Below is something I wrote up for a colleague who was trying to figure out some quick and dirty ways to keep data. She was trying to count in her head and train at the same time, and that was not working. This kind of picks up in the middle of a conversation, but hopefully you'll figure it out.
Thought this might spur some more ideas, maybe be helpful.
You could use those carpenters aprons and drop incorrects into a separate pocket. Another was to video whenever possible and get feedback from other trainers re. efficacy of measuring as well as training. Also- if you video you could do all kinds of measuring, just off the video.
Another idea that I thought was pretty cool was to measure out 12 sets of 10 treats, and put each set in one part of a muffin tin. Pull out the ten for your session, and put your incorrects back into the tin.
The following is an example of how to count more easily than actually counting in your head or with fingers, etc while training. Obviously you can use 10 treats rather than 30, set the timer for 5 seconds rather than 20, etc.
Count out 30 treats; set timer for 20 seconds; every time you cue dog and dog responds correctly, feed the dog. Every time you cue the dog and dog responds incorrectly, move a treat over to another pile.
I am working on cue response to sit. I start the timer, and cue the dog. When the dog responds to the cue, I take a treat out of the pile of 30 and feed the dog. When the dog does not respond (within whatever parameter I have set, say two seconds), I move a treat from the original pile of treats to an "incorrect" pile. Let's say that at the end of 20 seconds, I have the following:
18 treats in the original pile 6 treats in the incorrect pile
This gives me the following information:
In 20 seconds, I cued the dog 12 times; the dog was incorrect 6 times, and correct 6 times. 50% success rate.
Since you have time period, number of responses, number of corrects, you can figure: average time per response: 20 seconds divided by 12 responses gives me 1.67 seconds per response average time per correct, 20 seconds divided by 6 corrects, 3.333 seconds per correct response.
They are obviously not "real" numbers in the sense that they are actually telling you time per response. For a real number I would think it would be necessary to video with a time counter and use that to time.
However, I think those numbers can be helpful when looking at efficacy and progress of training.
==================== Generally speaking, when working on speed, the number of corrects vs. incorrects shouldn't even be an issue. It's my understanding that you shouldn't even be working on speed as a separate issue until you've got the behavior where you want it, and it is perfected topographically, etc. (though you should try to build the behavior with speed, if that makes sense). Anyway, figure you've already got about an average 95% success rate.
So when you are measuring speed, you just set a timer for a set amount of time (let's say a minute) and count responses (I do this by counting out a pile of treats ahead of time, so I don't have to count while I'm training). If you want to count corrects and incorrects to make sure you are staying at 95%, you can do so by the same method above. But if your success rate is not around 95%, you shouldn't be working on speed. ===================
A side note about speed: speed of response usually does generalize to other behaviors. So if you have a behavior that is very solid and is not likely to extinct through lack of reinforcement (like a retrieve for a natural retriever), you can work on speed on that one behavior. Determine your current average speed, and only reinforce those response faster than that. After a few sessions, re-test average speed. You should have a new (faster) figure.
Often working on speed will also improve latency, if that is a problem.
Copyright of all posts is the property of the original author. Please obtain permission from the original author before copying, quoting, or forwarding.
List and Site Owner: Melissa Alexander, mca @ clickersolutions.com