The long tail may not hit a target: High school teachers
I’m attending the NSF CE21 Community meeting this Thursday and Friday. I have been asked to lead a session on Friday afternoon on distance education in CS for teachers. I was encouraged to talk about just a couple concrete examples, then leave the session open for discussion. The question is which examples?
Here’s a more specific question that leads to this blog post: Are the on-line Stanford CS classes a good example to talk about? Clearly, they are a highly innovative example of distance education for computer science. But is it relevant for teaching high school teachers for the CS10K effort?
First of all, was the audience for the Stanford CS classes like the audience of potential CS10K teachers? I’m not convinced. First, when I read the comments to posts about the the Stanford classes, or Fred Martin’s post, I’m struck by how many people took the courses who already knew the content. They were curious about the course, or wanted a refresher. I wonder how many of the students who finished were novices to the content, and how many were old-timers? My guess is that the average completer in the Stanford classes was a lot more CS-savvy than a business teacher who had never taken a CS class.
Second, was the method of teaching right for reaching in-service high school teachers? I don’t think that the medium of the Stanford CS classes would work, at least as-is. I read the comments to my post about the effort required in classes like these, and I think about Klara Benda’s study. The people who dropped the course aren’t saying it was too hard. They’re saying it took too much time, the pace was too demanding. I can’t imagine that the technology behind the Stanford classes demands a rapid pace, but it’s clear that the pace was an issue for some of those who dropped out. High school teachers don’t have the spare cycles for that rapid pace — Klara’s study has us realizing that we get small chunks of an in-service teacher’s time in which we can provide learning opportunities.
What I’ve come to realize is that the Stanford classes were successful as a long tail effect. They enrolled a couple hundred thousand students, and some 20% finished. When you look at the big number of finishers, which is way more than probably all other students in all other AI classes in the world combined, it’s really quite remarkable.
On the other hand, 80% didn’t finish, and it may be that the students we most need to succeed for CS10K were in that 80%. A long tail effect can get you large numbers, but perhaps, none of the numbers that you might be targeting. A long tail covers a wide swath of the distribution of people, but those that you hit (who complete the course) are not necessarily randomly distributed. More likely, the course acts as a filter on the long tail and filters everyone who doesn’t meet a particular set of criteria. It may be possible to use a long tail approach and hit the target population you want to reach. But it’s not a sure thing.
I am not claiming that the Stanford AI classes were trying to reach teachers for CS10K. I am looking at that innovative work with a different filter. I’m exploring the question of how well that innovation meets the CS10K goals. As part of my talk preparation, I’m revisiting John Daniel’s book Mega-Universities and Knowledge Media. It’s an older book now (1999), but they report that the UK Open University with its reliance on printed books had over a 50% completion rate on average across their classes. I hope that advanced Internet technologies would lead to even better completion rates.