 2)  The handler took the treats one by one from his/her full hand to his/her other hand and fed the dog until the association had been developed and the dog expected to have the food with the above mentioned hand movement. During this feeding session, no orders were given to the dog.
• 3)  The handler took the food piece by applying the same hand movement but this time instructed the signal, i.e., the previously chosen vocal, with the normal tone of voice immediately before the hand movement had completed and subsequently retained the food in his/her hand. The dog was startled by the sudden absence of the food which it got used to having without exhibiting any performance and, therefore, got frustrated.

• 4)  As soon as the dog quit demanding the food from the hand and exhibited any alternative behavior, the other hand was opened and the handler gave the food piece to the dog. Thus, exhibiting the alternative behavior after getting the signal was the only solution for the dog to terminate the feeling of a frustration.

2nd step:

• 1) In this step, the same feeding procedure as the one in the first step was performed by a foreign person.
• 2) Together with the signal instructed by the owner, the foreign person retained the food in his/her hand.
• 3) As soon as the dog showed the alternative behavior, the owner rewarded the dog by serving the food.

Different sorts of treats such as dry food, sausage, cheese, or mixture of two or three of them etc. were used for the training. If food mixture was used, the dog’s favorite one was served as a reward.

3rd step:

• 1)  For this obligatory step, the dog was on a long leash (approx. 3 m) and the owner had two toys.
• 2)  The owner played with the dog by throwing a toy until the dog had the feeling of free access to the toy.
• 3)  The handler threw the toy so far that the dog could not reach it and instructed simultaneously the signal.
• 4)  As soon as the dog exhibited the alternative behavior, the owner played with the dog by using the other toy as a reward.

2. “Blocking” is a phenomenon in animal learning that occurs when a previously learned signal interferes with learning about a new signal. Here’s an example:

Let’s say that I taught a rat that a buzzer sound occurs right before a shock. After they have sufficiently learned that the buzzer predicts shock, I want to also teach them that a light turning on predicts the shock. If I present the buzzer and the light together, the rats won’t learn anything about the light, because the buzzer already told them everything they need to know. The light did not provide them with any new information. This is called “blocking” because the previously learned association of the buzzer “blocked” learning about the light. To take it one step further, we could even change the aversive event and still witness blocking. In other words, I could teach the rat that the buzzer predicts shock, and then use the buzzer to block learning that a light turning on will predict getting poked with a needle. The rat will not learn about the light/needle relationship because prior learning about the buzzer “blocks” it. The motivational representation is “buzzer predicts something bad is going to happen.” It doesn’t really matter whether the bad thing is a shock or a needle poke.

Sisyphus was the King of Ephyra, and he had a reputation for defying the Gods and being a bit of a trickster. One of his best known exploits came at the end of his life when Hades, the God of the Underworld came to claim him, bringing along a pair of handcuffs. Sisyphus, in all his cunning and mischief, managed to persuade Hades to demonstrate the handcuffs on himself. Sisyphus further took advantage of this turn of events by locking the handcuffed Hades in his closet.

Eventually Sisyphus’ shenanigans caught up with him and he was brought to the underworld to receive his eternal punishment. For all his transgressions, he was condemned to an eternity of rolling a massive boulder up a hill. What made this especially torturous
was not that the hill was infinitely tall; in fact by exerting all his strength Sisyphus was able to reach the top.  However, the moment he reached the peak and was ready to rest and rejoice in his accomplishments, the darn boulder rolled right back down to the bottom. Sisyphus, tired and frustrated, had to start the process all over again. And on it went for eternity….

Now, for lack of a clever segue, I’m going to abruptly shift gears. But don’t let the tale of King Sisyphus slip too far from your mind.

Negative reinforcement is one of the most widely used and versatile aspects of how animals learn. Technically speaking, negative reinforcement refers to the elimination of a stimulus (generally unpleasant), for the purposes of encouraging or strengthening of behavior. In dog training, negative reinforcement refers to when the dog learns to turn off (or escape) an unpleasant sensation, and later learns to avoid the unpleasant sensation altogether by responding to a specific cue.

Used properly, negative reinforcement can strengthen and solidify your dog’s response to known commands, and make that response far more reliable and resistant to extinction. The key, however, is to learn to use negative reinforcement properly. An incorrect understanding of negative reinforcement can make training stressful for the dog. At best, using negative reinforcement incorrectly can simply slow down your training progress and limit the overall reliability of the results.

While there are many mistakes that are commonly made when it comes to the use of negative reinforcement (which I will refer to as R-), I would like to use the story of King Sisyphus illustrate one of the most common ones:  During the initial conditioning, or instructional phases of training, when the dog is learning how their actions can control the stimulus (or pressure), no sooner than the dog completes the task asked of it, then they are instantly released and/or given another command and the dog has to escape the pressure again.  

To illustrate by way of example, let’s take the early stage of remote collar conditioning where the dog learns to go his bed in response to the stimulation*.  The trainer presses the button on the transmitter on a low setting (only a mild tickle or annoyance to the dog), and then guides the dog to his bed. As the dog goes to his bed, the trainer releases the button and the dog is praised and rewarded. Then, after only a brief moment, the dog is released and the exercise is started again (the trainer presses the button, guides the dog etc.).

What we must remember is that it is the cessation of the collar pressure that is reinforcing to the dog. In order to really take advantage of this reinforcement, the dog needs a moment to enjoy his accomplishment and the sense of relief and relaxation that comes with it. In other words, when the dog successfully removes the stimulation, give them a minute to savor it.

When we drill our dogs with a rapid succession of commands during R- training, we are essentially giving our dogs the same fate as Sisyphus. However, training should be a fun and enjoyable experience for the dog. The “curse of Sisyphus” erodes the value of the reinforcement, thus eroding the dog’s desire to work with us, causing them undue frustration, and slowing down our progress.

Don’t give the dog the curse of Sisyphus.

Moreover, the more motivating the stimulus or pressure is, the more important it is to give the dog this extra bit of time.

After all, if Sisyphus was given a chance to sit down and catch his breath between boulder rolls, perhaps an extended break at lunch for a Panini and a glass of wine, and two solid days off on the weekend, maybe his fate wouldn’t have been so torturous (heck, it’s just a solid days work!).

In addition to potentially causing undue stress during training, we may also be missing out on one of the potential benefits of negative reinforcement training.

For those with just a casual interest in training, you can probably stop here. For those dog nerds like myself, you may want to read on, I’m going to get all sciency for a moment.

As stated earlier, negative reinforcement training ultimately has two components. First, the dog must learn to turn off, or “escape” the pressure when they feel it. Second, they learn to avoid it all together by responding to a predictive cue (i.e. our command).  One of the unique and desirable qualities of this later avoidance  learning is that once the dog learns how to avoid the pressure, they continue to do so for many repetitions without needing to be exposed to the pressure again. In fact, done properly, this type of learning is one of the most resistant to extinction.

Early researchers postulated that what was maintaining the dog’s response in the absence of actual pressure was a classically conditioned fear response when the cue is given. This seems to make sense. The dog hears a command, and responds out of fear of the consequence for not responding. The problem was that the evidence simply did not support this theory. Dogs wear their emotions on their sleeves, and they are terrible liars. What researchers observed was that when dogs were properly conditioned through negative reinforcement and avoidance learning, not only did they respond reliably, but they did so with very happy and relaxed dispositions.

More research and a new theory were needed to explain this phenomenon. Along came the safety signal hypothesis. Several researchers (see M.R Denny, R.G Wiesman/J.S Litner, and D.F Tortora) recognized that after the removal of pressure, the dogs experienced a sense of relief and relaxation. Further, as the dogs learned to successfully avoid pressure, any potential unpleasant emotions faded quickly, but the sense of relief and relaxation remained. Thus it is the pleasant emotions of relief and relaxation which act as reinforcement, and account for the dog’s disposition and the continued maintenance of the desired behavior.

In fact, M.R Denny noted that the experience of relief occurs 3-5 seconds after the cessation of pressure, and lasts for 10-15 seconds, whereas relaxation requires approximately 2-5 minutes to produce full benefits**. He also noted that the effects appear to double when the dog experiences both relief and relaxation as opposed to just relief by itself.

In other words, if you give at least 2-15 seconds between reps, the dog experiences some reinforcement, but it if you give a full 2-5 minutes, the experience of reinforcement can effectively double.

What this means is that by giving ample time between repetitions during escape/avoidance training, not only are you avoiding giving your dog the curse of eternal damnation (a bit of an exaggeration I know), but you are doubling the pleasurable aspects of the training.

We can take advantage of this extra time. Research has shown that we can condition other signals to be associated with this sense of relaxation. Thus praising and interacting with the dog during this time can increase the value of your praise and help establish your interaction as a source of safety and comfort. The latter is immensely valuable for professional trainers who are regularly working with dogs with whom they are relatively unfamiliar.

Lastly, remember that this principle doesn’t only apply to leashes and collars. For instance, in the rehabilitation of dogs with social anxieties we are often working on how to relieve social pressures in appropriate ways. Taking a bit of extra time between exposures can help to amplify your results. The same applies to exposure to other forms of fear, phobia and anxiety as well.

Training with any kind of pressure is a responsibility, not a right. If you are going to do it, every effort should be made to do it well.  Avoiding the curse of Sisyphus is just one of the many ways you can ensure you get the most out of your training.

-Tyler Muto

*I recommend training dogs initially with the use of positive reinforcement techniques, and utilizing the electronic or remote collar only to solidify and reinforce the previously established training.

** Denny specifies that relief involves a strong autonomic factor, whereas relaxation involves striatial muscles and various motoric components.

References:

Denny M.R. (1976). Post aversive relief and relaxation and their implications for behavior therapy. J Behav Ther Exp Psychiatry, 7: 315-321.

Denny M.R. (1983). Safety catch in behavior therapy: Comments on “safety” training: The elimination of avoidance motivated aggression in dogs. J Exp Psychol Gen, 112: 215-217.

Lindsay S.R. (2000). The handbook of applied dog behavior and training. Vol 1,  295-296.

Tortora D.F. (1983).  Safety training: The elimination of avoidance motivated aggression in dogs.  J Exp Psychol Gen, 112: 176-214.

Weisman R.G. and Litner J.S.  (1969). Positive conditioned reinforcement of Sidman avoidance in rats. J Comp Physiol Phychol, 68: 597-603.