The ability to set and evaluate criteria is one of the foundation skills every trainer must have.
What is criteria?
A final, finished behavior is often very complex. Not only may be the behavior itself be quite complicated, but there are additional elements such as duration, distance, latency, generalization to different locations, and proofing against a variety of distractions. Both the behavior and each of these elements must be broken down and taught bit by bit.
When you are training, you have to know exactly what bit you are looking for in each training session. This bit is criteria (singular: criterion). Your criterion is defined by you at the beginning of each session. It must be specific and consistent throughout the session, which means:
At the end of the session, you can review your data and determine whether youre ready to move ahead.
Some examples of session criteria include:
Choosing Your Criteria
As I mentioned above, your specific criteria determines when you click. When you first begin working on sit, you click the sit itself the moment the dogs rear end touches the floor. Soon hes sitting regularly, but he gets up as soon as you click. So you want to add duration to the behavior. Now you click a second after the sit. Then two seconds after the sit. After a few days, hes holding his sit quite reliably, except when people walk by. So you change when you click again, clicking at the moment a person, who is getting increasingly closer and closer, passes.
Setting criteria is one of the most basic and most difficult skills a clicker trainer must learn. Improperly-defined criteria are frequently the root of training problems. Defining your criterion is easiest when the behavior is black and white. For example, touching a target. Either the dog touches it or he doesnt, and its easy to figure out exactly when to click when the dogs paw (or nose) touches the stick.
Unfortunately, not all criteria are that clear. For example, youre training your dog to hit a bell to let you know when he wants to go outside. Hes targeting the bell, but he needs to hit it harder. Harder is not black and white. A clearer criterion might be hard enough to make the bell move one inch to the side.
In the beginning, make your job as easy as possible stick to clear, discrete criteria as much as possible.
Making It Harder -- Evaluating Your Criteria
If you stay at one criterion too long, it can be difficult to move past it, because your dog has such a strong reinforcement history for that response. However, if you increase your criteria too quickly, your dog may become confused, and the behavior may fall apart. Fortunately, if you keep simple records, its easy to tell when to increase your criteria. The data will tell you!
Step #1: Set your criteria. It must be specific.
The more specific you are, the better your records, and the easier it will be to identify problems or holes in your training before they become serious. When setting criteria, think "bigger" than just the behavior, even in the beginning. Consider the dog's position to you, your body position, the cue or lack of cue, duration, distance, location, and distractions.
Also, don't forget rate of emitted behavior. There's a world of difference between getting 10 touches of a target in 30 seconds and getting 10 touches in three minutes. To judge rate, estimate how long it (ideally) should take to perform the behavior and consume the reinforcement. Multiply that times the number of reps in a session. Then time the session. If the session takes longer than that, your dog may not be as fluent at the current level as you want -- even if he performs the behavior correctly repeatedly.
When you first set (or increase) your criteria, rate is probably unimportant. It's not fair, for example, to penalize your dog for not lying down within three seconds if he hasn't yet figured out that lying down is the behavior that will earn him clicks. But before you make it harder -- before you begin working on duration, for example -- have your dog achieve a goal that includes rate, such as "10 downs in one minute."
Step #2: Do ten repetitions of the behavior. This is one training session.
Ten is an arbitrary number. I like it because it's easy to figure percentages. I have, on occasion used numbers fewer than ten in a single session. I have also done more than one session at one time. Sticking to multiple of ten makes recordkeeping easy, and even if you do more than one session at a time, it's easy to stop and jot a note and then return to training.
During the session click if, and only if, your dog achieves your criteria. If he doesnt meet your criteria, count the repetition as an error. If your dog makes two errors in a row, STOP and rethink your criteria. Don't frustrate yourself or your dog.
Step #3: At the end of the session, count the number of errors.
You can use the error percentage as one factor in determining whether you're ready to "make it harder."
Rather than strict error percentage, some trainers like to have 10 error-free reps in a row before increasing criteria. Susan Garrett requires 10 correct reps for *every* error the dog makes. Figuring a time-based goal, based on rate of emitted behavior, is another aspect of readiness to increase.
When you raise criteria, it makes sense for the number of correct responses to drop. However, the dog should still be successful the majority of the time. If the dog makes two errors in a row, stop and evaluate what you're doing.
The dog should not just be successful. It should be successful quickly and often. This is rate of reinforcement. If your rate of reinforcement drops too low, either because the dog isnt successful or because you fail to reinforce correct behaviors, your dog may become frustrated or quit.
Set your criteria at an achievable level, and dont increase until your dog is reliable at the current level. Keeping records and strictly following a system a like this -- using data instead of intuition -- enables you to make your training extremely efficient. You wont increase too quickly, building a weak behavior and confusing your dog, nor too slowly, making it hard to move past a heavily-reinforced temporary criteria.
Once youve increased your criteria, dont back up (repeat previously accomplished criteria in the same situation). Your dog would love not to work so hard for his clicks. At the beginning of a training period, he may indeed offer a lesser version of the behavior. If you accept that behavior, he will quickly train you to make it easier for him. Stick to your guns, and trust your data.
Note that I specifically said not to repeat the same criteria in the same situation. When you add a new distraction, a new location, or other new element, relax your standards for the previous elements a bit until the dog gets comfortable with the new requirements. Gradually, the dog will be able to put all of the elements together into a smooth, precise performance.
Leaps of Learning
A "leap of learning" is a very common occurrence. Youre carefully progressing minute step by minute step. Then suddenly the dog jumps ahead and performs well-beyond your expectations. For example, youve been shaping a spin. You started with a glance to the right. Then a head turn. Then a step with the front foot. Suddenly the dog turns a third of the way around. Then he does it again! Youre sure he has this figured out.
What do you do? Do you take advantage of this leap of learning and advance to the new level of competency?
No. Very often the offering doesnt reveal true understanding of what you want. Trainers who jump ahead like this often find the behavior falls apart later.
Stick to your current criteria throughout this session. If, indeed, the dog is consistently performing beyond whats expected, your percentage of correct responses will be high enough to justify moving ahead. Increase your criteria in the same small increment you would use even if the dog hadnt performed beyond your expectations. Again, if the dog has truly made a leap of learning, hell fly through these repetitions with a high percentage of correct responses. If not, you havent risked confusing your dog by moving too fast.
List and Site Owner: Melissa Alexander, mca @ clickersolutions.com