Thank you to Jennifer Titus of CARE for Reactive Dogs for editorial advice. All errors and awkward moments are mine alone.
Citing “stressed-out R+ dogs” in an argument is an old chestnut that comes around regularly. The writer usually describes a training session he or she witnessed where a dog being trained with positive reinforcement was exhibiting fear or stress. The goal of sharing this description generally seems to be to blur the real differences between training that is based on positive reinforcement (R+) and training that is based on escape, avoidance, and punishment. Sometimes it is a feeble attempt to argue with the ranking of methods in assessments such as the Humane Hierarchy.
Cherry-picking a moment out of any dog’s life to support a general point about methods is tempting but is not effective argument.
The “Stressed-Out” R+ Dog
So let’s consider the stressed-out dog in positive reinforcement training. What are some possible causes of stress in an R+ training session?
When using positive reinforcement, some metrics we use to assess the skill of the trainer and the effectiveness of the training are timing, criteria, and rate (or sometimes magnitude) of reinforcement. Let’s start our analysis there.
Bad timing can cause the dog some stress through lack of clarity. The trainer is marking and rewarding some incorrect behaviors while sometimes failing to reinforce some correct ones. If she cleans up her act and stops reinforcing the wrong stuff, the dog will go through an extinction process. Depending on the trainer’s skill, this can be stressful.
Raising criteria too fast means a higher failure rate. This can also cause some frustration. So while this is in an R+ training environment, what you have when you raise criteria too fast and the dog doesn’t do anything reinforceable is, again, an extinction problem.
If the rate of reinforcement is too low, you can actually put the desired behavior on extinction. So you may get a confused dog who starts throwing behaviors out of frustration, or a dog who will wander off and do something else more reinforcing, given the choice to do so.
Another stressor can be the use of negative punishment when the dog hasn’t learned the behavior. If the dog isn’t clear on how it can earn the reinforcer, it is frustrating to have it taken away contingently as it tries other things.
Note that none of the above errors is likely to hurt, scare, or startle the dog.
Two more types of stressors possible in an R+ training session are pressure of some type, and an accidental, momentary aversive. These two can indeed hurt, scare, or startle the dog, but are not linked to the positive reinforcement learning process.
- What I’m calling pressure could consist of anything in the environment, setup, or even mannerisms of the trainer that the dog would like to escape from. Is something too loud? Is someone pressuring the dog with his or her body? Is the dog being kept too close to something she is scared of? This type of problem comes from the unwitting inclusion of an aversive stimulus.
- Likewise, accidents happen, as they can in any training. A trainer might step on her dog’s tail during a stay, but again, this is an aversive accident, not an integral part of R+ training.
So our causes of stress are probably either technical mistakes on the trainer’s part or the presence of an unplanned or unrecognized aversive stimulus. Are these problems unique to positive reinforcement training? Absolutely not. They can happen in training based on aversives just as easily.
A Fair Comparison
Let’s compare apples with apples. Rather than focusing on the stressors in faulty positive reinforcement training, lets compare the net effect on the dog of R+ training vs. aversive-based training–with both done poorly. There is certainly no shortage of sloppy training done with aversive methods. I can find such a video on YouTube within a couple of minutes, and the trainer is often touting it as a success story.
So what happens to a dog being trained with escape/avoidance and punishment when the problems and errors I described above are present? Not only is the dog startled, hurt, intimidated, or at least irritated by the training itself, she will also be subjected to the additional stress resulting from trainer errors. Or she may experience aversives in addition to the ones the trainer is purposely using.
Here’s what it could look like.
- Bad timing: Imagine popping a dog’s collar when she is heeling perfectly, in addition to popping her when she makes an error.
- Changing criteria too fast: Imagine using duration shock to teach a dog to jump off a platform immediately after using it to teach her to jump on it.
- Unplanned aversive stimulus: Imagine teaching stays using your hands to force a sound-sensitive dog to hold her position while a delivery truck with a no muffler drives by.
Those make the possible stressors in R+ training look rather like small potatoes, don’t they?
A Real-Life Example of the Results of R+ Training with Errors
I will be the guinea pig. I have a video of my own training that demonstrates many of the stressors I listed above.
In this popular video of mine that demonstrates lumping, I raise criteria too fast for Zani. She gets visibly frustrated. You can see it around 2:25 in particular. She plants herself in front of me in a sit and makes what I call the “terrier frustration noise.” A sharp exhale through her nose. I don’t blame her.
In addition to the training errors that are the subject of the video, there are more. I often mark late. I mark and reinforce improper behaviors, both when she targets my bare hand instead of the tape, or does a “drive-by” and doesn’t connect at all.
My rate of reinforcement is not bad, but there are a couple of times when Zani is going through extinction, trying other behaviors, where I might have interrupted her sooner, or marked something approaching the right behavior.
My reinforcement placement is not thoughtful. I am generally tossing the treat in order to reset Zani, but think how much faster she could have gotten to the wall if I had treated in that direction instead of away from it?
Another criterion issue is my poor choice of tape color. Gray, even metallic, is not a good contrast on a tan/yellow wall. Zani probably couldn’t see it well.
Interestingly, there is a subtle aversive stimulus in the session as well, and I think we can see the effects of it on Zani’s actions. The tape on the wall is in a tight area. I think her reluctance to enter that small area (in other words, an aversive setup) is one of the reasons she targets the desk multiple times instead of going for the tape. She is extremely pressure sensitive and I am asking her to go by me into a tight little space. She tries to avoid it.
So in one video, we have many of the problems I listed above.
But even with the errors in the training and the slightly aversive setup, Zani hung in there with me and was wagging her tail in the last section. She successfully learned the behavior I was teaching and got 24 tasty food treats in the three minutes of training time shown. Not a bad rate at all, considering that there were two dry spells and also that she was spending a fair amount of time chasing down treats.
So here is a thought experiment. Imagine that instead of what you saw in the video, I used aversive methods to get the targeting behavior from Zani. You can imagine a combination of physical manipulation and body pressure, or a shock collar. No food in the picture. (If you are imagining Zani falling to pieces, that’s about right.) Now add to that multiple errors of timing and criteria, and an unwise setup that creates a tight space. How is Zani doing now?
That is a much fairer comparison of the results of different training methods.
The Proper Rejoinder
Evoking the scenario of the stressed-out R+ dog in argument invites the following response:
It’s a good thing the dog was being trained with positive reinforcement then. Adding training errors and aversive situations to any protocol can cause stress. Think how much worse it would have been if the dog were being deliberately trained with aversives to start off with!
The real illogic of the comment in the title is that in most examples described it’s the addition of aversive stimuli that creates stress. Blaming stress that results from the accidental inclusion of aversive stimuli on the process of positive reinforcement training is not only illogical; it’s a cheap shot.
Conclusions from Examples
Drawing conclusions from examples is tricky, and can easily lead to the logical fallacy of “missing the point.”
A couple of the valid conclusions that can be drawn from the “stressed-out R+ dog” scenario are that some positive reinforcement trainers lack mechanical or observational skills, and that it is possible for other learning processes besides positive reinforcement to be going on when we are trying to train with R+.
What the scenario doesn’t support is the idea that there is some unknown dark side intrinsic to positive reinforcement training, or that there are characteristics of training methods that are immune to analysis through learning theory, or that stressors from lack of skill happen only in R+ training, or that training based on the use of aversive stimuli can make for a happier dog.