Shaping is when we teach a behavior by succesive approximations, reinforcing behaviors that can lead to our desired behavior and gradually changing what we are reinforcing to closer and closer approximations. A challenge with shaping is choosing appropriate steps to increase criteria and choosing when to increase criteria. When we make a training plan, we are essentially guessing what step we will start with and which steps will come next. The more experience we have with shaping and our dog the better educated those guesses become, but it is still guesswork.
Raising Criteria: The 80% Rule
The guideline for when to increase criteria is when the current step is at 80%. The purpose of this is that you want the current behavior reliable enough that it doesn’t extinquish easily, but not so reliable that it becomes more difficult to change. Makes sense, although 80% of what total number is left unclear. And so the record keeping recommended for shaping is to keep track of how many times of 10 attempts you C&T your current criteria and to then raise criteria when it reaches 8 of 10 (80%).
How does this work in practice? It depends on how well you have guessed each step you will need.
Let’s use shaping a dog to pick up an object as an example. You reach 8 of 10 nose touches of the object, and your next criterion is opening the mouth. You stop C&T’ing closed mouth nose touches and hopefully the dog does some variations of the behavior that will include opening their mouth. The rate of reinforcement plummets. This process relies on the frustration of not being clicked to get ‘more’ or a variation of the behavior (an extinction burst).
One Step Forward
Often this training process becomes one step forward, one step back because the dog isn’t being clicked often enough, the rate of reinforcement falls too low and you have to go back a step in your training plan. Then you take only a half or quarter step forward to get the rate of reinforcement back up high enough that they are willing to keep trying. Frustrating and possibly aversive for the dog (unless you already have experience with that behavior and therefore can make a good educated guess to base it on and avoid this).
Removing the Guesswork
My solution for this difficulty is to keep track of not only the number of times I C&T or not, but what my dog is actually doing and how they are doing it. This gives me more information to base my decisions on. Most importantly, I can note the variations of my current criterion (there are always variations in the behavior) and base my next one on a variation I am getting. I can raise my criteria when I am getting enough of that variation, basing it on data instead of guesses and hopefully avoiding having to take steps back.
Back to my example of shaping my dog to pick up an item. I am at nose touches of the item, and I know they have to open their mouth in order to pick it up, mouth opening is my possible next criterion. I C&T each nose touch, my current criterion. While I am doing that, I will watch for and make note of nose touches that also include some degree of mouth opening (again, my current criterion is nose touch so all nose touches are C&T’d, but I can take data on more than just that).
In the picture above, you see my data on 20 attempts of nose touch and also my observations of nose touches with slight mouth openings. It shows that I have reached 80% on nose touches. It also shows that my observations on slight mouth openings tells me that at most only 3 out of 10 of those were nose touches that were done with a slight mouth opening. If I raise criteria to slight mouth opening, my rate of reinforcement will be pretty low. I decide to stay at nose touches for the next session, even though my nose touches have reached 80%.
My next session, I have more slight mouth openings, 6 of 10. If I change criteria now, a good portion of the behaviors the dog is already doing can still get a C&T and my rate of reinforcement won’t be as low when I stop clicking nose touches that don’t include the slight mouth opening. My percentage of correct nose touches has reached 90%, however it has only been a total of 23 repetitions. 23 reinforced repetitions is not all that reliable of a behavior, and it shouldn’t take too much for them to extinquish.
I hope my system of keeping track of more than just click or not helps others with their shaping plans. If you have a question or a tip of your own, please feel free to leave a comment.