I am mystified by one particular argument of those who use protocols for fearful or reactive dogs other than desensitization/counterconditioning (DS/CC). These other protocols often use negative reinforcement; if not that, then sometimes desensitization without counterconditioning; sometimes extinction; sometimes habituation.
It’s OK to expose the animal to a trigger at a potentially aversive level as long as we are not the ones who put the aversive there for them to be exposed to. We’re not adding an aversive; it’s already there.
Where the aversive came from is ethically irrelevant, since the trainer makes a choice whether or not to use it, however it got there. Most would agree that such a use is an ethical choice, to be carefully considered.
So the fact that people are still mentioning this irrelevancy about “who put it there” seems like a lot of hand waving to shoo away the real issue: choosing to use an aversive.
But wait–in case it matters–how did it get there?
How It Really Got There
My hand, my voice, my phone.
I have a formerly feral dog with whom I have been working for a few years, gradually getting her socialized to people, and making lovely progress with DS/CC.
Even though my goal is to keep the triggers (people, in her case) under the threshold of aversiveness, I realize that I am dealing with potentially aversive situations when we go out into the world. And I arrange for and seek out those situations for her sessions. For instance, I make phone calls at times to arrange for a controlled session with a person unknown or partially known to her.
If I do this and blow it and let her get too close or stay too long, I have exposed her to an aversive. How’d it get there? Me! Entirely through my choices! I arranged it. I deliberately sought it out with her. I made the phone call, drove my dog to the meeting place, and exposed her to the trigger. I added it to her environment, or added her to an environment where I knew it to be.*
People following any protocol generally arrange for triggers to be present in this way, including people, dogs, specific things like people on bikes or scooters, or other animals. So if someone is doing any type of exposure treatment, how can they claim that they are not responsible for the aversive being there? Did the Tooth Fairy bring it? Can their dog pick up the phone and drive the car?
It is not logical to claim to have nothing to do with the aversive being in the environment if you planned it, arranged for it, or sought it out in the first place. And that includes stealth sessions. If you are out there looking for triggers to use without their knowledge, you are still the one choosing to expose your dog to them. Finding = adding.
Empathy
You can probably detect that I find this irritating, but I seek to look at it in an empathetic way.
I have been reading some posts by Behavior Adjustment Training (BAT) practitioners in particular who express that they feel attacked and beleaguered by questions about negative reinforcement and humane training attached to their protocol. I get that they feel pushed into a corner.
I can empathize with that. Here is something you believe in, and people are asking difficult, pointed questions about it. Sure, anybody would be defensive. As a blogger, I have to deal with all levels of criticism. Even the most reasonable of criticism hurts.
There are people who react to these questions with dignity, though. They say yes, they are using negative reinforcement at times if they use certain protocols. They have thought it out, see good results, usually use other protocols as well, and are ultra careful about side effects. They don’t play like the presence of the aversive has nothing to do with them. Although I may not agree about all methods these folks use, I can appreciate their transparency and honesty about the science.
But it really worries me that there are still people who claim not to be responsible for getting the aversive into the environment. If they are trying to elude responsibility for that, even though it’s completely a side issue, what else are they willing to overlook, justify, or push out of their minds?
Thank you to all the people who do their best not to adjust the science (or even basic logical thinking) to justify their own preferences.
Coming Up:
The Girl with the Paper Hat Part 2: The Matching Law
Punishment is not a Feeling
Why Counterconditioning Didn’t “Work”
How Skilled are You at Ignoring? (Extinction Part 2)
* I am not confusing positive punishment and negative reinforcement, here. To use negative reinforcement, there has to be an aversive in the environment to be removed or escaped. We’re talking about how the aversive got there in the first place.
5/25/14 Addendum
This post is an urge to be honest about one aspect of the use of aversives. I believe that all trainers, regardless of method, should be honest about their training choices and philosophy. You do it: own it. That’s the message in a nutshell. And I directed it to an argument that I believe does the opposite of “owning it.”
However, one of the common responses I have gotten over the past week is comparisons of the ranges and setups of DS/CC protocols and those using negative reinforcement, often in an apparent attempt to minimize the differences.
I have previously provided a webinar and a movie on the differences and similarities of the major protocols for addressing fear in animals, with particular emphasis on their ranges and setups.
To review a few relevant points: Debating who starts further from the stimulus is a moot point. No matter how far away you start, you are required to go into the aversive range for a negative reinforcement protocol to work. In desensitization and counterconditioning you have no need to cross into the range of stimulus aversiveness in order to get effective results. In R-, aversive exposure is necessary. The protocol depends on it. In DS/CC, aversive exposure is by accident and hopefully rare. That is an important distinction between DS/CC and negative reinforcement-based protocols.
The other important distinction is that you can get a positive conditioned emotional response from DS/CC. With DS/CC and negative reinforcement there are two very different types of learning going on.
Zani keeps her eyes on me a large part of the time
Some dog trainers who use tools such as shock, prong, or slip collars, or startle the dog with thrown objects or loud noises, claim that these things are done only to “get the dog’s attention.” They may further insist that the dog is not hurt, bothered, or scared.
This sounds like the Holy Grail of dog training. It’s the Magical Attention Signal! It can get your dog’s attention, get him to do something, or stop doing something, all rolled into one. You don’t have to use those pesky treats or toys, and it certainly doesn’t hurt or bother the dog!
Gosh, who wouldn’t want that? Life would be so easy with the Magical Attention Signal!
Unfortunately, the Magical Attention Signal is utter nonsense.
Nonsense Clue #1
We almost never want only our dog’s attention.
Let’s say that your Magical Attention Signal is tossing a lightweight coaster towards your dog. Your dog doesn’t particularly care about coasters. (Folks with disc-crazy dogs, hang on, I’ll get to you.) So you toss the coaster and the dog looks up. Yay, success! You’ve got the dog’s attention. Mission accomplished!
Um, no. Of course we don’t want only the dog’s attention. When we want their attention, it’s for a reason. The reason is almost always one of two things: to get them to do something or stop doing something. Getting their attention is only the bare beginning.
Nonsense Clue #2
Non-predictive stimuli are subject to habituation.
Habituation: A decrease in response following repeated exposure to a non-threatening stimulus.–Klein, Thorne: (2006) Biological Psychology
Virtually all of us have experienced habituation to something that was initially novel. Let’s say you move to a new house. It’s barely within earshot of an elevated train or metro track. When you first move in, you notice the sound of the train regularly: maybe a whistle, or just the rumble.
At first it gets your attention. However, it gradually sinks in that there are no relevant consequences to that sound for you. The train schedule doesn’t affect vehicle or pedestrian traffic in any way. You don’t have to arrange your day around it. None of your loved ones ride it or work for the railroad. The noise is faint and there aren’t any noxious fumes. It doesn’t predict danger. In fact the train noise doesn’t predict anything for you, good or bad.
So what happens to the stimulus of train noise? Habituation. You stop noticing it. It fades into the background. Our minds sift through stuff all the time to determine predictors of good and bad consequences. Things to seek and things to avoid. Low-intensity stimuli with no consequences fall to the bottom of the priority stack.*
Animals, including dogs, do this sifting too. Some dogs are noticeably good at it, like my Clara, who often knows my behavior patterns better than I do. And when you think about it, loads of the stuff we humans do has some kind of predictive value to our dogs. Turning on the TV. Getting dressed. Opening the refrigerator. Sighing. Even pulling down a book from the bookshelf.
I had a hard time thinking of a regularly occurring non-predictive stimulus in my life with my dogs, but here’s one. For my own dogs, the automatic switching on and off the the central heating and air means nothing. They hear it intermittently all day long, but it is just background noise to them. If the temperature weren’t well controlled, or if one of them was extremely hot- or cold-natured, she might start to notice and take the opportunity to go lie next to the air vent. Then the sound of the heat and air clicking on would become predictive, and start rising up in the stack of “things to notice.”
So the upshot is that if we want our dogs to keep responding to a stimulus, it generally has to be quite strong in itself, or have a consequence. Good or bad, your choice. But not neutral.
What Really Happens?
So how might our thrown coaster stimulus work? We have determined that if it were non-predictive, it probably wouldn’t continue to get the dog’s attention. So if it works consistently to get the dog’s attention, what’s going on?
There are four relevant possibilities:
Yay!
Having a coaster suddenly land nearby could be intrinsically desirable to the dog. Maybe you have a loopy goofy retriever and he loves having something thrown near him, even if it’s just a coaster. He probably grabs it and plays with it. However, it may have failed as an attention-getting device. He’s playing with the toy, not looking up at you. And if you threw it when he was doing something you didn’t like, you would have accidentally reinforced the bad behavior. “Yay! I got a toy when I barked at Grandma!” (This can happen when people try to interrupt or punish with squirt bottles. Some dogs think being squirted is wonderful.)
Yipes!
It could be intrinsically aversive to the dog. I would wager that this is the case for many dogs, especially at first. Something flying through the air, appearing suddenly close and making a noise could startle them. Some dogs would habituate to it, and some might never do so. If they didn’t habituate, this could work as a way of getting your dog to pay attention to you. There’s a big drawback though: that startled, fearful response would likely become associated with you. You become the scary person who throws stuff.
It could predict something desirable for the dog.
Good stuff coming!
Maybe your dog is not turned on by coasters. But what if, every time you tossed the coaster, you then threw a treat or a toy? The dog would quickly learn that the coaster toss predicted great stuff (in the same way that clickers are typically used). If you were to toss the coaster a number of times, pairing it with good stuff, after the dog learned to the association you could use it to interrupt undesirable behavior. This is the principle of the “positive interrupter.” But you don’t have to throw anything. If you are close enough to toss a coaster, a simple noise or word would do. And it’s pretty clear that the promoters of the Magical Attention Signal are not using it this way.
Oh oh!
It could predict something aversive for the dog. Like Cesar Millan’s “Tsst!,” it could predict a kick or a jab in the neck. Or something less dramatic, like being yelled at or handled roughly. This might not have been the trainer’s or owner’s intent from the start. But if the startling effect of the thrown coaster wears off (version #2), a stronger consequence will need to be added. Then the thrown coaster would become either a punishment marker (“Fido, you are about to get it”) or a threat (“Fido–hop to it or you are going to get it”). This is also how most shock collar training works. When a trainer brags that he uses only an extremely low, non-aversive level, that is because the dog has already been taught that the shock can easily be escalated if he doesn’t comply. Otherwise we are left only with the Magical Attention Signal.**
By the way, #4 illustrates the concept of the “punishment callus.” One of the paradoxical problems with using an aversive is that most people want to start out light. But if you try that on strongly entrenched dog behaviors like barking, digging, or jumping up, the behavior may well prove to be too strong. Then you will be in the position of having to escalate. And often the dog’s ability to tolerate the aversive will escalate right alongside.
No Magical Attention Signal
Many promoters of aversive tools to use in dog training don’t want to say that they ever hurt or scare or startle or nag or bully dogs. And our mythology about dogs is so strong that most of us want to believe them. Hence, the lure of the magic signal that works all by itself, with no other consequences. I hope this post will bolster your “nonsense detector.” Behavior is driven by consequences. If no change in consequences occurs, there is no reason for a behavior to change.
If someone says that Tool A, B, or C is designed to “get the dog’s attention,” ask them what happens next. Once they get the dog’s attention, how do they actually get the dog to do something or stop doing something? Also ask them what happens if the “painless” tool fails to get the dog’s attention.
The Magical Attention Signal is not going give any lasting help on its own. Learning theory and common sense (if only we could apply it when we think about dogs!) tell us that behavior has consequences. We take actions for a reason. We act to get stuff we want. To avoid stuff we don’t like. All creatures with a brain stem, and more primitive creatures as well, from what I hear, do this.
Rapt attention in the back yard
But the good news: if you keep conscious control of the reinforcers in your life with your dogs, use those reinforcers to strengthen behaviors you like, teach alternatives to behaviors that you don’t, you will have a head start on getting great attention from your dog.
All photos except the one of my dog Zani and the one with my three dogs are courtesy of Wikimedia Commons. The boxer photo was cropped.
**We could also add, looking at the four quadrants, that the thrown coaster could predict the cessation of something aversive, or the removal of something good. But I think these are pretty unlikely usages.
Summer at the instant she reaches threshold of fear
I have made a movie about thresholds in dog training. It gives a quick overview of the work that I presented in my webinar for the Pet Professional Guild. (Click here for a complete script of the video; or expand the audio (only) transcript below the video.)
The threshold webinar is still available as a recording ($10 members/$20 non-members of PPG) and I encourage anyone who is interested in thresholds to view it.
If you are a visual learner, the movie will probably be helpful. I spend a lot of time explaining the diagrams, and have an animation of what happens to the thresholds as we train. The movie also has video examples of dogs and stimuli over the thresholds. (Plus it has a threshold of hearing test! How cool is that?**)
Threshold Movie Script
[Dogs barking]
>>EILEEN ANDERSON:
Have you ever heard a dog trainer use the term “over threshold” and wondered what it meant?
A threshold is the point or level at which something begins or changes. That’s the standard dictionary definition. But the interesting thing is that there are actually three physiological and psychological thresholds that are important when we are training our animals.
The first threshold we need to know about is the sensory threshold as defined in psychology. Here’s a definition: “The faintest detectable stimulus, of any given type, is the absolute threshold for that type of stimulus.” Have you ever heard the term, “threshold of hearing?” Right now during this slide I am playing a high frequency hum. Can you hear it? If not, it is under your threshold of hearing for that frequency. If you can hear it, it is over the threshold.
The sensory threshold is involved when our dogs are able to see, hear, or smell something new in their environment.
Another threshold is the threshold of reactivity or fear. This is the one people usually mean when they say their dog is over threshold. The most general definition of this threshold is the point at which the sympathetic nervous system responds when the animal is afraid. This causes chemical changes in the body and overt behaviors usually falling into the categories of fight, flight, or freeze.
Dogs who are aggressing are generally over the threshold of fear. Here are two other examples of dogs over that threshold.
[vet clinic noises]
>>EILEEN:
She’s panting, but it’s not hot.
She’s hyper vigilant.
Trying to escape, or hide.
And trembling.
This dog is practically paralyzed with fear.
But there’s one more threshold, and it’s located behaviorally between the other two. If one threshold is where the dog sees something, and another is where the dog freaks out about it, what’s in between?
The point at which the thing becomes aversive, where the dog starts to be uncomfortable with it.This could be called the threshold of stimulus aversiveness.
Here is an example of a dog in a situation where a stimulus is over the threshold of aversiveness. In other words, she is stressed about something in her environment, but so far she is holding it together.
[Neighborhood noises: siren in distance, children talking, birds, a sudden thump]
>>EILEEN:
She repeatedly licks her lips and looks behind her.
She’s responding to my cues, but she’s worried about the noises.
In my webinar on thresholds in dog training, I made diagrams of these thresholds, and discussed where each of our common training protocols falls among the thresholds. Here is a summary of those diagrams.
The black line represents distance from or intensity of the stimulus.
All three of the protocols discussed here take place over the threshold of stimulus perception, since the animal has to perceive the stimulus to learn about it.
The combination of desensitization and counterconditioning is correctly practiced under the threshold of stimulus aversiveness.
Protocols that use negative reinforcement straddle the threshold of stimulus aversiveness. The animal is exposed to the stimulus at an aversive level, and escape from the aversive level of the stimulus is used as a negative reinforcer for appropriate behaviors.
The closest proximity to the aversive stimulus may be more or less than I show here; the important point is that negative reinforcement protocols have to cross the threshold of stimulus aversiveness to work.
Flooding takes place at or above the threshold of fear.
The thresholds aren’t always spaced out nicely. For example, if the threshold of perception and the threshold of aversiveness are very close together in space, a trainer using desensitization/counter conditioning would probably not use distance as the initial way to keep the stimulus non-aversive. The trainer would probably use a different form of the stimulus first. This configuration of the thresholds is probably common with wild animals.
Likewise, if the threshold of stimulus aversiveness and the threshold of fear are very close together, a negative reinforcement protocol would be very difficult to perform without risking flooding.
Finally, the thresholds move because of environmental factors, the animal’s stamina and psychological state, and of course as we train. This is what we hope will happen as we train.
For more information on thresholds, please see the links to my webinar and blog in the video description. Thanks for watching!
Coming Up:
BarkBusters: Myths about Barking
Surprising Progress on Thunderstorm phobia
Why Counterconditioning Didn’t “Work”
How Skilled are You at Ignoring? (Extinction Part 2)
** For the auditory people, musicians, and nerds among us (I’m all three): I used an iPhone app to generate a high frequency sinusoid (15.5 kHz) and recorded it for the movie. I used an oscilloscope app to make sure that the sound was playing during that part of the movie, through my own computer anyway. It’s just below my threshold of hearing. Younger people can probably hear it, if their computer speakers can generate it.
I didn’t give today’s post a cute title, because this situation makes me very, very sad.
There are some strange claims going around the dog training community. They are not being made by shock trainers, although I am sure they appreciate them. Instead I am hearing them from many people in the force free community. The statements minimize the problems that can be caused by using negative reinforcement.
There is truly a continuum in the severity in the applications of R-. In the human world, it can run the gamut from putting on a coat, to a staredown, to torture. Negative reinforcement happens a lot in the natural world, too, often at very low levels of aversiveness. So people are correct if they say that some situations are more aversive than others, or that using negative reinforcement is not always a catastrophe. The trouble begins when they make blanket statements–especially blanket incorrect statements–that include all negative reinforcement.
Following are two related versions of the statement about negative reinforcement that I keep seeing.
Version 1
The reason some trainers object to negative reinforcement is that when people add the aversive, there can be fallout.
This statement omits the majority of the problems known to accompany the use of negative reinforcement and aversives in general. The fact that an animal’s response to an aversive can get generalized to the handler is only one of the many problems with using negative reinforcement.
I rewrote the statement to be more complete.
The reasons some trainers object to negative reinforcement include that it employs an aversive, the association with the aversive can be generalized, it is on the undesirable end of the humane hierarchy, it is linked with reactivity and aggression, and has other undesirable side effects for both the animal and the trainer.
The main issue isn’t whether there’s a human wielding the aversive, it’s that an aversive is being used in the first place.
If the only problem with negative reinforcement were that the animal might make an association between the icky thing and the human, all that would be necessary to make negative reinforcement acceptable across the board would be to prevent the animal from making that association.
The shock trainers must be delighted whenever they hear this statement come from the mouths of force free trainers. If it were true, all they would have to do for their training to be acceptable would be to make sure the dog doesn’t know that they are controlling the shock. (And shock trainers with skill and knowledge of learning theory take care to do just that, by the way.) Poof! No more criticism of shock!
I know that this is not the intent of the force free trainers who are defending negative reinforcement. But as long as they make blanket statements about that quadrant, it is the logical conclusion.
It also strikes me as very self centered to mention only this particular problem with negative reinforcement. Really? It’s OK to deliberately use something unpleasant to get the dog to do stuff, as long as the dog continues to like us?
Version 2
Negative reinforcement is ethically OK as long as the handler isn’t the one who adds the aversive to the environment.
On the surface, this sounds like the same thing. But in general, the people who say this are discussing ethics, not behavioral fallout. I have seen probably a dozen people write that using an aversive that is “already out there” is ethically acceptable, while adding one oneself is not. It’s a tempting rationale, but there are some real problems with it.
Let’s go straight to examples on this one.
Photo credit: Wikimedia Commons
Let’s say my dog and I are out in the yard and it starts to storm. I notice that my dog is cowering at the door; she is scared of the thunder. Instead of letting her in immediately, I require that she sit and give me eye contact for 10 seconds. If she can do that, her reinforcement is that she gets to go in the house where she feels safer from the storm.
My dog and I are again in my back yard. I have bought a new sump pump for the crawl space in my house. I turn the pump on while my dog is watching. It will run for 2 minutes as a test. I notice that my dog is cowering at the door; she is scared of the pump sound. Instead of letting her in immediately, I require that she sit and give me eye contact for 10 seconds. If she can do that, her reinforcement is that she gets to go in the house where she can get away from the pump.
Now compare the two experiences for the dog. She is sitting there at the door trying to figure out how to get me to let her in, away from the scary noise. If the noises are equally aversive, the two situations are just the same.
I don’t see a difference ethically. The thunderstorm exposure is no more humane than the sump pump. In both cases I chose to use an aversive and required my dog to stay longer than necessary in a situation that scared her. And I did have another option in each case, one that is almost always ignored by people defending negative reinforcement protocols. I could have just let her in the house without requiring a particular behavior.
Natural vs Contrived Negative Reinforcement
There is a recognized difference between two types of reinforcement: natural (or automatic) negative reinforcement and contrived (or socially mediated) negative reinforcement. I have written a post about them. Paul Chance’s definition is as follows:
Natural reinforcers are events that follow spontaneously from a behavior… Contrived reinforcers are events that are provided by someone for the purpose of modifying behavior. Paul Chance, Learning and Behavior, Seventh Ed., p. 140-141
Getting inside a house is not a natural consequence of sitting and offering a human extended eye contact. Both of the above examples are contrived, even though one utilizes a phenomenon in nature, and the other a sound from a machine deliberately turned on by the human. There is no stipulation about the stimulus for these definitions, only the reinforcer.
A related example of natural negative reinforcement would be if my dog were in the back yard, it thundered, and she came in the doggie door under her own power. In this case, the reinforcer of getting in the house is a natural consequence of the dog going through the doggie door.
A Message from My Heart
Making glib claims that minimize the harm in negative reinforcement can result in dogs being hurt.
Please remember that when you make blanket claims about negative reinforcement, you are not necessarily talking about the more benign end of the spectrum or just one instance. If you have stature as a trainer, you are giving blanket permission to countless people to be cavalier about using aversives.
For whatever reason, most people are primed to believe it when told that X, Y, or Z method “doesn’t hurt” the dog. Many of us pet owners have had this experience. I would venture to say that most pro trainers have come across it in their clients. People are ready to believe that things that hurt dogs don’t hurt them. And they are ready to believe that practices that harm dogs are not harmful.
It is responsible to urge caution in the use of aversives. It is not responsible to minimize the fallout.
Regarding Comments
This is a post about speaking truthfully when making general claims about aversives. It is not about any training method. It does not “damn” anyone who uses negative reinforcement when training their animal. It urges them not to make blanket statements about the acceptableness of R- in general or to argue in favor of its acceptance as a general practice.
In 2014, I gave a webinar entitled, “Over Threshold: The Changing Definition,” for the Pet Professional Guild (PPG). Thank you to everybody who came! That was so cool to recognize some names!
I am outlining here some basics from the webinar. The reason I created the webinar in the first place was to allay confusion in the dog community about definitions and to aid in communication. I think some straightforward definitions and a “map” will help many people and their dogs.
Something is over the threshold of Rounder’s visual sense
I argue that there are three distinct, necessary, and useful definitions of the word “threshold” in dog training. They are:
Sensory threshold as defined in psychology. Here is one textbook definition: The faintest detectable stimulus, of any given type, is the absolute threshold for that type of stimulus.—Psychology, Peter Gray, 5th edition, 2007. In the webinar I go into the history, discuss types of thresholds in psychology (yes, even within the discipline there is more than one), and cite several more definitions and supporting information. There are many sensory thresholds: hearing, different aspects of vision, smell, touch, etc. One way the psych definition of absolute threshold is unfamiliar to our common use of “threshold” in training is that it is the stimulus that is above or below the threshold of the organism’s sensory capabilities. We say, “That sound is under the threshold of hearing,” not, “The dog is over its threshold of hearing.” Organisms’ responses to sensory stimuli are often invisible to the naked eye, and the responses generally do not correlate to overt, “dramatic” behavior.
Threshold in the common training sense. You can find definitions of this usage in books by Leslie McDevitt, Debbie Jacobs, Laura VanArendonk Baugh, and more. These definitions roughly agree. Here is Debbie Jacobs’, which is very straightforward: The threshold is the point at which your dog can no longer deal with a trigger before reacting in a negative way (with fear or aggression).” — A Guide to Living With & Training a Fearful Dog, 2011. I argue that this usage of the term generally corresponds to the response of the sympathetic nervous system: the fight/flight/freeze response or the fear response. In this usage, it’s the dog that is said to be over threshold, not the stimulus. I also cover usages of the term “threshold” by behaviorists including Steven Lindsay, Dr. James O’Heare, and Dr. Karen Overall. They discuss dozens of sensory, physiological, and behavioral thresholds, and always specify which one they mean.
Threshold of stimulus aversiveness. This threshold is the point at which an existing stimulus becomes aversive, generally because of the intensity of the exposure. I gave this threshold a name, but I didn’t invent the concept. It is necessary to locate this threshold through our dogs’ behavior when doing both desensitization/counterconditioning protocols and operant learning processes. Jean Donaldson defines it thusly: In DS/CC, “under threshold” is: An intensity of stimulus that elicits no fear (and so the intensity of stimulus that would not function as R-)…. Not “mild fear” or “manageable fear,” it’s NO fear.–The Pitfalls of Negative Reinforcement, PPG Webinar, 2012. Attention to this threshold is necessary also in negative reinforcement protocols that involve escape, since the stimulus exposure must be over the threshold of aversiveness in order for escape to function as negative reinforcement.
I believe that most of the problems in discussion of thresholds in the dog training community are due to two major points of confusion: 1) Whether we’re talking about the dog or the stimulus being over threshold (confusion between Definitions #1 and #2); and 2) The fact that many of us haven’t realized there is a difference between Definitions #2 and #3.
Mapping it Out
Here is a graphic showing these three different thresholds, and where the major types of protocols for working with fearful, aggressive, or reactive dogs fall among them:
This graphic may be shared for educational purposes with the copyright and credits included, and I would appreciate online citations to link back to this post or the webinar page.
The combination of desensitization/counterconditioning is performed under the threshold of aversiveness, as Jean Donaldson describes. Protocols using negative reinforcement (escape/retreat from the aversive stimulus) must take place partly over the threshold of aversiveness, or else the movement away is not reinforcing. They generally go back and forth over that threshold. Finally, flooding takes place at or above the threshold of fear response. In the webinar I split things out further and map six different protocols separately; I have roughly grouped them for this graphic.
I also talk about the ways that some of the thresholds move and change, through training, external events, and the emotional state of the dog. I describe how certain protocols become more difficult if two of the thresholds are very close. For instance, for wild animals the threshold of sensory perception of a stimulus and the threshold of aversiveness of that stimulus may be practically on top of each other. I address the often-heard claim that there is no negative reinforcement happening if the dog is “under threshold.” (Which threshold?) Also, I discuss which protocol is likely to take place closer to the stimulus—but also why the absolute distance is not a good point of comparison.
Conclusion
I will be working further on this topic, in part because of some great questions asked at the end of the webinar, so you may expect more from me about it at some point. The webinar is no longer available through PPG, but I plan to release a similar one in the future. I’m working on it in 2022.
Please note that the webinar will not have information on how to perform the above-mentioned training/conditioning protocols, on reading dog body language, or other training tips. It’s not about how to train a fearful or aggressive dog. But the feedback I have gotten from viewers is that it has clarified an area of considerable confusion and that it will help their training.
One of the classifications in Dr. Susan Friedman’s Humane Hierarchy that is pretty unfamiliar to most of us dog trainers is called “Antecedent Arrangements.” And look, it is on the more desirable end of the hierarchy! There’s no speed bump, caution sign, or stop sight. There’s an inviting little arrow. Worth looking into, don’t you think?
The Humane Hierarchy
We are accustomed to manipulating consequences when trying to effect behavioral change, but that’s not the only thing we can do. We can make changes to the antecedents, the things that set the stage for behaviors. Antecedent arrangement is on the desirable end of the Humane Hierarchy because it is less intrusive. You are not actually trying to change the animal’s behavior via reinforcement, punishment, or extinction. You are manipulating the environment to enhance the likelihood of the behavior you want.
How do you do this? The three types of antecedents are cues, setting events, and motivating operations.
Cues: You can remove something that serves as a discriminative stimulus for a behavior that you don’t want, or don’t want right then and there. Or you can add something that will better signal the behavior you do want.
Setting events: You can make the behavior you want easier by changes in the environment, and make the undesired behavior more difficult.
Motivating operations: You can do something that affects the animal’s motivation, either to perform the behavior you want more, and/or to do the behavior you don’t want less.
I have an example of antecedent arrangement in my second post about the Humane Hierarchy. But another one fell in my lap lately, so I thought I would share it.
The Dread Back Door
Undesirable door behavior
Since Clara became an adolescent, then a young adult, I have struggled with back door behavior with my three dogs. Actually, since before then, since Summer is reactive and sometimes can’t respond well when she’s worried about what might be down in the yard.
My goal has always been for Clara and Summer to lie down in assigned places close to the back door. Zani can sit or lie down wherever she wants, because she already has nice door manners, isn’t pushy, and had no agenda other then earning a treat if one is available. Summer needs to be back from the door to help her keep calm, and Clara is back from the door to keep her from bashing everybody else. Theoretically.
This is a generalization of a known behavior. I teach my dogs to get on mats and stay there as a stationing behavior, starting the day they come to me, in all sorts of situations. All around the house I use soft bath mats with rubber backing as dog stations, and they are like magnets to my dogs since they have been reinforced so highly for getting on them, lying down, and relaxing. But I was not able to use them to mark the places I had designated for Summer and Clara at the back door. This was because the den was the one room in the house in which Clara had free range as a youngster, and she would chew them up if not completely supervised. So I bought a couple of rubber non-skid bath inserts, like you put in the bottom of your tub or shower. They made decent station markers but were not attractive for her to chew.
I worked for a long time to get Clara to stay on her mat at the door. It was an “expensive” behavior for her, as Sue Ailsby calls it. There was just too much fun to be had dashing towards the door and knocking the other dogs aside like bowling pins. So it took a high level treat at first and some very consistent practice to get a nice wait on a mat. By the way, using going out of the door as the reinforcer didn’t work as an initial training strategy. Much too exciting. I needed to build the behavior up using high value treats. And since we went out the door many times a day, sometimes with very little preparation, Clara did get some chances to practice the undesirable things. I.e., I couldn’t always have great stuff and I had a hard time being consistent.
An old photo of Summer trying to make eye contact at the back door when her whole body and mind are already outside
Finally I did some intensive work over a couple of weeks and got some pretty consistent behavior. Once I got Clara’s behavior in shape, I started working on Summer. That was just as hard, in a different way, because I was working against some emotional patterning. Summer is anxious and predatory, and easily gets worked up into quite a state, anticipating what kind of animal might be in the back yard, especially at night.
So I finally got the general idea across to both of them (along with perfect little Zani), but the reliability of the behavior was not where I wanted it. My walking toward the back door was the main cue, but we were a long way from three dogs slamming into their places. I was still putting up with charging ahead from Clara every once in a while and glassy eyed standing around from Summer more often than that.
Then I had a bright idea. I got our door behavior very close to 100% without a struggle. The short video shows the solution. With one change, I got an improved cue and setting. Note that in this example, as in much of life, there is not just one learning process happening. The change in antecedent worked in tandem with the positive reinforcement (and differential reinforcement of an incompatible behavior) that had already been going on. But it sure gave it a huge boost!
What happens when you don’t have retrieve on stimulus control?
This post was updated and republished on January 31, 2019.
I’ve mentioned before that I’m not very good at stimulus control. I’ve included in this post a great video from when Clara was younger that demonstrates that really, really well.
Stimulus control in training is all about response to cues, and goes like this:
The behavior occurs immediately when the cue is given.
The behavior never occurs in the absence of the cue.
The behavior never occurs in response to some other cue.
No other behavior occurs in response to this cue.
Pride, the Rhodesian Ridgeback, sitting pretty on cue
This means, for example, if I have trained the behavior, “Sit pretty,”:
When I say, “Sit pretty,” the dog immediately sits up with his front feet in the air.
He doesn’t ever do that unless I cue it.
He doesn’t do it if I cue something else like down or stand.
He doesn’t down or stand when I say, “Sit pretty.”
Most everybody’s first question is about #2. If this were a natural dog behavior like lying down, he would still do it at other times, right? Sure. And although I’ve seen some discussions about that, I don’t know in what situations it would be a “violation” of stimulus control for the dog to lie down without a cue from a human. The common answer is to append “in a training session” to the above rules. But how do we expect a dog to draw a line between “training session” and “not a training session”? And aren’t we training for real life? Do we say that behaviors like sit and down are never on true stimulus control? Probably.
You may choose not to reinforce downs that you don’t cue, but they are reinforcing to a dog who wants to rest and relax. We can’t help that.
For most trainers, there is a period where we are teaching cue recognition and stimulus control where we do not reinforce uncued behaviors. After that is taught, though, we may change the rules a bit in real life.
There are behaviors for which one needs strict stimulus control. I have a friend with a service dog. “Gigi” has a special setup so she can do the equivalent of calling 911 if my friend falls down. Falling is actually the cue. My friend needs absolute stimulus control on this behavior because it is completely not cool if Gigi “offers” hitting the call box at any other time.
My dogs are not like Gigi. Or more to the point, I am not as skilled a trainer as my friend.
Lack of Stimulus Control
Even a gate doesn’t stop them from offering eye contact
If you put aside Rule #2 and reinforce your dogs for uncued behaviors, you get dogs who offer behaviors frequently.
One of the stereotypes of clicker trained dogs is that they offer behaviors all the time. Dogs trained with positive reinforcement tend to do stuff. And they’ll go wild with offering stuff if their people reinforce it. But it doesn’t have to be that way all the time. You can have a dog who is a virtuoso shaper and completely unafraid to offer behaviors, but who has also learned when that pays off and when it doesn’t.
We can set up some environmental cues and change our own behavior to let a dog know when we don’t want a bunch of offered behavior.
I do have those crazy behavior-offering dogs. If my dogs come running up to me in the yard for no reason to check in—I like that! They’ll usually get something from me. If I walk through a room and someone is lying nicely on a mat, they’ll get a treat.
I also reinforce offered eye contact. It usually comes along for the ride with other behaviors. Reinforcing this in real life means I have dogs who sit and stare at me.
I am OK with the results of this, but some people wouldn’t be. If you are regularly going to reinforce uncued behaviors, then you’d best be willing to do so even when it’s inconvenient. Because it’s just not fair to change the rules on your dog without warning. If you do that, you can put behaviors into extinction. This is unpleasant for the dog and doesn’t serve our overall training goals well.
My dogs are good at chilling since one of the offered behaviors I reinforce is lying down with relaxed muscles. This is nicely incompatible with trying a bunch of stuff to get my attention. I don’t mind tossing a treat around every 10 minutes while I’m working at the computer. But if we are really out of sync and they are tuning up to bug me to death, I just use management. I get behind a gate.
One of these days I may set up a cue for “The Bar is Closed.” There are a couple of situations in which I never reinforce my dogs and they have learned that perfectly.
In the following movie, the bar was definitely open. I was reinforcing Clara’s offered retrieves, and you can see the amusing outcome.
I’ve reinforced Clara for “trading” since she was tiny. But she started it. She always had a tendency to bring me things. I liked that, so I reinforced it. Still do. It means when she has something dangerous, I can immediately get it from her with no stress. This is a good thing since everything goes in her mouth. She was an outrageous chewer when younger, so I managed very tightly about this then.
When Cricket was alive, Clara was limited to only half the house most of the time. Clara was just under 2 years old when Cricket died in May 2013, and it seemed appropriate to open things up a bit after that. It went very well. About the worst thing that happened was that Clara snitched napkins off the table to chew up. I was careful where I put food, so she didn’t develop a counter-surfing habit. She did have certain items of my clothing—a hat in particular—that she kept a constant eye out for. But almost everything she picked up other than napkins she brought straight to me. She still does this, “busting” herself for picking up contraband.
There are good reasons to do the opposite, by the way. Some people teach a default “Leave It.” What if there is someone in your household who is prone to dropping pills or leaving sharp tools around? Then reinforcing a dog for picking random things up in her mouth and bringing them to you is not a good idea. But it has been a good choice for us, I think. You can see the rusty nail Clara brought me above. If she hadn’t, she would have been chewing on it in the yard.
By the way, the movie shows pretty impressive distance behavior. Clara was bringing items to me clear from the back of the house!
Does your dog have any behaviors on good stimulus control? Or any behaviors with an embarrassing lack of stimulus control, as mine do?
You could actually get my answer to this question by reading this other post: Only If The Behavior Decreases! But of course I’m writing some more anyway.
This issue was a big stumbling block for me when I first started studying operant learning. If negative reinforcement requires for an aversive to be removed, then it had to get there in the first place, right? That’s only logical. So whenever it appeared, there must have been punishment, right? I used to argue about this in my head all the time. But the answer is no, it doesn’t follow that there necessarily was punishment. There absolutely could have been, but it is not logically necessary after all. Here’s why.
Positive punishment: Something is added after a behavior, which results in the behavior happening less often.
Look at the second half of the definition. I’ve harped on this before, largely because I need a lot of reminding myself. The definition of punishment (and reinforcement also) is recursive. We can only know if punishment has occurred by traveling to the future. If the behavior didn’t decrease, there was no punishment.
Aversives, Negative Reinforcement, and Positive Punishment
Employing an aversive and using negative reinforcement do not mean one is also positively punishing a behavior. I have this from the words of Susan Friedman, PhD, in one of her lectures in her Living and Learning with Animals Professional Course. For punishment to have occurred, a behavior must decrease in frequency. That’s the definition. And if one is in a negative reinforcement scenario, the behavior the animal is performing at the onset of the aversive could very well be randomized, because that is not the focus of the training. The animal may not always be doing the same thing when the aversive (e.g. shock, pinch, pressure, nagging, or appearance of scary monster) starts.
Examples
Some typical implements and examples of negative reinforcement. Photo credits: Wikimedia Commons (cat o’ nine tails; man scratching), Alan Cleaver (alarm clock), Eileen Anderson (shock collar and stuffed dog with tight leash)
If I decided to teach my sensitive dog Zani to back up by consistently walking into her whenever we were both in a certain hallway (note: I don’t teach backing up that way), her behavior of coming into that hallway when I was there would definitely decrease. Coming into the hallway with me would have been positively punished.
But let’s say in another training scenario I play a mildly aversive noise whenever I want Zani to come get on her mat in the kitchen, and the noise stays on until she gets there. The mat is a “safe place” and getting on it turns off the noise. (Again, I would not do this.) Zani could be anywhere in the house when the noise starts. Her behavior at the onset is randomized, so nothing gets punished. But if I started doing it consistently when she was sitting in a particular chair, she would likely stop getting in that chair.
Since we humans fall into patterns so easily, it is very easy for positive punishment to start happening when we regularly use an aversive. But the point is that it does not have to happen.
Let’s face it, people use noxious stimuli all the time without behavior decreasing. That’s one of the many problems with trying to use positive punishment: unless the noxious stimulus is strong enough (and well timed enough, and several other criteria), the original behavior may maintain its strength.
Ramifications
So does this mean that negative reinforcement is OK? No. An aversive is an aversive. Just because there is no positive punishment going on doesn’t mean that the training is humane.
If you wanted to reword the question in the title, you could say, “But if you use negative reinforcement aren’t you also using an aversive, just like in positive punishment?” As long as it is recognized that the aversive is used in a different way, the answer is yes.
But the funny thing is, I’ve heard the relationship between the two processes that use aversives used to make two opposite claims, neither of them true in my opinion.
First is the claim implicit in the title. It usually goes like this:
If you use negative reinforcement, therefore you are using positive punishment, so neeter neeter neeter.
I have dealt with that one above.
The second claim goes like this:
Yes, you can have negative reinforcement without positive punishment. And that’s the GOOD kind of negative reinforcement. As long as there is not positive punishment going on, negative reinforcement can actually be kind of nice.
Excuse me? The existence or non-existence of concomitant positive punishment is irrelevant to how aversive a stimulus is. It is possible to train with a shock collar using a negative reinforcement protocol, for instance. Again, as long as the point at which the shock is turned on is fairly random with regard to the dog’s behavior, you may not necessarily see a decrease in a behavior as a result of the commencement of the shock.
Negative reinforcement is only one notch up from positive punishment on the Humane Hierarchy, and for good reason. It always involves an aversive, and employs escape and avoidance, pure and simple. So don’t anybody dare use my argument above showing that there is not necessarily positive punishment in order to say that negative reinforcement is OK.
I have an elliptical training machine in my back room. I’ve had it for three years. Zani loves that room because that’s where the kibble, some human food, and other interesting things are stored.
But when I get on the elliptical to exercise, she’s outta there. It doesn’t really have alligator jaws attached to it, but I think that’s a good portrayal of the way Zani used to see it.
For those who aren’t familiar with these exercise machines, here is a video of an elliptical in motion. It is similar to mine. You can see where a sensitive dog could be alarmed with the motion.
I have mats in that room for dogs to hang out on, and Summer and Clara stay on their mats and get the occasional treat while I exercise. Cricket did so too in her day. But not Zani, until now.
The other day I realized I could probably help Zani get over her fears. It took all of 5 days, and I did it while I was exercising.
Desensitization and Counterconditioning
The techniques of desensitization and counterconditioning (DS/CC) are often used together to help animals, including humans, recover from fears. They are not bandaid solutions that mask the symptoms. When done correctly, they change the animal’s emotional response.
Systematic desensitization is a procedure in which learned fear of a neutral stimulus is extinguished by exposing the animal to the stimulus so gradually that involuntary fear responses are never triggered. –Standard definition, as worded by Susan Friedman in her professional LLA course
This is the technique where you start with the thing the animal is scared of (the stimulus) at a distance or intensity where the thing is not scary. When the animal is OK with that, you gradually bring it closer or intensify it. Dr. Friedman points out that desensitization can only get the animal from scary to neutral. It doesn’t make the animal delighted or happy with the stimulus. But it can get the animal OK with it.
With counterconditioning, the animal’s respondent behavior to a stimulus is replaced with an opposite automatic response.–Standard definition, as worded by Susan Friedman in her professional LLA course
OK, counterconditioning is the frosting on the cake. Counterconditioning is the technique that can actually replace fear or another undesirable response with a positive emotional response. This is done by associating the scary stimulus with something wonderful, while the animal is under threshold, consistently over time.
Starting point: Typically when I would get on the elliptical, she would leave and go into the bedroom across the hall. She often went out of sight and got on the bed. That was her comfortable distance from the elliptical, so that’s where we started.
Picture 1: Since I was tossing treats to the other dogs anyway, I started tossing some into that bedroom (bank shot!). She learned to hang out by that doorway and get the treats. She could be out of sight of the elliptical if she chose. Distance: 16–19 feet.
Pictures 2a and 2b: Soon she started waiting in the hallway instead of in the doorway to the bedroom, so I started aiming the treats into the hall. I could tell she was comfortable because she didn’t retreat to the bedroom anymore or show any signs of concern, just happily chased down the treats. Distance: 12.5–16 feet.
Picture 3: All my dogs are trained to get on mats. The mats have good associations with relaxation and treats. So I threw a mat down in the area where Zani was already comfortable. She immediately got on it and stayed there happily when I got on the elliptical and started tossing treats. This was a big step because previously she had been on her feet and moving. If she had had any residual fear, she was free to trot away farther. Staying still in the presence of the elliptical was a big step. Hence, I didn’t cue her to get on the mat. I gave her the choice. I would have tossed treats either way. But she immediately plopped down on the mat and stayed there.
Technically this was a switch to the technique of Differential Reinforcement of an Incompatible Behavior, or DRI, from counterconditioning, since Zani was now also being reinforced for getting on her mat. However, as I mention below, operant and respondent behaviors can be all knotted up at times. Distance 12.5 feet.
Picture 4: Over two more days I moved the mat closer. I could tell that Zani was fine with that because her body language was comfortable, and she always got on the very front of the mat. Final distance: 10.5 feet.
That’s it! She can now participate in the “Eileen exercises and dogs lie on their mats and get treats” event. She is within my treat throwing range and she is completely comfortable. I don’t want any of the dogs closer than the current “front row” while I’m exercising since the moving parts of the elliptical could be dangerous for them. Since Zani generally likes a front-row seat, it will be interesting to see if she moves up to try to join or displace another dog. I’m betting she will. I may have to train her to stay back from the elliptical!
Here is a slide show of the steps we took.
I apologize for the poor photos. I wanted to show my view from the elliptical, and the actual distances involved. The light (and the clutter in the room) was not conducive to that. Pictures 1, 2a, 2b, and 3 are reconstructions by the way. I didn’t take photos during the process.
This slideshow requires JavaScript.
Slow Techniques?
To review: the desensitization part was moving Zani gradually closer to the doorway of the back room while the elliptical was in motion. The counterconditioning part was the yummy treats that accompanied the process. DRI came into play when Zani started settling on the mat. It worked because I didn’t rush. I watched Zani to be sure she wasn’t scared and just venturing forth to get the treats, then retreating to safety again. She had to feel safe with every step.
Desensitization and counterconditioning are often said to be slow. They can be. Depending on the history and intensity of the adverse reaction, the attractiveness of the counterconditioning item (usually really good food or fun play), correct timing, and the skill of the trainer, these techniques can take a while.
But guess what? Not always. I was frankly amazed at how fast Zani got over her mistrust of the elliptical, and felt bad that I hadn’t tried to help her in any systematic way before.
Even when these techniques take time, they are always my choice, along with operant learning using positive reinforcement**, for helping an animal overcome its fears. They are completely humane, and science has supported for decades that they create a true emotional response in an animal. I am privileged to watch my formerly feral dog Clara, under the care of a skilled trainer, blossoming into a comfortable, sociable dog, using these methods.
By the way, the treats I use for the elliptical mat game are pieces of Prairie kibble (which is small) and the undersized leftover pieces from when I cut up Natural Balance rolls. My dogs don’t typically need much encouragement to hang out on mats, and since I am throwing the treats while in motion, I don’t want to cause a scuffle if I were to toss something high value right between two dogs. But if I had planned better I would have used something higher value during Zani’s rehabilitation. The animal’s ultimate conditioned response can only be as positive as its response to that particular item, so one usually uses something really spectacular. Luckily it turned out I didn’t need to. But perhaps to cap things off I’ll surprise them all with a piece of liverwurst (hand-delivered) now and then.
** Operant learning played a role in this protocol as well. Once the good feelings associated with the unconditioned stimulus start spreading to the previously feared items, the animal will often on its own develop behaviors to hasten its access to the goodies. In this instance, Zani performed the operant behaviors of chasing treats and lying on her mat. Both of these are familiar, comfortable, and pleasant for her. Sometimes just having a job to do is a great help. On the other hand, the mat itself has already been classically conditioned as a very nice place to be.
