Pairing: The Silent Saboteur of Reactive Dog Training?
Many dog trainers who work with reactive dogs work with pairing up good stuff with the bad stuff. The aim is to help dogs overcome their feelings about stuff that they’re worried about. You’ll hear this called counterconditioning.
When I first started training, I also used to use this method. Working in a large shelter setting, however, often made it very hard. When you have 160 dogs, a cattery, people, cars, wheelbarrows, staff members, people coming to adopt, volunteers and a colony of outdoor cats coming and going, as well as all the wildlife in the forest next door, clean and clear pairings are not easy to achieve.
It was also achingly slow. I just wasn’t seeing any progress. In many ways, this problem with pairing has driven my journey with Lighten Up.
I also saw a lot of dog trainers in the same position. Working with a dog for eight months (!) and seeing g-l-a-c-i-a-l progress… that just didn’t feel right to me. Firstly, there were technical errors. Pairing can be really tough. Then there were also the limitations of this method.
Now, I got good at the technical stuff. I can create a pairing in five individual trials in three or four sessions over a week or so. Working in a huge and busy shelter will help you get there faster than most.
I also share these technical skills widely in my work. It’s important we can do those trials quickly.
Our dogs deserve trainers who have that technical prowess.
We shouldn’t still be working on that eight months later. That to me suggested huge failings in training methods.
It also sets up the potential for huge welfare issues.
What are the welfare issues of poor training?
Your dog hates dogs, so you’re going to show him dogs every other day for eight whole months? 241 days, so 120 training sessions, give or take? 5 dogs (or the same dog 5 times!) in each session? So 600 dogs over 34 weeks – and the dog is still struggling? 34 whole weeks later?
My training sessions take around 5 minutes – that’s the quickest I can do them – but they get longer by the end of the training programme. Say they didn’t, that’s at least 10 hours of being shown dogs and the dog is *still* barking at other dogs?
I simply cannot live with that!
We get better by first understanding the technical side. When we understand that, we can reduce ethical and welfare concerns considerably.
That means understanding the theory though. Don’t worry. It’s pretty simple. In fact, it’s one of the simplest forms of learning there is.
The theory is purely Pavlov. You know, that mean guy with the bells and the dogs and the meat paste. Let’s forget that he caused huge suffering for the dogs with whom he worked in the name of science for a minute. Let’s also forget that he didn’t use a bell.
We have a theory.
Pavlov was working on reflex responses.
The reflex is simply a relationship between one thing and a bodily response. Those things that cause a response are known as stimuli. I get very tired of this word, so I often say ‘stuff’. Some of that stuff is intrinsic. For instance, when our bodies detect low levels of oxygen or high levels of carbon dioxide in the blood, it causes a response. It alters our body’s behaviour.
Reflexes are often protective. What this means is that they have evolved to protect our bodies and ensure optimal functioning. Think of the ‘dive reflex’ among diving mammals for instance. This reflex prepares warm-blooded mammals to prepare for sudden submerging in cold water. Think about times you dived into cold water: you held your breath. Your blood vessels in your limbs constricted, and your heart rate and breathing slowed. Despite that, the body keeps up arterial pressure, meaning that your organs don’t die of oxygen starvation as the body prioritises oxygen to the brain and heart.
Reflex responses are usually involuntary. We can’t stop them from happening.
When do reflexes fail?
Reflex failure is very interesting. Let’s take your eye. Normally, your iris sphincter muscle contracts in bright light. This constricts the pupil and makes sure the bright light really doesn’t damage your retina.
A reflex is made up of three components known as the reflex arc. The first is sensory. The body senses something and the nerve cells that do the sensing carry that information to the central nervous system.
There, it passes through an interneuron that passes the message on to a motor neuron. The sensory neuron passes along information to say, ‘Hey, this is happening!’ and the interneuron and the motor neuron say, ‘Oh, okay… We’ll do this then!’
When a reflex fails, say for instance your pupil does not constrict when a bright light is shone at it, then this gives medical professionals useful information.
Perhaps the problem is on the sensory neuron side. Is it sensing the stuff? If your eye is covered in a patch, maybe not!
Or the problem could be in the transmission of the sensory message to the interneuron. That’s important information for professionals too. Maybe the interneuron isn’t getting the message, or isn’t passing it on. That suggests a neurological problem somewhere.
The problem could also be in the message going out to the muscles. Maybe that’s malfunctioning. Thus, if a paramedic shines a light in your eye after a headbump, it’s a useful way to detect if there’s any trauma, but it might not say what that trauma affected.
Reflex failure with dogs
One reflex vets might test is the knuckling reflex. Normally, if you tuck your dog’s foot under when they’re standing, they’ll right it straight away. However, if they don’t, or if they’re very slow, then it could indicate problems that the vet can then investigate.
You can see that here in this vet test:
This in itself isn’t massively informative. That problem could be sensory. Maybe the sensory neurons aren’t passing on the message to the brain. If so, that could be the receptors, or it could be the signalling that reaches out to the brain. It could also be the interneurons that receive, process and pass on the message. Reflex failure, like the ‘righting reflex’ above could also be a motor problem. Perhaps that’s on the signalling bit and perhaps that’s on the actual muscle itself.
If you want to know more about the fascinations of the reflex arc, by the way, I highly recommend Guy Leschziner’s book, The Man Who Tasted Words. Once you’ve read that, read Ed Yong’s book An Immense World. Both explore the physiology of sensory input.
Back to Pavlov
Pavlov was investigating the digestive system, a complex process of different reflex actions. It really annoyed him that the dogs he was using as test subjects kept drooling, even when there was no food. What was a nuisance variable in his experiments became the thing he is perhaps best known for.
We call this process conditioning, but whether you call it ‘classical’ conditioning, ‘respondent’ conditioning or ‘Pavlovian conditioning’ is largely your own education and preference. I call it ‘respondent’ to remind me that it is a response. That helps me remember that it is (largely) involuntary. The world says Jump and our bodies say How High?
I usually also swap out ‘conditioning’ with ‘learning’ every time I read it!
Our own physiology, it turns out, can learn. They learn by association.
What he found was that we are quick to work out pairings that predict the imminent arrival of things. If a bell predicts the imminent arrival of food, then our body will start responding as soon as it hears the bell.
Later researchers theorised that it wasn’t just physiological reactions but also emotional responses as well. Signs in the world prepare our body to respond. That works on a physiological level and with basic emotions.
Our dogs learn more than just drooling for treats. A harness comes to predict a walk, so dogs get excited when we pick up their harness. The word ‘walkies’ predicts the same thing, so we end up spelling it out.
Arriving at the vets can predict pain or restraint, so dogs might start showing signs of fear the moment you arrive at the vet. I’ve worked with a couple of dogs who’d been ‘trained’ with shock collars which beeped, and came to be as fearful around beeps as they were after they’d been shocked.
Believe it or not, the scientific jury is well and truly out on emotions as to whether they are involuntary reflexes, like a knee-jerk reaction, or whether they are constructed in more complex parts of the brain.
Or both, of course. I’m in the ‘both’ camp myself.
There’s even debate about what emotions are for.
There’s definitely debate about which emotions animals share with us, and what the emotional experience of an animal would be like.
Most researchers agree that there are basic, physiological processes that occur. They disagree about how many there are and how complex they are. Even researchers who think emotions are mainly complex and constructed believe that various things that happen in the world usually fall into ‘positive’ or ‘negative’ basic emotions. We see a snake and we’re more likely to feel fear.
Professor Dan Siegel uses the terms ‘upstairs brain’ and ‘downstairs brain’ to distinguish between the subcortical ‘downstairs’ structures that we share with other mammals, and the cortical ‘upstairs’ structures which tend to be more complex for humans. Professor Walter Mischel calls these ‘hot’ responses and ‘cool’ responses, which I also find helpful.
Most researchers would generally agree that there are ‘downstairs’ or ‘hot’ brain emotions. We share these with other animals and they are very much about survival. They can also help animals adapt better in social situations. These tend to be ones that are more instinctive – the kind of BIG emotions our reactive dogs have. We share these ‘downstairs’ responses.
Other emotions are more complex and involve more cognitive processes. They are learned. Think about anticipation and disappointment, for example. Right now, I am anticipating my lunchtime sandwich with delight. I’m also disappointed that a contractor has not returned a call. Both of these are more complex emotions that are individualised.
Pavlovian problems with our dogs
Understanding that emotions can be simple ‘hot’ responses or more complex cognitive ‘cool’ processes is our first step. Don’t forget, it can of course be both!
This is very relevant for our reactive dogs.
There can be ‘hot’ Downstairs Brain reflex responses which tend to be very big, largely involuntary and very invariable, and there are also more complex elements that involve consequences. These are the ‘I’m afraid for my life’ kind of feelings. They’re also the ‘Cats are the Mortal Enemies of Dogs’ kind of feelings.
But I also believe that dogs have ‘cool’ Upstairs Brain (or, half-way up the stairs kind of stuff) responses. These are more influenced by reward and consequences. What happens next kind of stuff. By the way, just so I’m not seducing you with science, there are scientists out there who are sceptical that dogs have these ‘cool’ Upstairs Brain kind of feelings.
There are groups of researchers who are looking into this kind of thing, but there’s a lot of dispute which is pretty healthy, I guess.
Even so, BIG behaviours and responses may have ‘cooler’ elements. For instance, the reactive dog who has learned that if they bark and lunge, you will let them off the lead to run over to say ‘hi’ to other dogs. This kind of response is much more complex. It involves learning, history and expectations. Frustration and anticipation as well as disappointment and surprise are four ‘cooler’ emotions that I’m pretty sure most of us with dogs would consider that we see (see Pickersgill et al., 2023 for more).
It’s difficult then for trainers and behaviour consultants to always know whether they are working with complex emotions and behaviours that have consequences, or simple emotions and behaviours that are a reaction to the world.
Problems with excitement
First are the problems with excitement. I’m currently working with a dog who starts fixating on the distance as soon as we get near a road. He’s looking for cars, because when cars arrive, it sends him into a frenzy of chasing delight which is very tough for his guardians to control.
Here, you can see some mild interest as my phone alarm went off to remind me to give Heston his pills. As you can see from the video, both dogs understand that the alarm means something, and they know where to go when they hear it. The phone alarm (X) predicts cheese (Y).
As time passed, my dogs also learned to predict the time that the alarm would go off. They were actively waiting for the moment the alarm would go off. Once or twice, they would even sit waiting, looking at me for it to ring.
This kind of response is more complex. It involves expectations and learning history.
Similarly, if our dogs learn that chasing pigeons is really fun, they can be constantly scanning the world for pigeons. Instincts and reflexes and ‘hot’ Downstairs-Brain kind of stuff provides the ingredients, and life turns them into a fine behaviour cake.
Problems with fear and unpleasant situations
Then there are the problems with fear or unpleasant situations.
For example, our dogs can learn that the approach of other dogs causes feelings they don’t like. This is perfectly normal in lots of ways. As soon as they see a dog, they then start barking.
You pull up at the vet and the dog goes into a panic. Nothing has even happened – yet!
You call your dog back as it’s the end of the walk, and they’re reluctant to come back because they know – because of Pavlov! – that this time, you intend to clip that lead back on and put an end to both their off-lead running about as well as the walk itself.
You’ve got a dog who doesn’t like getting in cars and the moment you open the door, they’re trembling.
You’ve got a dog who struggles when you go out, and the moment you pick up your keys, your dog goes into a full-on panic.
There are many situations when our dog learns that X predicts Y, especially where bad stuff is concerned. The brain is wired to focus on us being able to predict bad stuff will happen. It’s a very useful thing to happen. Fear, then, is great at motivating us to run away or seek cover.
Because our dogs are very good at learning that X predicts Y, we can use this in our training. It’s not all about problems! Lidy is so smart she knows that if I take my phone in the kitchen, we’re going out in the car. How mad is that?! I didn’t even realise this predictable pairing I was making. Pairings can be useful as much as they can be a nuisance.
Predictability like this can be a nightmare if you have an anxious or frustrated dog. Expectations – good or bad – can leave us having to do a lot of work to disassociate the pair.
On the other hand, it can be really useful if our dogs are struggling, especially if they’re anxious.
We can use it in our training.
In fact, this is how a lot of reactive dog training has worked: we’re trying to get X (the appearance of a scary dog on the horizon) to predict that Y (food will appear).
It’s more than just about predictions, though. We’re also trying to cause a different emotional state for the dog. We call these states ‘incompatible’. You can’t feel fear and feel relaxed at the same time, for example. Or you can’t feel afraid and also feel excited to get food. In theory anyway!
Incompatible physiological and emotional responses
This is where many dog trainers are working. One well-known procedure is known as ‘Bar is open/Bar is closed’ and it comes from trainer Jean Donaldson. When the bar opens, food rains from the sky. When it closes, the food stops. This is the very best and clearest example of respondent counterconditioning.
Trainers are aiming to pair up the bar opening with the appearance of the dog. When a scary dog appears, the bar will open and treats will rain from the sky in abundance. When the dog disappears, the bar will close.
In experimental settings like Pavlov had in his lab, animals often learn that X predicts Y really quickly.
A light predicts food will be available. A tone predicts a footshock will happen.
The light not only causes the animals to run to the place food is delivered, but the body starts preparing to eat by salivating even before food arrives. The tone causes the animal to start trembling before the footshock even happens.
So where does it go wrong?
Firstly, we’re not working in a lab. Trying to make these pairings can be really confusing to the dog.
If it’s not really, really super-abundantly clear that X (the arrival of the dog) predicts Y (the bar opening), then they’re not learning anything. That’s one reason dogs are still struggling eight months in. We need a clean and clear environment so the dog can’t fail to make the connection. I’m a huge fan of very carefully engineered staging for the first pairings for this reason. Get that right and everything else is easy.
Pairing things up needs to be strong from the start. If the pairing isn’t clear to the dog, then they’ll struggle to understand what is happening. In reality, the scary stuff needs to be at low intensity for the learner. If it is not, we may not see sharp learning as the dog is too focused on other things. Again, another reason that how we set up the learning needs to be really carefully done.
Thirdly, we don’t switch from ‘Bar is Open’ quickly enough into other behaviours that we can shape. These days, when I train, the bar is open for 3 – 5 trials maximum. A trial is simply the brief presentation of the scary stuff followed by the good stuff.
So for a dog who is barking at people, that would be the brief appearance of a person followed by food raining from the sky.
That’s one trial.
Shortly afterwards, like a minute or so later, another person will appear. The bar will open again. Food will rain from the sky.
That’s two trials. Our dogs should pick up on what we’re doing pretty quickly if we’re working at low intensity and with high enough value treats. Then we can ask for different behaviour.
But there’s more…
We also have to be really clear WHEN the bar opens. It can’t open before the dogs or people or cars appear. It has to open after our dog has clocked them.
If we open the bar before the stuff that triggers our dog’s behaviour, then we’ve got the order the wrong way around. The bar opening will predict dogs arriving. In the same way as tuna came to predict pills to my cat, we could end up causing our dogs problems.
And we have to be careful with timing. It has to be really quick between our dog clocking their trigger and the bar opening. I don’t mean minutes. I don’t even mean seconds. Milliseconds.
We also can’t put other stuff before the food and before the dogs. So for instance, if we go stand in front of our dogs before the triggers arrive so we can cause the bar to open easily, then that becomes the predictor of food arriving, not the dogs. That’s just plain confusing for our dogs.
So technically, it can be tricky for us to get this pairing malarky right unless we’ve got some serious skills in creating and using staged set-ups to make it clear for our dogs.
The biggest problem
Many dog trainers use what they say is ‘counterconditioning’ and it’s actually not.
The term in itself is a confusing concept.
Partly because there’s two different types of counterconditioning: one related to incompatible bodily and emotional responses as we’re hoping for, and one related to incompatible behavioural responses. It’s also confusing because it’s often mixed up with “rewards” when the pairing isn’t a reward at all. The surest sign I have that a trainer has misunderstood the concept is that they’re trying to use a clicker or that they call food a ‘reward’. To be clear, we’re not rewarding anything here.
It’s also confusing because some dog trainers try to use it for a process of helping dogs feel relaxed around their triggers. Relaxation isn’t the only feeling that’s incompatible with fear and it might be really hard to achieve. Let’s face it: if you’re a tiny bichon and you’re faced with an enormous Great Dane, relaxing is perhaps not easy to do.
Sometimes trainers also erroneously use it to help dogs get used to stuff they were sensitive about.
But a lot of trainers think they’re changing the dog’s feelings about stuff and actually, they’re just making a pairing for the dog. They’re conditioning, not counterconditioning. This is a big issue. Unless your dog spots every single spotable dog, and reacts to every single spotable dog, then you probably don’t need counterconditioning.
We accidentally end up making triggers even more noticeable for the dog.
Why is this?
We have to be really, really sure that it’s a reflexive emotion… that it’s a hot, Downstairs Brain kind of thing. If there are clear consequences to our dog’s barking and lunging – even occasionally – like the people go away or the dog backs off, then we will need to use other methods.
How then do we know it’s an involuntary emotion? A knee-jerk response?
Because it’ll happen every single time our dog sees their trigger.
Now their trigger might be very specific. It might be young male dogs who whiff of testosterone. That might account for why it seems to be intermittent behaviour. But if our dog is chewing things over and doesn’t like this or that about a dog, say it stares at them a while too long, then this isn’t really all that reflexive or involuntary.
It’s much more thoughtful. Or, it’s a very, very specific trigger and we haven’t found the specificity yet.
Now let’s imagine our dogs have a very specific trigger: being eyeballed by other dogs. Every time they are eyeballed by another dog, they start barking. If we’re going to pair up food and eyeballing, then that bar can only open when our dog gets eyeballed. If we make a pairing with all dogs and food, then we’re not likely to change our dog’s feelings about being eyeballed, and we’ll be turning all dogs into a predictor of food.
We can’t afford to be clumsy.
What’s the consequence of clumsiness?
Sometimes we can also inadvertently make things even more important to our dogs rather than less important. If our dog is not reacting to every single dog, human or car, then we can inadvertently be teaching our dog to look out for things because they get food for doing so.
You might know this yourself. When it’s coming up to time to finish work or for a lunch break, you might start clock-watching in anticipation. Or if you’re waiting for a parcel delivery, you might end up spending a lot of time looking out of the window for the van to pull up. In this case, free time, lunch or parcels bring us a lot of joy. The clock and the van arriving predict the joy. Just like we hope our dogs’ triggers will predict the joyful arrival of food or toys or games for them, they too can end up scanning for signals that predict when food will happen.
Lots of our dogs do exactly that right before a meal time, for example. They are waiting for us to give the signal that it’s food time.
We don’t want our dogs’ triggers to become something that our dogs are looking out for, but this can be an unfortunate consequence of using these methods. Just like Pavlov’s dogs were looking out for his assistants entering the laboratory, our dogs can be looking out for the things that also predict food.
How do we avoid problems?
The first is by moving quite quickly from pairing to phasing out positive things like food and toys for the least interested looks.
We can also move quickly to building new behaviours. If you’ve been following the series of articles on the roadmap to resilience, you’ll know that a crucial skill for reactive dogs is being able to disengage. We can very quickly turn our dog’s triggers into cues to move away.
For example, the first couple of times the bar opens, our dog is going to be surprised and delighted, but they won’t realise yet what’s causing it. Just like if you were given a jackpot lottery win on some random Tuesday afternoon, you wouldn’t know why you’d won it. You’d be delighted, just as our dogs are, but you wouldn’t connect it to the world.
The next time it happened, on a Wednesday morning, you’d probably also be delighted, but you wouldn’t always know why you’d won it.
Imagine that you got a Jackpot Lottery win for completing your timesheet at work. On that first Tuesday, you pressed ‘send’ on your timesheet and a fanfare happened. ‘Wow!’ you thought. ‘How cool!’
The next week, you’re a little late, so you only press ‘send’ on the Wednesday morning. Then there’s a fanfare and the penny starts to drop.
You can see this in this fun episode of The Office:
When the third week, you press ‘send’, you might be waiting to confirm your theory.
And when the Jackpot noise goes, you’d think you’d cracked it. The fourth week just confirms your theory. After that, you’re expecting a jackpot every single time you press enter on your time slip.
So how does this resolve problems?
As you see, Dwight puts his hand out in expectation of the delivery of the treat. Our dogs are going to also change their behaviour in anticipation of the food, toys or play. If we’ve been working on disengaging and we feed them away from the things that trigger their barking and lunging, we’ll also get that behaviour in anticipation of the food.
Jim does exactly what we’ll do. We pause a microsecond to see if our dogs are anticipating the food. What we’ll find is that if our dogs have made the connection and we hesitate, they’ll turn to us with the same, ‘Hey! Where’s the stuff?!’
And that delicious moment is when we can either ask them for a different behaviour, like a u-turn, or we can just take a step back before we feed and our dogs will probably move with us.
Over time, as our dogs spot the stuff that makes them feel all ruff, we won’t even need to say anything. They’ll turn to us expecting a handout.
We can keep this up a while, but then start phasing out the handouts for the lesser disengagement. You can see that in this video of my dogs:
I love this video as it shows so clearly that my dog Heston is simply treating worrying noises outside the home as a signal that food will happen. He’s clearly still on alert, but you can also see that he is listening out for noises – the opposite of what I want! – because he wants a treat. We don’t want to make those noises even more important or significant than they already are to our dogs.
Pavlov can be very powerful, especially if we use it to create voluntary behaviour rather than just trying to change how our dogs feel about things. It is at its best when we have clarity in our techniques and when we’re not sloppy in our methods.
We’ve also got to know when and how to use it to create behaviours that will help our dogs with their barking and lunging, like looking to us rather than racing to the gate to bark at neighbourhood noise. Pavlovian methods are very much a bridge that we want to use as efficiently as possible. It shouldn’t take eight months to change our dog’s behaviour.
We may never be able to change how they feel about the world. In fact, we may not want to – it’s as valid to have negative emotions about the world as it is to have positive ones. This way, we validate their emotion and also we help them find other ways to deal with it rather than those enormous, big behaviours.
At the same time, it is really important we don’t create a monster. Turning the triggers into simple signals that good stuff will start is one thing; causing our dog to start scanning the environment for the very stuff that bothers them is another. As you see in the video with Heston, my girl Lidy just ignores all the tractor noise. That’s our goal. Our anxious dogs should be so used to the world that we’ve normalised everything in it. We definitely don’t want to make it more interesting!
To find out more
You can check out the roadmap to resilience
You can also find the Lighten Up self-study courses here