Respondent Learning for Dogs
A series of articles for professional dog trainers, those who want to become professional dog trainers, and those who want to become certified dog trainers.
For those of us involved in changing behavior, it is crucial to understand respondent and operant learning. These are the two models used to describe how a behavior is learned. Although they are manifested differently, they are still learned the same way – through experience! Let’s start with respondent learning and we’ll do operant learning in the next article.
Respondent learning is also referred to as Pavlovian or classical learning. You’ll also often hear it called “conditioning” instead of “learning.” Any of these terms are correct and mean the same thing.
Respondent learning is “the procedure of pairing a neutral stimulus (typically referred to as the CS) with a US.” (Chance, Learning & Behavior, 5th ed., pg 453.) “CS” is an acronym for conditioned stimulus and “US” is an acronym for unconditioned stimulus.
So, what does all that mean? Simply put, it means that an animal has created an association between two events: a song popular during your “coming of age” summer plays on the radio and your eyes fill with tears; the electric can opener runs and the cat begins salivating; you hear a police siren and your heart beat increases. So, one event predicts another event and elicits a response over which the animal has little or no control – thus the term “respondent.”
Respondent learning affects reflexes – pupil dilation, heart rate, perspiration, panting, etc. Again, we are responding to a stimulus and we don’t have control over the response. We generally think of respondent behaviors as having an emotional basis; we think about using respondent learning when we are dealing with emotional behaviors such as fear, anxiety and aggression.
The difference between respondent and operant learning is in the contingency. The contingency for respondent learning can be stated as: When “a” happens, “b” follows. The contingency for operant learning can be stated as: If I do “a,” then “b” will likely follow.
Other Factors Affecting Respondent Learning
Pairing the CS & the US
There have been a lot of studies done on the most efficient way to create that respondent association. There are four ways they can be paired:
- Trace: where the CS comes before the US
- Delayed: where the CS comes first, but the US overlaps it
- Simultaneous: where the CS and the US come at the same time
- Backward: where the US comes before the CS
It’s been discovered that simultaneous and backward pairing are quite ineffective. Trace and delayed are both effective; however, the effectiveness is dependent on a variety of variables.
For animal training purposes, we generally use trace pairing and the rule of thumb is to allow as little time as possible to elapse between the CS and the US. It is probably best to aim for no more than ½ second of elapsed time.
“Failure of a stimulus that is part of a compound stimulus to become a CS. The stimulus is said to be overshadowed by the stimulus that does become a CS.” (Chance, Learning & Behavior, 5th ed., pg 453.)
If there is more than one stimulus presented at the same time, the animal will make the association with the more salient stimulus. Probably the classic example is the owner who teaches his dog to sit thinking he’s teaching a hand signal, but in reality he’s leaning over from the waist down and the dog recognizes the lean, and not the hand signal. The owner has overshadowed the hand signal with a more salient signal – the movement of his entire torso.
“Failure of a stimulus to become a CS when it is part of a compound stimulus that includes an effective CS. The effective CS is said to block the formation of a new CS.” (Chance, Learning & Behavior, 5th ed., pg 448.)
If an animal has already learned an association between two events, and the stimulus is then paired with another stimulus that the animal has not learned, the new stimulus is unlikely to become meaningful to the animal because the conditioned stimulus he already knows is the predicting stimulus. So, if you’ve conditioned your dog to the clicker and the clicker predicts food, if you rang a bell at the same time that you clicked the clicker, your dog would be unlikely to associate the bell with the food, because the clicker already predicts the food and he has no reason to learn the bell.
Prior Experience with CS & US
Finally, there’s the matter of prior experience with the CS. If an animal has had a history with a CS, but the CS did not predict anything, then it will take longer to create an association between the CS and a US. For example, if you ran around the house clicking your clicker, but never paired it with food, your dog would learn that the clicker doesn’t predict anything meaningful to him. So then, when you decide to pair the clicker (CS) with food (US), it’s going to take longer for your dog to create that association than if he had never heard the clicker before.
There are many more respondent learning concepts, but these are the concepts that will most often affect your training. We’ll deal with other concepts in later articles.
In real life, as with so many things, operant and respondent learning are very difficult to separate – as Bob Bailey says, “Pavlov is always sitting on your shoulder.” What he means by this is, even though you are using operant methods to train a behavior, you are really using both operant and respondent methods, and vice versa.
When you teach a dog that when you say “sit” and he puts his rear on the floor he gets a treat you are teaching an operant behavior – the contingency is “if I put my rear on the floor, I’m likely to get a treat.” However, you are also creating an association between the word “sit” and treats. So, whenever you use operant learning, you’re getting “respondent bang for your buck.” (I don’t know where I first heard this phrase, but I’ve been using it for quite some time. If I didn’t come up with it on my own, I suspect it came from Jean Donaldson, as it sounds like something she would say.)
Let’s look at a sample case and see how we might use respondent learning to change a dog’s behavior.
Scenario – Dog Growls at Big Hats
Let’s say you have a dog that growls at people wearing big hats. Although we’re making an assumption, this is a common enough occurrence that it’s probably safe to say that big hats frighten the dog. Although growling is an operant behavior, it is usually an emotionally-based behavior. Because we can’t readily measure the dog’s reflexes, we use the growling to assess his level of comfort.
We know the dog is scared of hats, and we know that this is an emotional reaction; therefore, we know that this behavior is a good candidate for respondent learning. The thing to remember when starting a respondent learning program is the contingency – “a” predicts “b.” Therefore, when the dog sees a big hat, good things happen. In this case, good things are going to be yummy treats.
This is a pretty easy set-up for training. We get a helper who will wear a big hat and appear on cue. The trainer and the dog can remain in one location and the helper will come in and out of the picture. Every time the hat appears, the dog gets treats.
Because we are using the respondent contingency, the dog’s behavior is irrelevant. He can growl, bark, roll over, or anything else – it doesn’t matter! What matters is that when big hats appear, the dog gets yummy treats.
Over time, as the dog learns that big hats are followed yummy treats, he’s going to change the way he thinks about big hats. Big hats are no longer scary – they now are a predictor of good things and he is happy when he sees a big hat!
Counterconditioning & Systematic Desensitization
Counterconditioning is “the use of Pavlovian conditioning to reverse the unwanted effects of prior conditioning.” (Chance, Learning & Behavior, 5th ed., pg 449.)
Systematic desensitization is “combining relaxation with a hierarchy, of fear-producing stimuli, arranged from the least to the most frightening.” (University of Southern Florida Behavior Analysis Glossary http://www.coedu.usf.edu/abaglossary/glossarymain.asp?AID=5&ID=3101)
What we’ve described in our case study is also called “counterconditioning.” We are re-conditioning (or countering) the dog’s emotional association with hats.
Counterconditioning is the broad category for changing prior conditioning. Within counterconditioning are other, more specific techniques. The one we will use most often is “systematic desensitization.”
Although systematic desensitization is primarily used with humans and requires teaching humans relaxation techniques, the dog training world has co-opted the term and as much of the technique as possible for use in counterconditioning emotionally-based behaviors. Although we haven’t yet figured out how to teach animals relaxation techniques, when using systematic desensitization, we do our best to keep the dog sub-threshold. This means that he is kept far enough away from the unpleasant stimulus to be comfortable and relaxed.
If you go back and read the above scenario, you’ll see that there’s no discussion about the dog’s proximity to the hat. In counterconditioning, that is not necessarily a consideration. However, you will usually get faster results if you incorporate systematic desensitization into your program. Following are three good reasons for using systematic desensitization.
- As a rule, using systematic desensitization will speed up the learning process. Because the animal is sub-threshold, he is able to pay more attention to surrounding events and stimuli and thus make a quicker association.
- With some behaviors such as resource guarding and handling, it is safer to use a hierarchy. By starting with an object that the dog does not guard or moving your hand toward him but not touching him, you “teach him the game” before moving to more difficult objectives. If the hierarchy is arranged properly, the dog should always be sub-threshold, and therefore comfortable with your proximity.
- Finally, using systematic desensitization allows the trainer to measure the progress more easily. By beginning at a comfort level, you are able to define the dog’s baseline behavior. This is the behavior you will be looking for throughout the program that indicates you can move to the next level in the hierarchy. If you start with the dog over-threshold, it may be quite some time before you are able to see any progress. Although this does not mean that learning is not taking place, it can be discouraging for the trainer.
Be aware that systematic desensitization simply makes the process faster and easier; it is not absolutely necessary for counterconditioning. The scenario that comes to my mind is large, leash-reactive dogs living in Manhattan. They have to go out onto the street, and it is very difficult to manage that environment – there are dogs everywhere! In a situation such as this, it may not always be possible to use systematic desensitization, so simple counter-conditioning may be the answer. It may take longer, but it can still get the job done.
Although this article did not cover all respondent learning concepts, these are the concepts most useful when changing emotionally-based behaviors. Other respondent concepts will be covered in future articles. In the next article we’ll discuss operant learning.
Raising Canine d has a school for dog trainers which focuses on operant training for dogs, dog behavior, working with clients and addressing client compliance, and the science behind behavior modification.