Clicker Training : By the Numbers
At a seminar I gave in Wichita, Kansas, one of the attendees was Dr. Ogden Lindsley – brilliant pioneer behaviorist, inventor of “Precision Teaching” and donkey trainer extraordinaire. Og’s influence on clicker training and operant conditioning and training has been significant though now always recognized. When you hear clicker trainers talk about “fluency”, they are unknowingly using Og’s terminology and frame of reference. Og is one of the few behavior analysts who is an expert at actually teaching real animals in real surroundings to perform dependable and intricate behaviors. His jack donkey, Silver Butte Jack, plays basketball, carries a full size miner’s pick in his mouth on cue and fetches the mail from the roadside mailbox – about 200 yards from the house. Other than a little drool and slobber, the mail arrives in better shape that some scent articles I’ve seen.
One of the secret’s to Og’s insight into behavior is his long-time focus on behavior and it’s relationship to time. Rather than just counting how many times a rat presses a lever, Dr. Lindsley counts how many lever presses occur over a recorded period of time. In Precision Teaching, the ratio between behaviors and elapsed time is the important aspect of teaching and learning. By charting this type of data, Og can determine a great deal about how well a student is learning and how well a teaching program operates. While Precision Teaching uses complex charts and rigorous data keeping, clicker trainers can benefit from Og’s “Real World” tool for training – a hand-held tally counter in one hand and five fingers on the other.
During the Wichita seminar, I was demonstrating how to vary reinforcement with my dog, Tuggy. The purpose of the demo was to show the relationship between varying reinforcement and varying behavior. Tuggy was knocking down a target that bounces back up again. Over the course of the demo, he knocked the target down many times, but was reinforced only occasionally. At the end of the presentation, Og described what had happened, based on his use of a hand-held tally counter in one hand to count repetitions, and the fingers of his other hand to count reinforcements. He timed the event by keeping track of the second hand of his watch. The results were interesting.
For the first part of Tuggy’s performance, the rate of response was about 15 per minute. The rate of reinforcement was a ratio of 4:1 – four behaviors for one reinforcement. At the mid-point of the demo, I intentionally withheld reinforcement for longer periods of time. Og knew that Tuggy’s performance shifted to a ratio of more than 30:1, at a rate of 26 per minute. When the reinforcement was stopped entirely, Tuggy did 50 repetitions in 94 seconds. Now that I have that information, I have objective data about Tuggy’s performance that I can refer to over a period of time. If I suspect his behavior is falling off, I can easily test him and see if it’s my imagination or if something is really affecting his behavior. Monitoring current levels of performance is not the only way to use behavior/time counting. Using simple statistics while teaching new behaviors can come in handy, too.
Recently I was working with a 12-week-old Husky puppy on “down from a distance”. The previous session, the owner had asked me how long a training session should go. I had answered him with a general statement that I regularly train puppies for 30 minutes at a time, but it can be as low as a couple of minutes, or no session at all if the pup is sleepy, distracted, tired or hot. The following week the owner gave a progress report based on comparing the pup’s performance against the 30 minute capacity. He had spent the week worrying that his pup’s performance was less than average. Every time the puppy worked for 30 minutes the owner was happy, but if he quit after ten minutes, the owner classified the session as a failure. He was so focused on only the duration of the session that he really didn’t know how well his puppy was doing. The solution was to use two criteria for judging performance – how long the puppy worked and how many behaviors he performed during that time.
During the following week’s session, we continued to work on “down from a distance.” The pup was still under the impression that had to be at “front” when he dropped, in order to be reinforced. As we worked I noticed the time and started triggering the tally counter each time the owner clicked and treated. After 15 minutes I looked at the counter – 62 repetitions, or roughly four per minute. At the 30 minute mark, the counter had added an additional 43 repetitions – about 2.8 reps per minute. At 35 minutes the rate of response dropped to zero – the pup was more interested in a butterfly and then flopped down in the shade. Now we had something to work with.
After looking at the records, it was clear that just focusing on “how long” the puppy was willing to work wasn’t appropriate. The first 15 minutes of the session were 150% more productive than the final 15 minutes. The owner started to realize that holding his pup to just one criterion was foolish. For the next few sessions, the owner used the tally counter and tabulated the number of repetitions, every 15 minutes. During the following week, the difference in the dog’s performance was observable. The length of time the pup offered high rates of response was increasing to more than 20 minutes – even though the overall length still bounced around 30 minutes. The high rate of responding allowed the pup to gulp information at a faster rate than before. If the behavior was physically fatiguing, the pup’s rate of response stayed high, but the length of the sessions shortened to about 20 minutes. If the owner had been tracking merely the length of each session he would have remained disappointed, even though his pup was acquiring knowledge at an increasing rate.
In the world of experimental psychology, behavior is often reduced to an absurdly narrow focus. Rats and pigeons are studied with great scrutiny as they press levers and peck keys, never learning anything new and never performing outside their electronically controlled chambers. While trainers may not be able to directly use the charts, graphs and data produced by these studies, we can all benefit from borrowing a bit of their methodology. The next time you really want to know if clicker training counts, get yourself a tally counter
and count what you train.
| Back to Clicker Project Index