Thursday, September 15, 2011

Kathy Sdao Seminar: Why That "Treat Thing" Doesn't Always Work

Classical counter conditioning is a great foundation for working with a reactive dog. It's also very easy to do: a scary thing happens, and then you give the dog a treat. Over time, the dog learns that the scary thing predicts good things, which changes his feelings. He's no longer scared, which will hopefully mean that the fear-based behavior will disappear. Still, sometimes you'll hear someone say “I tried that whole treat thing, and it just didn't work...”

And you know what? I believe those people. Classical conditioning, while pretty straightforward, does require that you get some of the details straight. Kathy shared the errors that people commonly make that can prevent progress, or even worse, create even more problems. If you're struggling with that “whole treat thing,” here are eight things you should check.

(If you aren't familiar with the basics of classical conditioning, check out this post, and then this one. Today's entry will make a lot more sense if you do.)

Make Sure Your Timing is Right
This is perhaps the area you can screw up the most. In classical conditioning, the second thing that happens is the most important because it is the one that will determine what kind of association is formed. If that second thing is positive, the association will become positive. If the second thing is scary, though, then the association will become scary, too. In other words, when you give the treat matters, and it matters a lot.

There are four basic ways to pair stimuli. Trace conditioning happens when there is a 1-2 second delay between the two stimuli. For example, if you want the dog to learn that a bell ringing predicts that food is coming, in trace conditioning, you would ring the bell, wait 1-2 seconds, and then give the dog the treat. Delayed conditioning occurs when the first stimuli is presented continuously for several seconds (the bell rings several times in a row), and then food is presented immediately. Simultaneous conditioning means that both stimuli are presented at the same time; you ring the bell and give the food at the same time. Finally, backward conditioning, like the name implies, is when you reverse the stimuli. You give the food and then ring the bell.

Although there are scientific uses for all of these, when it comes to training a reactive or fearful dog, both simultaneous and backward conditioning is basically useless. Simultaneous conditioning will simply have little to no result. Backward conditioning, however, is much worse. Instead of the bell predicting the treat, the treat predicts the bell. This can have devastating consequences. If a scary happens after the food does, then the food predicts the scary thing, and thus becomes associated with fear. When this happens, the mere presence of food can cause a dog to shut down or react. Not a good thing.

Most people don't set out to get this order wrong. We understand how it should work, but where we go astray is in assuming that the dog sees the scary thing at the same time we do. Therefore, you should be sure to give the treat only after the dog has perceived the stimuli/scary thing.

Make Sure You Avoid Rhythmic Trials
We humans are creatures of habit- when we set up training experiences, we tend to fall into patterns. Unfortunately, dogs are very talented at discerning patterns, so we need to make sure that our trials are not rhythmic. Kathy gave the example of working with a dog who is afraid of men. If she were to enlist a male friend and directed him to walk in and out of eyesight, it's likely that he would begin doing so at regular intervals. Unfortunately, what the dogs often learn in these scenarios is not that the presence of men predicts steak, but rather, that steak happens every 20 seconds or so, and if you get a positive association, it's very weak. You will get a far better result if you make sure that your trials are random and varied.

Make Sure There Are No Competing Stimuli
Since the second event matters most, we need to make sure that we know what it is. Sure, we might think the pairing is simple- see scary thing, get a treat- but dogs are highly observant. Does he know he's getting a treat as a result of the scary thing, or is something happening in between those two events? Common pitfalls include reaching in a pocket, or crinkling a treat bag as we get the snack out. Again, associations can and do happen in these situations, but they are weaker than if it was direct, with no middleman. Try to set up the situation to reduce as many competing stimuli as possible.

Make Sure You're Using a High Value Treat
Kathy advises that you use the absolute best thing you can in order to create the most intense response possible. Classical conditioning follows a pretty steep curve, which means that you'll get the strongest response possible fairly early on in training. Don't squander this opportunity by using kibble. Sure, the dog will learn that scary things predict food, and that's good... but it's not awesome. Think about it: do you want an, “Oh, yes, kibble, thank you” response, or do you want the “HOLY COW I JUST GOT RAW TRIPE” response?

Make Sure Your Treat Remains Special
Classical conditioning works best when the contingency is strong. In non-science speak, this means that it works best when the awesome thing that follows the trigger happens only when the trigger does. No matter how much your dog loves cheese, if cheese happens all the time, it becomes meaningless. For that reason, Kathy recommended that you have a treat that you use only for counter-conditioning. Actually, she said that it would be best if all food happened only as a direct result of the scary thing, but since that's pretty much impossible, practically speaking, she recommended reserving the high value treat for your counter-conditioning work.

Make Sure You Aren't Lumping Criteria
Remember what I said about competing stimuli? Well, that counts for the first stimuli as well as the second. Make sure you aren't trying to counter-condition against too many things at once. It really works best to tackle one scary thing at a time. So, if your dog is afraid of men, and he's afraid of people riding bikes, men riding bikes might be too much. Try breaking it down more if you can.

Make Sure the Trigger Always Predicts Good Stuff
Classical conditioning is not operant conditioning. As such, it is not maintained by intermittent reinforcement, and is much more prone to extinction as a result. Instead, for classical conditioning to work, the scary thing pretty much always needs to be followed by the treat. Every time the dog sees a scary toddler, a piece of chicken should follow. Every time. If toddlers appear and chicken doesn't, then the dog learns that toddlers are not a very reliable predictor. If you're lucky, the dog will still develop a lukewarm response. If you're not, though, the dog will not make an association at all, meaning that you're doing all that work for nothing.

Your job is to find a way to avoid the trigger when you're not training. This means either carrying chicken and going everywhere with your dog, or not doing certain things with your dog. And yes, this includes walks and going potty. Speaking of which...

Make Sure You Switch to an Operant Technique
Since classical conditioning requires a near 100% trigger-treat response, it simply cannot be considered a long-term strategy. At some point, Kathy said, you just have to switch to an operant training strategy. This might be something like auto-watches or emergency U-turns, or it might be something “fancier” like BAT or Look at That. The technique you choose doesn't matter as much as the fact that you choose an operant behavior for your dog to perform at some point since that can be reinforced intermittently, allowing you the ability to just relax and enjoy your dog sometimes.

As you can see, there are plenty of opportunities for you to undermine the work you're doing. This is just part of the reason I advocate working with an experienced trainer when you're trying to modify your dog's behavior. I know I certainly made plenty of mistakes along the way- I never did reserve the absolute best treat possible solely for counter-conditioning, and so of course, those treats happened all the time. I'm also often guilty of lumping criteria, and I'm sure I screwed up my timing now and then. Still, we muddled through, and while my efforts could have been better had I understood this information then, we did okay. But what about you guys? Any memorable mistakes? I'd love to know that I'm not alone in my training shortcomings...


Anonymous said...

I believe this line is backwards:

"If a scary thing predicts food, then the food becomes associated with fear."

I believe you meant:

If food predicts a scary thing, then the food becomes associated with fear.

Otherwise great article and very useful summary of using classical conditioning.


Crystal Thompson said...

Oh my gosh, THANK YOU NOEL. I don't know how I missed this in my editing process, but you're absolutely correct. I've fixed that (huge, blatant, awful) error! THANK YOU for pointing it out. :)

Ninso said...

Great info, but I think this is tough to pull off in real life. If you can't isolate your dog 100% from the scary thing while training, and if you can't set up situations that you have 100% control over the stimulus on a very frequent basis, I think it's going to be really difficult to get the result you're looking for. At least, that has been my experience with Jun. Set-ups once a week at the most is not enough and incidental contact between those will undermine all your work. Unforuntately, this makes CC not a particularly viable option for my dog and her particular fears.

Another interesting point is that you are supposed to feed regardless of the dog's reaction to the stimuli, however in some cases this can actually reward the undesirable behavior. Interestingly, it still has the effect you want on the dog's emotional state, so you could end up with a dog who is no longer fearful but whose behavior hasn't changed. Or if your dog was not actually fearful to begin with (like Elo) you're just rewarding bad behavior. So this would be a situation where skipping CC or moving to an operant strategy very quickly might be more effective.

Crystal Thompson said...

Ninso, yes, you're right that you can't do this stuff 100%. Kathy was pretty honest about that in the seminar, and it's part of why she said CC just can't be a long term solution. It's just not practical at best and it's impossible at worst.

I've also had the experience of feeding regardless of the dog's reaction (and then later getting that behavior offered, sigh). I want to say this is why it's so vital to keep your dog under threshold while doing CC work, but I also know how hard that can be. Maisy's threshold was often quite unpredictable pre-medication.

What it boils down to, I think, is that the science is pretty clear, but the science was discovered in pretty sterile environments. So, while it's good to understand, it's more difficult to use in the real world. I think that's why we see so many operant counter conditioning strategies (like Look at That and BAT) that seek to both change the dog's emotions AND teach the dog what to do instead.

Ninso said...

Thanks! That makes a lot of sense. And good to know it's not just me that has a hard time with it!

Anonymous said...

Crystal, I have just discovered your wonderful Blog, it fantastic, like going to all those seminars that i would like to have been able to go on.
I also have big problems with avoiding the scary triggers. sometimes my dog is freaking out on a daily basis. I live in the country in the south of spain and i'm on a regular cycle route where people often pass by on bikes completly out of the blue and my dog doesn't cope. so even if i am trying to cc in other training sessions it just isn't having any affect because the daily scenarios are hard to control : (
it's very helpful to read your notes from kathy's seminar. makes a lot of sense and i feel encouraged too that i'm not the only one having problems with cc : )

Msminnamouse said...

What you described as trace conditioning is actually forward conditioning.

Trace conditioning is presenting the stimuli first and then giving the treat US while the CS is still present.

For example: CS is present for 5 seconds and the US is presented during the last few second of this and a little bit after that.