This post includes discussion of animal experimentation from the 1950s and 1960s using shock. It is unpleasant to contemplate. But to me, it makes it even worse that the knowledge gained by those studies is not widely known. Studying that literature gives one a window on how punishment works. I hope you will read on.
The studies I cite are all included in current learning theory textbooks, and my descriptions are in accord with the textbooks’ conclusions. The results are different from the common assumptions about punishment.
This is a typical response to application of a mild-to-moderate aversive. I created this graph because I don’t have rights to the ones in textbooks–but they look like this. The X-axis represents sessions over time. The Y-axis show the ratio of behavioral decrease. The shape of the graph roughly correlates to the frequency of the behavior and shows that the suppression of behavior was only temporary.
I’ve written a lot about making humane choices in training and about the fallout that accompanies aversive methods. But there are other problems with the use of aversives besides the immediate fact of hurting, scaring, or bothering your dog. It turns out that using positive punishment is tricky.
In the term positive punishment, positive doesn’t mean “good” or “upbeat.” In learning theory usage it means the type of punishment in which something is added and a behavior decreases. The added thing is something the animal wants to avoid. If every time your dog sat you shocked her, played a painfully loud noise, or threw something at her, your dog would likely not sit as often. Those things I mentioned would act as “aversive stimuli.” If the dog sat less after that, then punishment would have occurred.
There is another type of punishment called negative punishment. It consists of removing something the dog wants when they do something undesirable. I’m not discussing that type of punishment in this post. For the rest of the post, when I refer to punishment, I am referring to positive punishment.
The Punishment Callus
Some trainers and behavior professionals warn about something called the punishment callus. A punishment callus is not a physical callus. It is one name for the way that animals (including humans) can develop a tolerance for an aversive stimulus. When that tolerance is developed, that stimulus does not decrease behavior. It is not an effective punisher. The animal has become desensitized to punishment.
This is not just a piece of folklore. It has been demonstrated repeatedly in studies, and it happens way more often than we realize in real life. I’m going to describe some of the research.
The first thing that happens in most punishment experiments is that the animal is taught a behavior using positive reinforcement. The pigeon learns to peck a disk to get some grain. The rat learns to press a lever or run down a chute to get food. There will be dozens, hundreds, or even thousands of repetitions. Then, after the behavior is strong, the researchers introduce punishment. This is usually in the form of shock. The shock is generally contingent on the animal touching the food or performing the behavior that gets access to the food.
At first glance, this seems weird, not to mention wildly unfair. Why would they be starting off a punishment study with reinforcement? Then why would they punish the same behavior?
Think about it a little and it makes sense. You can’t use punishment if you don’t have a behavior to punish. Reinforcement is what makes behaviors robust. You can’t measure the effects of unpleasant stimuli on a behavior unless you have a strong, consistent behavior to begin with.
In some studies, they cease the reinforcement after the punishment starts. In others, the reinforcement continues. In these experiments, the animals and birds get shocked for trying to get their food in the same way they learned to get it through many repetitions of positive reinforcement.
But this is not at all unique to lab experiments. A hard lesson here is that we do the same thing when we set out to punish a behavior. Animals behave because they get something of value (or are able to escape something icky). The behavior that the dog is performing that annoys us is there because it has been reinforced. It didn’t just appear out of the blue. So if we start to punish it, the animal is going to go through the same experience that the lab animals did. “Wait! This used to get me good stuff. Now something bad happens!” And punishment and reinforcement may happen together in real life, just as in some of the studies.
How We Imagine Punishment to Work
I think most of us have an image of punishment that goes something like this:
The dog has developed a behavior we find annoying. Let’s say he’s knocking over the trash can and going through the trash. The next time Fido does that, we catch him in the act. We sternly tell him, “No! Bad dog!” Or we hit him or throw something. (I hope it’s obvious I’m not recommending this.) The next time he does it, we do the same thing. In our minds, we have addressed the problem. In our mental image, the dog doesn’t do it anymore.
But. It. Doesn’t. Work. That. Way.
Real life and science agree in this. It’s much harder than that to get rid of a reinforced behavior.
Many studies show that the effectiveness of a punishing stimulus correlates to its intensity (Boe and Church 1967). The higher the intensity, the more the behavior decreases. Very high-intensity punishment correlates to long-term suppression.
Skinner was one of the first to discover that low-intensity punishment was ineffective. He taught rats to press a bar to get food. Then he discontinued the food and started to slap the rats’ paws when they pressed the bar. For about a day, the rats whose paws got slapped pressed the bar less than a control group. Then they caught up. Even though they were getting slapped, they pressed the bar just as often as the control rats (Skinner 1938). Other early punishment studies also used mild punishment, and for a while, it was assumed that all effects of punishment were very temporary (Skinner 1953). This was determined to be incorrect in later studies with higher intensity aversives.
Dog owners who try to use low-level punishment are faced with an immediate problem. Ironically, the problem is born out of a desire to be kind. Many people do not feel comfortable doing anything to hurt or startle their dogs, but these are the methods they have been told to use. So they figure that they should start with a very low-intensity action. They’ll yell just loud enough to get the dog to stop. They’ll jerk the dog’s collar just enough to interrupt the pulling on leash. They’ll set the shock collar to the lowest setting.
But if a behavior is valuable enough to a dog (i.e., it gets reliably reinforced), a mild punishment will barely put a dent in it. It may interrupt the behavior at the moment and suppress it for a short time, and people are fooled into thinking it will continue to be effective. But it almost certainly won’t.
So the next thing the humans do when the dog performs the behavior is to raise the level of the punishment a bit. They yell louder, jerk harder, or turn up the dial on the shock collar.
Lather, rinse, repeat. If this pattern continues, the humans are successfully desensitizing their dogs to punishment. The desensitization can continue up to extremely high levels of punishment. That is the punishment callus, and it has been excruciatingly well documented in the literature.
In one study (Miller 1960), hungry rats were trained to run down a walled alleyway to get a moist pellet of food at the other end. The rats repeated this behavior many times as they got acclimated to the setup. Each rat’s speed of running down the alley was recorded as they gained fluency. The behavior of running down the alley was reinforced by access to food. This continued (without punishment) until the rats were determined to have reached their maximum speed.
A shock mechanism was then initiated so the rats’ feet would get shocked when they touched the moist food. The rats were divided into two groups. They were referred to as the Gradual group and the Sudden group, indicating the way the shock was introduced. The Gradual group started with a shock of 125 Volts, which caused virtually no change in behavior. The shock was raised in each subsequent session. The rats’ speed slowed down somewhat each time the shock was raised. Then it recovered and leveled off as they got accustomed to the new intensity. The shock was raised through nine increments up to 335 Volts.
The rats in the Sudden group didn’t experience the gradual shocks. Their first introduction to the shock was at 335 Volts. Their movement down the alley slowed drastically. Often they would not touch the food.
In the last 140 trials (5 trials each for 28 rats total) the results were telling. Out of 70 trials at 335 Volts for the rats in the Gradual group, only 3 trials resulted in the rat not going all the way to the food. In the Sudden group at the same voltage, 43 trials, more than half, resulted in the rat not going all the way to the food.
To repeat: These two groups of rats responded differently to shocks of the same high voltage due to how the shock was introduced.
Now take careful note of the differences in their behavior:
The [subjects] in the Gradual group flinched and sometimes squealed but remained at the goal and continued to eat. Those in the Sudden group seemed much more disturbed, lurching violently back, running away and crouching a distance from the goal (Miller 1960).
There’s the clincher. At 335 Volts, some rats were still approaching the food and eating while getting shocked. In other words, those behaviors were not effectively punished. For the other rats, the behaviors were definitely punished–and the rats were traumatized.
So there you have it. Two of the most common outcomes of using punishment are:
- a spiral of ever-increasing punishment intensity that the animal learns to tolerate; or
- a shut-down animal.
This information has been available for 50 years. Yet aversive techniques are still casually recommended to pet owners with no education in learning theory, no exposure to the mechanical skills involved, and most important, no clue of the harm to the animal.
The Resilience of Behavior
One of the things I finally “got” about punishment as I studied the graphs in these studies is that complete cessation of a behavior is rare. Again, our mental image of the results of punishment is incorrect. In the Miller experiment, the traumatized rats in the Sudden group did sometimes approach and eat the food despite intense punishment. The rats in the Gradual group consistently did so.
The rats in the Gradual group correspond to dogs who are trained with gradually increasing punishment. They acclimate and the behavior continues. They get a punishment callus. The rats in the Sudden group probably resemble the heavily punished dogs I describe in my post Shut-Down Dogs, Part 2.
One more thing about the graphs. When punishment is initiated or taken to a higher level, there is an immediate drop-off of behavior. It’s usually of short duration. The rate of behavior generally rises back up again. This is what I modeled in the diagram above. You can see a bunch of these graphs in the Azrin study linked below.
Increasing the punishment intensity seems to have the same general effect as the initial addition of punishment. In both instances, the new punishment intensity produces a large suppression at the moment of changeover, with substantial recovery after continued exposure to this new intensity. Only at severe intensities of punishment has further increase failed to produce an abrupt decrease in responding (Azrin 1960).
One of the tragedies of this pattern in dog training is that the drop-off causes the human to believe the punishment is working. The human’s act of raising the level of the punishment is reinforced.
The deliberate use of positive punishment as a training method is already ruled out of consideration for most positive reinforcement-based trainers. This is because of humane concerns and its known fallout. But I believe it is also important for us to know how difficult it would be to use effectively and that it does not work the way most of us imagine it to. We can see desensitization to punishment all around us once we learn of its existence. My takeaway from the studies is how vastly superior and straightforward it is to build behavior in our pets than to try to squash it down.
Note: Please don’t quote this article to claim “punishment doesn’t work.” High-intensity punishment does work. But it has unacceptable side effects that are destructive of our dogs’ happiness and wellbeing, not to mention their bonds with us.
Azrin, Nathan H. (1960). Effects of punishment intensity during variable‐interval reinforcement. Journal of the Experimental Analysis of Behavior 3(2), 123-142.
Boe, E. E., & Church, R. M. (1967). Permanent effects of punishment during extinction. Journal of Comparative and Physiological Psychology, 63(3), 486-492.
Miller, Neal E. (1960). Learning resistance to pain and fear: Effects of overlearning, exposure, and rewarded exposure in context. Journal of Experimental Psychology 60(3), 137-145.
Skinner, B. F. (1938). The behavior of organisms: an experimental analysis. Appleton-Century. New York.
Skinner, B. F. (1953). Science and human behavior. Simon and Schuster.
Copyright 2016 Eileen Anderson
Easy tweets! (They include a link to this post)
- What's a punishment callus and why do we need to know?
- Why is using light punishment ineffective in the long run?