Originally published in June 2013
I had lunch with a colleague the other day and we got to talking about attending to behaviours, applied behaviour analysis and reinforcing with attention. She works with children and I work with dogs. How often do we hear parents, teachers and other adults talking about children misbehaving because it gets them attention? We hear the same thing in the field of dog training. “He is just doing it to get your attention” is a statement I often hear from my clients when their dog is doing something fairly benign that will result in a mild unpleasant consequence.
|Learners never work to get an unpleasant consequence, but if the unpleasant consequence is not a bad as being uncertain then the known unpleasant consequence may increase the likelihood of a bad behaviour happening. Image credit: Cole123RF / 123RF Stock Photo|
If you misconstrue the attention issue, you can completely miss what is happening when you are training. To begin with, it is important to understand that no learner will work for the opportunity to have an unpleasant consequence. It is also important to understand that just like reinforcers can be arranged on a hierarchy, so can punishers. For D’fer, throwing a bumper into the water is better than throwing a Frisbee on dry land which is better than a piece of liver which is better than kibble. There is a hierarchy of what is important to him. Likewise, getting a dirty look is less annoying than being yelled at which is less annoying than losing a turn and being put in your crate which is less annoying than being shocked with a shock collar. Perhaps the most annoying punishment for many learners, both children and dogs, is uncertainty. In the hierarchy of punishers, uncertainty can be as unpleasant or more unpleasant than electric shock.
|Reinforcers have a hierarchy; out of these five things, each dog will have a most favourite, a next most favourite, a third most favourite, a fourth most favourite and a least favourite option. Punishers or aversive stimulus work the same way. If the choice is between going to your room or being uncertain about the outcome, many children will choose going to their room, as evidenced by their behaviour. Dogs faced with a squirt of water in the face or uncertainty will also often choose the least unpleasant option, and choose to have water sprayed in their faces by doing the behaivour that will result on the known outcome. Image credit: kentoh / 123RF Stock Photo|
Behaviour analysis is a tool that helps us to understand how we change behaviour, and it works when applied to any learner, human or non human. Consider the scenario of a young child who is told “If you spit at your sister, I will send you to your room”. This leaves a huge window of uncertainty. The question in the child’s mind is “If I don’t spit at my sister, what will happen?” In an anxious child, the uncertainty builds and builds and builds and builds until being sent to one’s room is a relief. When there is an unpleasant consequence to a behaviour, the learner needs to know what the consequence is for the other or desired behaviour. If instead of framing the contingencies for a child in terms of “if you do X, this bad thing, Y, will happen” you frame the contingency in terms of “if you don’t do X, this good thing, Z, will happen” you allow the learner to make a better choice. Now you can say to the child “if you don’t spit on your sister, we will have ice cream.” For children who are deeply affected by uncertainty, the second statement can be very helpful.
What about dogs? Dogs are not able to understand complex language (if/then statements for instance) so we cannot outline outcomes and contingencies. Dogs learn by experimentation and experience. How often a client has come in and said “but he knows that I will spray water at him if he jumps on the counters!” In this case you have several forces at play on the behaviour. Jumping on the counters sometimes results in a treat; even the chance to smell the roast you put there the night before can be a potent reinforcement. The spray bottle is contingent on your timing and ability to guard the counter, so the dog is gambling that you won’t be able to get the spray bottle in time to prevent him from jumping on the counter. Not only that, but let’s say that for this dog, being sprayed with water is annoying, but not VERY annoying. And there is no clear picture of what will happen if he stays away from the counter. So much uncertainty and the dog has a dilemma; jumping on the counter may alleviate the uncertainty of what will happen if he doesn’t jump on the counter. In this case, simply marking when the dog is close to the counter but not jumping up and then treating him away from the counter, you remove the uncertainty.
Extinction, the process of changing behaviour by waiting till the behaviour ends is a very sound practice when done well, but can completely backfire if you don’t understand the mechanics, or if you set up contingencies where there are uncertainties. When a dog is howling in his crate for instance, and you want to wait him out, the fact is that you can only wait so long before you must let him out to toilet him. If the first time you set up the dog to wait him out, you wait an hour, but you have to leave the house and you need to toilet the dog before you leave, so you let him out, then what you have done in effect is reinforced one hour’s worth of howling. If the next time you try, you wait 90 minutes, and then your roommate lets him out because she just cannot stand it any longer, then you have taught the dog to try even harder, and that if he doesn’t try for long enough the door won’t open. The next time you try you may be able to wait for two hours and in effect, you teach the dog to howl for longer instead of getting rid of the howling altogether. The dog is left in an uncertain state of mind; what will happen if he barks? He might get out. What will happen if he doesn’t bark? No one has explained that. In order for extinction to work, you have to set up two contingencies; what will happen if the target behaviour happens, and what will happen if the target behaviour doesn’t happen. If there is only one contingency, then there is uncertainty about the other contingency, and if there is uncertainty and then the first contingency doesn’t play out, then you have created a situation where the behaviour won’t change and the dog will remain uncertain.
|With a dog who has practiced barking for a very long time, you may need to look for a loophole in the behaviour in order to be able to change it. Waiting this little fellow out may take longer than you can do. Image credit: reddogs / 123RF Stock Photo|
With behaviours such as the dog howling in the crate, few people have the where withal to sit out a dog who will howl for hours on end, and reward the lack of howling. Luckily there is a loophole in this particular behaviour, and we can use that to our advantage. In fact, most behaviours have loopholes if we look carefully. The loophole that is relevant to crate barking is that a dog cannot bark without inhaling at some point. When the dog inhales, we have a little tiny hole of silence, and we can mark that and reinforce. There is another loophole too; if we reinforce barking very early in the sequence, we can teach the dog that it doesn’t take much barking to get the treat. By interrupting the barking early and often with a reinforcer, we actually weaken the behaviour. In effect, what we do is remove the uncertainty of the situation for the dog. When people try and use extinction, and then they don’t follow all the way through, the uncertainty problem arises and then the dog doesn’t know where he sits and the behaviour gets worse instead of better.
When you are working with learners who have undesired behaviours, looking for the loopholes can really help you out. The child who spits at his sister has to turn his head in order to aim. If you mark that head turn by calling the child’s name and reward him for controlling the spit, then you are reinforcing something other than spitting, even though it was the behaviour that happens just before spitting happens. Do that often enough and the uncertainty of what happens if he doesn’t spit goes away. Over time, the need to mark and reinforce the desired behaviour also goes away. You don’t actually need to reward good behaviour forever; it eventually falls into a bigger context of self reinforcing cycles where not spitting makes the learner’s sister more pleasant to him. Looking for loopholes allows you to choose an alternative that doesn’t leave the learner in an uncertain state.
When we have behaviours happening that we don’t like, then we need to think about a number of things when we try and get ahead of those behaviours. The first is to define what behaviour we do want. Saying that we want the child to refrain from spitting or hitting leaves the door wide open for a child who is creative to find other ways of being socially unacceptable. Maybe that child will try kicking, or biting instead. Looking for the loophole like head turning, gives us a definable behaviour we can mark and reinforce. The same is true of the dog who was want to refrain from jumping on guests or climbing on counters. Defining what we don’t want can be a helpful starting point, but it won’t really move us forward in terms of developing a good training plan. Stating the problem in terms of what we DO want is a really important starting point.
The next point in making a training plan to change behaviour is that if the behaviour must be stopped immediately, and you are going to choose to use an aversive, the aversive must be strong enough to work in the first three tries or you are not actually making a change. In human learners, telling the learner the possible outcome without telling them what the outcome might be if they choose to do something other than the undesired target behaviour sets up uncertainty and uncertainty may be more punishing than the consequence that is offered. If you are inconsistent about your contingencies in your canine learners the same is true. The canine learner determines that some of the time a bad thing happens and some of the time nothing happens, but has no information about what will happen when they make a different choice.
Finally if you are going to actually use extinction, you have to go all the way or not bother at all. If you have taught your dog that barking opens doors, then you are likely going to get more barking before you get less. If you change the rule the dog will need to learn this by experience and the first thing he is going to try is not going to be silence. Let’s say that you have a determined barker. If they bark for forty five minutes and then you get frustrated and open the door, you have just set the bar for the next time you try extinction. In really determined cases, you might be waiting for hours. If this is the case, it is easy to argue that it is more humane to either reinforce shorter spells of barking, or use a very significant aversive with a solid plan for when the dog isn’t barking so that he can identify the desired behaviour.
Uncertainty is perhaps the most aversive thing I see used in training, regardless of the species, and I see it in the world of positive reinforcement from time to time. I have seen it in dogs who are being offered as many choices as they could possibly think up to get a click and treat. Without parameters, some dogs stop thinking about the next step and start to throw behaviours out like confetti. This can be especially true of dogs who have been inconsistently trained in the past. I am working with a dog right now who has been through a number of different training systems; we know he worked with a balanced trainer at one point, and he has been clicker trained for about a year now. When we are shaping he is sometimes really anxious. Recently, we use a mild aversive when he got too aroused and started throwing himself around and grabbing his leash. He had a sudden lightbulb moment; don’t do THAT! When he understood both what was going to get clicked and what was going to result in the aversive stimulus, he was able to settle down and think. If a learner understands the outcome for both the target behaviour and the alternative, then he can make a choice. But if one of his choices is uncertain, then he may choose the behaviour you don’t want in order to control what happens to him, because certainty is more reinforcing than uncertainty.