What Part Do I Reward?
Terry's Tips > What Part Do I Reward?
Criteria Selection in Reward-Based Training
When teaching something new to a dog, there are decisions, decisions and more decisions to be made by the trainer. It's helpful to split the behavior into steps, put the steps into a sequence and decide which step or aspect of the behavior will be rewarded during a given training session.
Recently I traveled to the east coast to attend a 15-day course on applied behavior analysis and operant conditioning. It was taught by Bob Bailey, a figurehead in the field of behavior analysis and operant conditioning. Here are some details from that course about criteria selection:
Criteria is the term used for the specific, trainer-defined response wanted. The instant the dog achieves the criteria, that response is rewarded (reinforced). There is usually more than one dimension of a behavior. Let's use the sit as an example. Most owners care about:
Physical Response - DOG PUTS REAR END TO THE FLOOR IN A SIT POSITION.
Latency of the Response - HOW FAST TO SIT WHEN THE CUE IS GIVEN.
Duration of the Response - HOW LONG DOG SITS.
Work with only one criterion at a time. When changing criteria, increase the rate of reinforcement. More rewards will help keep your dog focused and let him know this new twist is a good behavior.
Temporary Criteria - When Perfect is not better!
The dog and trainer should move quickly through the steps of an exercise (the temporary criteria) and not perfect those intermediary steps on the way to the goal. For example, in the beginning of teaching "High Five" there might be 10 or more steps to shape the behavior.
- Reward when dog shifts the weight of one front paw to another.
- Reward shifting weight and bringing the paw up from contact with the floor.
- Reward for a higher paw lift.
- etc.
If the dog is rewarded too many times at one step, he might not be willing to offer anything else. The dog might decide to stay at the behavior that got so many rewards. In a nutshell: Temporary criteria should be trained only to about 80% reliability before adding a step or changing criteria. If a temporary criterion is reinforced for too long, the animal may be reluctant to change its behavior. The rewarded behavior of weight shifting will have to go into extinction, which in most cases means getting stronger before getting weaker, before the trainer can successfully move on.
Permanent Criteria - the Finished Product.
Permanent criteria, your final goal, should be trained to a higher level of reliability than temporary criteria. Your "fluent" dog is not distracted by new environmental features.