When is failure okay?

Okay, so this is something that’s been on my mind a lot lately. When is failure okay in training?

If life were as dry and clear-cut as a textbook, the answer would be pretty easy to figure out. A 20% failure rate is ideal in operant conditioning. If you’re doing classical conditioning, you don’t want any failure: the dog should be kept under threshold all the time.

A study in avoidance and calming signals, this is pretty unhappy body language.

Avoidance and calming signals: stressed body language.

But what about real life? What about my dog, where everything we train is a combination of classical and operant conditioning? Layla’s baseline reaction to anything new is anxiety and distrust. The first performances of any new behaviors are usually either very hesitant, oftentimes with stiff body language or whale eye, or accompanied by frantic movement and repetitive yapping. This means that any time I teach a new behavior, I’m mixing the two (OC/CC) up in a hodgepodge of what probably looks like very poor training. Jackpots are the rule, not the exception. Even when my dog makes a mistake, my voice remains happy. I praise her for trying even if she fails. If I stop doing so, she’ll shut down or stress out on me. But it’s not easy to explain why I seem to be praising my dog for doing it wrong.

I can’t imagine what a mess this would be if I began to train coercively. As it is, I try to capture most behavior or lure through targeting. I really like shaping, but her tolerance of failure in shaping is very low. I’m not often skilled enough to build a behavior through successive approximations while also keeping her engaged and calm.


Alert but relaxed body language.

As we’ve continued to work together, this fearful stage of new training has become shorter and shorter in duration. Layla is able to gain confidence more quickly. But she still has a hard time accepting failure, and sometimes I wonder if my CC/OC hodgepodge isn’t confusing her more. Right now we’re working on stand-stay for Level 3 Rally. The first couple short stays were very uncomfortable to me: her tail was tucked, her eyes were wide, her ears were pinned back, basically she looked like I was torturing her. I jackpotted the short stays with a bait-bag dive (one of her favorite rewards). After a couple more trials in a few different environments, she started to look more comfortable and was able to start holding her feet completely still rather than shifting nervously. But if I raise criteria when she’s successful 80% of the time, that 20% failure rate stresses her out.

This isn’t just something Layla and I struggle with, either. I worry that I might be doing my students a disservice by not differentiating between CC/OC. In reactive dog class, we do a lot of classical conditioning. Yay, there’s a dog walking across the room! Here, have some cookies. But we’re also asking for operant behaviors: Look at that dog. Watch me. Keep your leash loose. Take a deep breath. If we never have a reaction in this class, we define the class as a success. However, if the dog is able to bounce back from a reaction well, is it really that bad to push them slightly? We’re asking for 100% success before allowing them to continue to the next level of training. Is this unrealistic? And how do we teach our students what to do when their dog reacts if we never  allow a reaction to happen when we’re there to help? And then there are the dogs who have been working on their reactivity for some time. They may react, but that emotional component seems to be missing. It’s almost like they’re reacting out of habit: it’s become a learned behavior. They almost beg for that 80% success rule in their training plan. If it’s not emotional, classical conditioning just doesn’t make sense. But how can we tell for sure which cases are which? When is failure okay?

So, who do I listen to for training advice? Susan Garrett and Bob Bailey are amazing gurus who have trained animals to do complex tasks based on the 20% rule. Leslie McDevitt and Ken Ramirez are incredibly gifted people who get to high levels of difficulty by keeping the animals completely under threshold and helping them be successful. The waters between classical and operant conditioning seem very muddy to me for so many behaviors.

So, how do you train? 80% success? 100% success? Something else entirely? When is failure okay in your training plan? What do you do when you’re working on behaviors that have both a classical and operant conditioning goal?



Filed under Layla, Training

6 responses to “When is failure okay?

  1. Crystal (and Maisy)

    This is such an interesting question/topic.

    I think failure is, generally, a good thing. It’s feedback, if nothing else. When Maisy fails, it tells me something useful about the situation. It might mean I raised my criteria too quickly. It might mean that I underestimated how scary that person/dog/whatever is. It might mean that she’s not feeling good. Whatever the case, it helps me learn how to adjust the way I train to help Maisy more in the future.

    I read this great book this weekend, and it had a quote that went something like: If at first you don’t succeed, redefine success. I’ve seen a t-shirt that says “CU students never fail, we just lower our criteria” (or something similar). In both cases, I think it’s true. Failure gives us information, which makes us successful in the long run. It teaches us to try again in a different way.

    How can we learn what stay really means without testing that boundary? How can we learn to bounce back from stressful events without it? I think with our dogs, what’s important is setting things up to minimize failure, and then deal with it when it happens. I don’t think we should plan for whatever an “ideal rate of failure” might be, but I think we should look at how often we are failing, and adjust our behavior to help minimize that stress in the future.

    I’m sure I’ll have more thoughts on this later, but for now: vet chiro and training class! 🙂

  2. Great post, Sara.

    Classical conditioning work is not a percentage kind of a deal. There is no behavior involved in classical conditioning whatsoever.

    Leslie McDevitt’s LAT game is very interesting to me because at once it’s both classical conditioning and operant conditioning at once – it’s the only behavior I know of that work that way.

  3. Hey Sara,
    I have a question… Do you use variable reinforcement? If so, how do you use it?

    Mark no cookie or no mark no cookie?

    • Hi Ron,

      Yes, we use variable reinforcement. How I use it varies (haha).

      When I’m using the clicker, the click is always followed by a treat, toy, opportunity to chase squirrels, or something equally potent. Always. I use the clicker when I need that precision or the very powerful positive CER to a marker.

      However, I will variably reinforce by using a verbal marker (yes) as needed. The verbal marker is usually, but not always, followed by a potent reinforcer (food tug etc). Sometimes I follow the verbal marker with a cue to do another behavior that she likes as well, sometimes I use it with no immediate reinforcer sort of as a keep going signal. She still responds positively to the verbal marker, but it’s less powerful than the clicker.

      The other thing I will do with behaviors that are conducive to this is to just provide the reinforcer with no marker (sometimes accompanied by praise). For example, when teaching her to ‘get out’ I may just chuck a treat or toy past her as she’s running away from me. It’s pretty clear what’s being rewarded in that instance and the reward placement serves to clarify the picture, so no marker is needed.

      I agree with you about the LAT game, but would respectfully disagree that it’s the only game that combines classical and operant conditioning. I believe many of the other CU games will also serve the same purpose when introduced and used appropriately. That’s why I love the program so much. If we know Pavlov is always on our shoulder, we might as well utilize him rather than just hoping he’ll go away if we ignore him! 😉

      • So, when you are using a clicker, the click is always followed by a treat.

        Does that mean that you are sometimes ignoring the correct behavior (meaning no click) to get a better expression of it?

        Do you find the same degree of anxiety with your verbal sessions as you find with your clicker sessions?

        I think that the click always followed by a treat is not variable reinforcement, but variable marking of behavior. I know I run afoul of some very important trainers on that, but I’ll be happy to chat more about it.

        Also, I think you misunderstood my comment about the LAT game.

        I understand that classical and operant are quite inseparable and that Leslie’s stuff is very good at hooking the two up with continuity, but LAT is different.

        The appearance of the stimulus = treat which is classical and the act of looking at the stimulus = treat which is operant. The other stuff is either classical + operant or operant + classical.

        LAT is at once both classical and operant. I think it is a fascinating technique because .

        Do you know of any other training method that is at the same time classical and operant like that?


    • Ron,

      If I’m in the middle of a clicker session and want to variably reinforce, I just throw in a verbal marker in place of the click. I know a lot of trainers like to just withhold the click, but I find that is too stressful for Layla. No click makes her think she’s done something wrong. Verbally marking it helps her know that she’s still on the right track, she’s just not getting a big prize for that particular instance.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s