Why you should Run from Engage Disengage - Best Practices for Reactivity Training Your Dog

Recently while training my horse Samwise (who has a serious fear of people), I came to the conclusion that thinking in quadrants and protocols is a fatal flaw in applying first principles. I also have spent the last several weeks doing a lot of behavioural analysis on a really specific use case that I think applies to many dogs and horses. Before I dive into it, let’s go over contingencies and really make sure they make sense. When I first explained this to a friend of mine, she told me I sounded like Charlie Day in It’s Always Sunny in Philadelphia and sent me a meme of him in front of his cork board. 

 When you truly understand behavioural contingencies and how to apply them, the quadrants and protocols we’ve come to obsess over become a much smaller piece of the behaviour puzzle. When we work from quadrants and protocols before understanding first principles we aren’t applying good learning techniques for the human. Failing to understand the “why” behind the “what” is a weak starting point. Instead, I propose starting with the famed contingency analysis. 

In contingency analysis, a la BF Skinner, the three term contingency consists of an Antecedent, a Behaviour and a Consequence. In that order. The consequence always follows the behaviour. The consequence selects for and acts on the behaviour. An antecedent is anything that precedes the behaviour and is context for the behaviour.

I know we all know this, but I'm hammering in our definitions early and for good reason because this gets very wild very quickly.

In training scenarios, we are quick to presume that click-and-treat is an R+ contingency (A→B→C [R+]) ; however, in one very specific and very common training scenario this is, in fact, completely incorrect.

Let's take a look at a simple example to start with.

A Target is presented → Dog looks at the Target → Food is Delivered  A → B → C

If an R+ contingency is present, we can expect the rate or relative frequency of the dog looking at the target to increase over subsequent trials. In less "behaviourism" terms, what we're doing is reinforcing looking at the target with a well timed click-and-treat and we would expect that the behaviour we want (looking at the target) to increase and become more robust on subsequent training sessions.

Because we've proven in this scenario that the consequence (food delivery) has increased the rate or relative frequency of the behaviour (dog looking at the target), we can say that the behaviour of looking at the target has been positively reinforced.

In reactivity cases, we often see the following set up by trainers:

Trigger occurs → Food is delivered in place

This is what is commonly referred to as an Open Bar/Closed Bar protocol. Note here we're very much missing something in our behaviour column because what we are wanting from the dog is to not react or to not bark or to not run away. Ah, the slippery slope of non-behaviours…! But of course there's always behaviour happening, so if we think about what the dog is doing in this scenario it would be more accurate to think of it as follows:

Trigger Occurs → Food is eaten [the replacement behaviour for lunging, barking, running away] → Trigger Disappears

Dear Gentle Reader, may I challenge you to answer the question of what is actually reinforcing what in this above scenario? Think back to our A-B-C contingency and where B and C are in this scenario. If it’s not clear to you yet, don’t worry, just keep reading! Protocols like Engage / Disengage  are clever because they have solved for the non-behaviour problem we identified above by adding in a behaviour in our contingency plan:

Trigger appears → Dog Looks at Trigger → Food is Delivered. A → B → C. Perfect!

Now, in many cases this is perfectly appropriate! We can see that the dog is successfully increasing looking at the trigger and the trigger is successfully paired with the delivery of food and reactivity signs decrease over time we can pat ourselves on the back and be very proud of ourselves for having changed the emotional punch of the trigger into something a lot less explosive.

In other cases, however, this is not how this is actually operating.

Scared animals want distance from the scary thing.

If you've got a phobia of spiders, and I tell you that everything is fine because I brought your favourite cake to the spider den, I'm not sure the cake is going to have any effect. Regardless of how much you love cake, how much cake is being offered or how impressive the cake looks you're just not likely to be cool with a cake in front of the spider. In some cases, a careful desensitization protocol using cake could work. We get your consent and we start at a very far distance from the spider to the cake and we work our way closer and closer. Ideally your fear subsides and you can now tolerate the spider in these controlled scenarios. BUT, if we happen upon a big hairy spider in the wild who just jumped in front of you, chances are eating cake - even the most wonderful cake you've ever had - in this scenario is not going to change your feelings about the spider. And, distance from that spider is going to be your most desirable reinforcer. Fear based behaviours are really distance seeking behaviours. If we use that distance as a reinforcer instead of trying to suppress unwanted fear-related escape and avoidance behaviours we can design some very cool training.

Let's write this out as a contingency analysis to see what's actually going on when escape is the motivating operation.

When escape is the MO, the aversiveness of the trigger increases the effectiveness of distance as a negative reinforcer and increases the likelihood of all behaviours that have previously produced distance. The animal's natural escape contingency looks like this:

Trigger → Escape behaviour (flee/lunge/bark) → Distance gained (–R)

But in a typical Open bar/Closed bar setup, those escape behaviours are unavailable. The dog can't leave. The trigger doesn't go away until the food routine is complete. So what we've actually arranged is:

Trigger → Eat food → Trigger removed (–R)

The food is not functioning as an added reinforcer here. It is the trigger being removed that is selecting for the behaviour. Food eating is the behaviour being negatively reinforced by the disappearance of the trigger, and escape behaviour is being blocked.

Engage/Disengage protocols follow the same logic:

Trigger Occurs → Dog Looks at Trigger → Food Delivered in place for multiple reps

and then distance is added.

Now let's look at what this means in practice. When avoidance and distance are an animal's number one desire in a situation, we run into an issue whereby:

Trigger Appears → Dog Looks at Trigger → Dog Eats → Distance Gained 

A → B1 → B2 → C

In this scenario, the DISTANCE from the trigger is contingent on a behaviour chain of looking and then eating. We cannot forget that food eating is a behaviour!

In this scenario, the dog learns that eating the food leads to the trigger disappearing or gaining distance from the trigger. And as the trigger leaving is the most desirable outcome, we are training food eating and increasing the rate and relative frequency of food eating in the context of the trigger.

What's really problematic about this set up is the fact that we have blocked the act of escape behaviour happening. We are not replacing behaviours and we are not using a DRO. Why? Because the trigger disappearing is squarely in the consequences column of our behavioural analysis and, as we discussed earlier, food eating is the behaviour being operated on.

So, distance seeking behaviours are not being “replaced” by food eating when the dominant MO is to gain distance – and this is especially true if we're withholding food for duration near a trigger. Remember that we've made leaving contingent on food eating so by waiting to feed the food, we're not changing escape or avoidance behaviours we're simply not allowing the dog to leave or for the trigger to disappear. This is my number one issue with open bar/closed bar in highly stressful situations. 

It's worth noting the irony here: the standard advice for these protocols is to keep the animal 'under threshold' and at an 'appropriate distance.' The fact that appropriate distance is considered essential tells us that distance matters but it is often treated solely as an antecedent not as something the animal is actively motivated to seek. The moment you're not at the right distance, the animal's desire for distance doesn't just disappear! 

Sarah and I have a set up at The Dog School for reactivity cases where we know gaining distance is the safest option and where dogs learn that leaving a situation and/or taking a few steps away is a much better answer than lunging and barking. Turns out you don't need to make the thing go away if you yourself can go away!

In our set up, Sarah and I use the following:

Trigger appears → Dog looks at the trigger → Handler and Dog U-Turn away → Food is fed away from the trigger

In this way, the behaviour chain includes walking away from the trigger. Look, Gain Distance, Eat.

We've put food at the end of the behaviour chain and squarely out of the behaviour box. We're actually operating on 4-term contingency here:

Trigger Appears → Dog Looks → Gain Distance [R-] → Food Delivered [R+]

The most clever thing about this set up is if the dog doesn't require distance (because they are perfectly fine to work for food in this set-up) there is no fallout as gaining distance in the behaviour chain does not diminish the food reinforcer; gaining distance is simply a behaviour in your chain (i.e.: Instead of operating as R-, our contingency becomes Trigger Appears → Dog Looks (B1) → Gain Distance (B2) → Food Delivered [R+]) and in cases where distance is the most or only available reinforcer to the learner, we use distance as a negative reinforcer and then reinforce the entire chain with food. It is my opinion that using distance from the trigger in this way is a much safer and more animal-friendly way to work with fear than open bar/closed bar types of operations. This set up allows for both R- and R+ contingencies to occur and is especially important as we can't know which reinforcer the animal desires if we don't have both options available.

It’s also easy for non-trainers to implement as it doesn’t require expert body-language reading to determine the optimal distance to train in and offers a practical and applicable way to use this “in the real world.”

I think it can be quite dangerous for us to not consider that R- may be the most reinforcing scenario as in some cases, if we are using food eating as a behaviour by accident we may in fact simply be suppressing an escape behaviour. It’s important for me to note here that this blog is in no way encouraging or endorsing the use of human mediated aversives so long as you use food at the end of your behaviour chain… I’m advocating for a very specific functional reinforcer-first approach that is environmentally mediated.  

In summary, when there's a strong Motivating Operation like predation or escape in play, it is my belief that the functional reinforcer has to complete its contingency before R+ can operate cleanly as an actual added reinforcer. If you deliver them simultaneously, the reinforcer gets absorbed into the MO-driven contingency and loses its function. If you sequence them as functional reinforcer first, R+ second you can operate both contingencies cleanly.

We already accept functional reinforcer first, R+ second in many of the fantastic dog-training protocols out there. Predation Replacement is one such example, BAT is yet another. Even Engage-Disengage, when used at an appropriate distance, partially captures the functional reinforcer before layering R+. So I don't actually want you to run away FROM engage-disengage, I'd just like you to add some running away in engage-disengage as part of your protocol (see what we did there?).

What I'm really keen on having the reader understand here is the way in which the contingencies can cause problems when we stay in a quadrant without understanding the learner and when we stay stuck in a protocol without understanding functional reinforcers. We know distance from a trigger is important when training: it's the first step in nearly every single protocol! Why are we then not operating on the behaviours that gain distance instead of avoiding all triggers or spending years in desensitization protocols that don't work when the spider jumps out at you without warning (my reactive dog people… why is there always a loose dog around a blind corner?). 

We can't always control the environment but we can be smart about contingencies and when the proverbial shit hits the fan I personally think it's deeply important to create intelligent, thoughtful distance seeking behaviours that encourage safety, autonomy and which can be modified so blind panic doesn't ensue. I also am firmly of the belief that food eating behaviours in these situations aren't that important and that we can spend a lot of time training food eating by accident and this feels to me to be a bit of a waste of time.

A practical thing to note… If you notice you're wanting to go up in treat value for something difficult for your animal, it could very well be that distance is a competing operation in your equation and that you can and should increase it either as the behaviour you are working on as in the above example or as a factor in why you may want to change your antecedent arrangement or environment.

So what does this mean for our quadrants and protocols? It means that in some cases, the best possible scenario is to use negative reinforcement. It means that we have to let go of dogma and do behavioural analysis correctly and let go of beliefs that may actually not be true. It also means that when something isn't working, return to contingency analysis. Identify what's actually being reinforced and ask whether the behavior you think you're training is truly in the behavior column.

I’ve created a companion tool to this blog for anyone interested in exploring contingencies and motivating operations in more detail. The program is currently in development and very much in a BETA phase. If you see any errors or have concerns about the explanations for specific contingencies, please contact me at info@thedogschool.ca  Please have a play around with the Contingency Analyzer Tool! www.thedogschool.ca/contingency-analyzer