Unhinged Contingencies (A Rant by Sarah)

The First Principles of Dog Training

At The Dog School, Katrina and I look to train people in the first principles of dog training, rather than simply guiding students through a series of rote protocols. It’s a bit of a “teach a man to fish” situation, except it’s more like “teach a dog to sit, and he’ll sit. Teach their owner how to train the dog to sit using first principles of dog training, and they’ll teach the dog to do lots of cool behaviours given enough time to practice”. That’s a bit of a mouthful, though.

The first principles of training are the fundamental truths which make animal training possible. First principles do not care if you’re a cowboy or a cat lady, whether you’re “balanced” or “force-free”, or even whether you actually know what you’re doing or not. If it’s working, there’s a reason why. It's not magic. You don’t have to understand the first principles of training to gain success with training animals—it’s sufficient to follow good advice and sound protocol. This is a huge problem, though, because protocols and sound advice tend to get garbled through the retelling, especially if they’re repackaged to be “punchy” for Social Media. A lot of internet training advice is just repacked or tweaked protocols which have lost touch with the first principles which make them work, and, as a result don’t work at all anymore or work very inefficiently. If you’d like an example of this, go read my blog post on teaching “settle”.

When I learned to train animals, I had a really large and diverse group of mentors and fellow students to discuss training topics with at length. These were essential to my development as a trainer, because I was able (and encouraged!) to question why everything worked. We would frequently take apart training protocols, tweak them, and put them back together again, just to see what would happen. Each time I did this, I learned more and more about the first principles of training which make the protocols run smoothly. My goal with this blog is to bring some of the long-winded, conversational rambles I have with my friends (mostly Katrina) to the internet, so that they are accessible to anybody brave enough to step into this discussion. In doing this, I hope you’ll learn to look at a training protocol and ask “Okay but why does this work?”... You might even be able to answer, because you read about such and such a first principle in my blog. 

In this article, you’re going to come on a journey with me about contingency. It’s a word we use a lot in dog training that means very little to non-trainers, and yet I rarely see it taught or spoken about at any level. Newer dog trainers don’t always get it, and advanced professionals seem to take it for granted. Not only that, but there are some side-bars to explore which are often left unaddressed. Understanding contingency is fundamental to understanding how to train any animal, so it’s frankly insane how little discussion there is out there on this topic, especially given that it’s a first principle of dog training. In fact, I know multiple very talented dog trainers who have re-invented this wheel on their own, simply because nobody was around to tell them. This blog post is my attempt at filling in some of these gaps, and at least providing some discourse. It’s not neat or tidy, and there are no real take-home messages. Come hang out with me in theory-land for a while.

Contingency in Training

Contingency, in the context of dog training, refers to the links between antecedents (the events that happen immediately before the behaviour) , behaviours and consequences. Before a contingency is established, two or more events happen together in such a way that one event can predict or bring about another. For example, I might click my clicker, then feed my dog a treat. This can be annotated as:

Click - Treat

Once the pattern has been understood by the animal, they will begin to link the events and understand that getting a treat is contingent on the clicker sound happening. Established contingencies (known as functional relations - for those wanting the extra nerdy version) can be annotated with arrows, as:

Click -> Treat

At The Dog School, Katrina and I tend to write all of our training exercises down in the form of flow diagrams with dashes linking each step with the next. These dashes (or arrows, if the contingency is now established as a functional relation) represent the established contingencies required for the completion of the skill. For example, if we are training a dog to lie down using a food lure, we might present the food lure so that the dog lies down, then we will click and deliver a treat. The annotation would be as follows:

Present Food Lure -> Dog Lies Down -> Click -> Release Treat

Each step hinges on the preceding step. If a step breaks, the whole sequence breaks because you can no longer proceed with the flow chart. This is what we mean by contingency.

Loopy Training

Our flow chart annotations are nothing new, trainers have been using these for decades to represent antecedents (cues), behaviours and their consequences. Really good trainers also use loopy training, which takes a linear flow chart and makes it a circle. The advantage of working on a loop rather than a line is that by the time you finish a repetition you’re back to the beginning of the loop, automatically reset to begin over again. This can keep training fast paced and fun for your dog, while enhancing the clarity of the exercise because the dog is always anticipating the next step in the loop. A good example is capturing eye contact from your dog. Toss a treat on the ground, let your dog eat it, and then wait for them to look back up at you. When your dog makes eye contact, click and then toss another treat on the ground. See? Now we’re back to the beginning, and ready to click and reinforce eye contact again. The annotation looks like this:

Toss Treat -> Dog eats treat -> Call Dog’s Name -> Dog makes eye contact -> Click -> Toss Treat 

Loops are super powerful because all the steps are contingent with all the other steps. The first couple of loops might be slower, but, as your dog understands the game, the momentum grows. You’ll find yourself rapidly speeding up as your dog spins through the loop, and that momentum is a potent signal that your dog has successfully learned the cue(s) and behaviour(s) within the loop. Without the loop, it’s harder to track the learning momentum, and you can lose some clarity of instruction.

Beyond Loops: Taking off the Training Wheels

Loops are amazing, but they are for learning and not for real-life. I like to think of loops as “training wheels” because they’re really important for early phase learning, but they should be deconstructed for the real world. It’s also one of my better attempts at word-play, so I hope you enjoy it. 

Real life training is not done on a loop. Real life is a really complicated series of events with multiple behavioural options available to both yourself and your dog (or, in extra nerdy terms, there are many competing contingencies and motivating operations in play at all times). It’s your job to guide your dog through the options to pick the best choices available for both of you. Training loops give you behaviours (on cue!) which you can call on later when needed. For example if my dog sees an exciting dog across the street, I can call her name and ask her to look back at me for a treat. It’s not a loop anymore because I don’t need to reset it for another rep, I only need to use it one time, but it was taught using the loop above.

For most dogs, loops are comfortable and predictable, so we phase them out gradually rather than all at once. Loops can be phased out gradually by:

• Switching between different training loops within the same session

• Making loops bigger (more behaviours within them)

• Making loops longer (adding more duration between steps)

• Adding additional “check-points” within loops, where your dog controls when the next step happens by offering the next behaviour off-cue. This is particularly excellent for adding distractions.

Degrees of Freedom Increase with Number of Contingencies.

As a rule, training loops provide extreme clarity (the dog understands what we’d like them to do), at the expense of our dog’s ability to express a wider variety of behaviour. Breaking the loop breaks the exercise, so the only route we give our dogs is a choice between co-operation or non-compliance. Even if there’s no bad outcome if the loop is broken, giving any animal only a single pathway to success is at best unrealistic and at worst coercive. I’ll give an example. Let’s say I offer you $5 to clean my toilet. That’s somewhat unrealistic. What about if I offer you $500 or $5,000? The higher “value” the reinforcement, the more compelling the behaviour becomes even if you didn’t want to do it. What if I take away your option to say no by locking you in my bathroom until the behaviour of toilet cleaning is complete? Now it seems quite coercive. Forcing behaviours to happen by using high value food or heavy management which prevents the dog from making any other choices is training tyranny, and it happens all the time. I personally see this most commonly in behaviour modification; the dog is asked to perform a loop under stressful conditions. If they fail, the value of reinforcement is increased (just use better treats!) or some kind of management is used to prevent opt-out behaviour (using a leash to prevent the dog from simply leaving). Running high-stakes loops with no way for the dog to opt out of them is pretty coercive even if there’s no punishment.

The first thing you can do to prevent this kind of training tyranny is to simply build in an “off-ramp” behaviour for every training loop you practice. Training an offramp is something that your dog can do to say “I’d rather not do this right now”. It’s you knocking on my bathroom door saying “I’m not interested in cleaning your toilet, Sarah. Let me out please!” and knowing that I will let you out and won't be mad about it. By adding an off-ramp to your loop, you add a new set of contingencies which earn the same (or equivalent) reinforcement.

1. Toss Treat -> Dog eats treat -> Dog makes eye contact with handler -> Click -> Toss Treat 

2. Toss Treat -> Dog eats treat -> Dog goes to place -> Treat on place -> Training session ends

Adding an additional set of contingencies to a training flow chart increases the “degrees of freedom” the dog has. Degrees of freedom are a measurement of how many choices the dog has available, and it’s measured by taking the number of behavioural options with the same outcome (e.g. earning the treat) minus one. If a dog has 2 ways to earn the same treat, they have 1 degree of freedom. The more degrees of freedom the merrier for choice and agency, but at the cost of clarity in learning. A dog given 20 ways to earn the same treat likely won’t learn very many of them, and there’s a strong chance that they’ll become confused in the process because they don’t understand what you want. It’s a balance.

I’m not saying you should always give your dog a degree of freedom. There are times when you have to make your dog do something they didn’t want to do (like experience a nail trim or an ear treatment). Degrees of freedom become important for things like reactivity training; if we expect our dog to do a simple counter-conditioning loop endlessly (look at the trigger -> eat a treat -> repeat) and don’t give them a way to leave when they’ve had enough, we trap the dog in a loop. Ultimately, this causes training to stall, or even regress.

Remember when I said that you don’t need to understand first principles in order to be a good dog trainer? Something I see very frequently in good dog trainers is that they are able to read their dog’s expressions and body language very well and vis-versa. This open dialogue channel allows the trainer to act on very subtle off-ramp behaviours which are intuitive to the dog. The dog might simply glance away, and the trainer will read that as “ok, time for a break”. The better you are at reading the dog, the more degrees of freedom you can effectively give them because you can recognise and respond to their needs effectively. Good trainers don’t trap dogs in loops because they’re intuitively aware of when they are doing so and take steps not to. The difficulty comes when a student focuses so hard on the loop, or their role in making the loop work, that they don’t read the dog well. In these cases, training a deliberate off-ramp behaviour can allow the dog to say “Hey! I’m struggling over here and you’re not paying attention”. It’s not a magic solution, but it is a practical one. We can’t all be flawless trainers all the time.

Broken Contingencies and Non-Contingent Consequences

Broken Contingenies can do interesting things in dog training. Usually interesting-bad. For this next part, you’ll have to broadly understand the operant conditioning quadrants (this blog is really aimed at fellow dog-trainers and more advanced students), but I'll try my best to make it accessible for everybody else.

Breaking a positive reinforcement contingency leads to a negative punishment. If your dog expects a behaviour to lead to positive reinforcement (adding something the dog wants), and it doesn’t, they’re going to feel confused, disappointed and frustrated. If I said sit on that chair and I’ll give you a glass of wine, you’d be pretty annoyed at me if you sat on the chair and I forgot you existed. 

Broken Contingency: Cue to Sit -> Sit -/-> Treat(sitting is punished). This happens all the time in dog training when our dog thinks they’ll be reinforced for doing a certain behaviour, but we fail to provide a treat. This is why cues and contexts are so important! Cues signal to your dog when reinforcement is (and isn’t) available, and context cues from the environment are important as well. Be consistent with cueing, so that your dog knows that your cues carry weight. If you do break a positive reinforcement contingency, your dog will be disappointed and both the cue and the behaviour may weaken. This can usually be recovered by being more mindful in the future.

Breaking a positive punishment contingency leads to negative reinforcement. If your dog expects a punishment to occur and it doesn’t, your dog will feel some relief. Furthermore, they will likely look for signs that a behaviour won’t be punished “this time”, because the un-punished instances are actively being negatively-reinforced. This is why aversive-inclusive trainers will tell you that consistency is really important (because it is). It’s also one of the (many) reasons I choose not to use positive punishment in training, the level of skill required to make it work fairly is actually quite high. While I’ve never been an aversive-inclusive trainer in the dog world, I cut my training teeth working with horses where aversives are the norm. It’s easy to see this effect in horse training; lesson horses are famous for squishing their novice riders into walls, ducking to eat grass, ignoring cues and slamming on the breaks. Because low-skill riders aren’t effective at using their tools to punish classic unwanted behaviours, the horses become cheekier over time (unless the coaches are working hard to counter this). If you’re not a horse person, let’s look at a human example: speeding. People know that the contingency is speeding -> get a ticket. Except, the enforcement rate is low. Everybody knows they can get away with speeding just-a-little-bit, so everybody does. Only flagrant speeding is punished, and even then not always. This leads to speed limits becoming effectively about 10% higher than posted. It also explains why some dogs pull “as much as they can get away with without you really noticing” when on leash.

Broken Contingency: Speed limit sign is “60 km/hr” (cue for speed) -> Speed at 70 km/hr -/-> speeding ticket(speeding is reinforced)Broken Contingency (A dog example): Dog on leash (context cue for loose leash walking) -> Dog pulls -/-> leash correction(pulling is reinforced)

Low-intensity punishment is particularly susceptible to this phenomenon, meaning that you have to use higher intensity of punishment to keep those unwanted behaviours suppressed. If you are very good with being consistent and clear with your contingencies, you can use low-intensity punishment more effectively (leading to fairer training). High skill aversive trainers can and will achieve better results with less frequent and less intense positive punishment than lower skill beginner trainers for this reason.

Non Contingent Reinforcement and Punishment happens when the dog isn’t expecting the consequence at all. From their perspective, the consequence happens “out of the blue”. This can happen for so many reasons: 

• We think we’ve established a contingency, but we have not

• We drop treats accidentally, or dole out punishment subconsciously

• We are trying to reinforce or punish an involuntary behaviour

I don’t often deal in absolutes (i’m not a sith), so I don’t usually make statements like this but I’m quite confident in this one: Non-contingent positive-punishment is always bad. I can’t think of any time it’s beneficial for a dog to receive a punishment that they didn’t see coming. I see the fall out of this when handlers have tried to punish a behaviour the dog is not doing intentionally. The saddest version of this is when an animal is acting out due to pain, fear or discomfort, and then experiencing some kind of physical correction (for example trying to punish a dog for whining in the car when they are carsick). On the other hand, non-contingent positive reinforcement (giving your dog treats, praise, play and attention) is extremely beneficial. Having fun with your dog for no reason, as it turns out, is actually a brilliant way to connect with your animal. Non-contingent positive reinforcement, particularly play, is extremely powerful for shy dogs in particular.

Broken contingencies and non-contingent consequences highlight some of the flaws with approaching dog training from a dogmatic operant conditioning standpoint. We don’t usually break contingencies on purpose, it happens when we aren’t clear with our training set up, or we aren’t being mindful of our dog’s behaviours and motivations. Operant conditioning isn’t some magic formula or equation that you can write on a board, it’s simply a guide, like scaffolding, to help us build a training picture. You have to understand your dog’s expectations and their mental state. If you don’t, those little arrows on our training flow charts will simply become lost in translation.

Conclusion

One of my favourite things about behaviour is just how complex it is, and just how adamant we humans are to try and simplify it. I’m reminded of my days as a cell biologist, looking at polysaccharide degradation pathways in the human gut. There are a mind bogglingly large number of proteins at work in the human body, doing thousands of independent little jobs, and here we are diligently cataloguing each of them in isolation. There’s so much value in doing this, of course. It’s necessary to understand each component in isolation, before we can investigate its larger role in the big picture of life. Contingencies are similar. We can look at one or two in isolation, we can train and study them. But don’t mistake this for simplicity. Dogs (and people) are complex beings, and there are more contingencies out there than we can hold in our minds, all overlapping and influencing our (and their) behaviours. 

As trainers, it’s our job to strengthen specific contingencies as a way of communicating with our animals, but don’t let single contingencies fool you into thinking training is robotic or mechanical. Behaviour is vast and complex, like the stars in the sky or the proteins in a cell. Focusing on one little thing in isolation does not give you the full picture of all behaviour. It’s important to step back and look at the whole picture, not just a fragment in isolation, to ensure you are on the right track.