You might have heard what happened to when Stanford professor Andrew Ng put his machine learning (a practical kind of statistical modeling) on line. He expected mild interest. One hundred thousand students signed up.
This was enough to excite even the New York Times, which dispatched Tom Friedman to investigate. He called Ng’s success a “breakthrough”, which it certainly is, in its way.
Friedman also puts us on to a new company called Coursera, which offers courses by professors from well known universities, in much the same spirit as Ng offered his. “The universities produce and own the content, and [Coursera is] the platform that hosts and streams it.” Too, many universities already have on-line courses housed on campus.
Several things. Stanford pays Ng’s salary. This money is only partly for teaching, and more so for Ng to publish papers—which contain the material which will eventually be taught. If Stanford didn’t pay him, he would have little time to think of what new to say. Also, Stanford rightly owns the content to Ng’s lectures. Pay for work, etc. Same thing at Coursera, which hires out the professors for a cut of the pie. All well and good.
Some courses are ideal for the web. The closer the course content is to cookbook recipes, the apter. I mean no disrespect. A courses which shows you how to install and run a certain program is nothing but a series of recipes, a well marked path with milestones and a known destination. Basic machine learning fits this scheme. As do the courses offered at Coursera: Algorithms, Calculus, Introduction to Logic, Vaccines, and Securing Digital Democracy (electronic voting schemes). Recipe does not mean easy.
But maybe not statistics. Unless you want a course in classical frequentist thinking, which is cookbook all the way. Coursera has one just like this. Taught, like it is in many places, by a psychologist (who I’m sure is a nice guy).
Don’t see any courses in history, poetry, literature, high-level philosophy, and the like; all classes which are not amendable to multiple choice testing. These are courses which don’t necessarily have an end, or have different possible ends depending on the mixture of students, or which require students do a lot of writing and talking.
On-line, it’s just as easy for a professor to grade one multiple choice or only-one-correct-answer test as it is to grade 100,000 such tests. But only if the class is recipe-based. One professor could not read through 100,000 essays, or listen to as many presentations. In fact, for these kinds of “free form” courses, he may not be able to handle as many students on-line as he could in person, since the material delivered remotely introduces some level of ambiguity and slows down interaction.
Student interaction for recipe courses, which are (and should be) more popular, is greater, because if the recipe says that “at time T, add X cups,” then it’s likely many students will have this information at hand when they are queried at some forum. This is not as likely in the free form class, where the answers are rarely as firm.
The way I run my introductory class is free form. It is also not a regular course in statistics, but Bayesian from the get go, and in the predictive sense (“all parameters are a nuisance”) regular readers will understand. This marks it an oddity. I get away with it because of a certain rare confluence of events. But it would never fly at most universities, where professors at their professions are more conservative than Rick Santorum (“What! Teach Bayesian probability before frequentist? Never!”). Just ask any professor how easy it is to introduce a new course into the system.
I lecture, but not in contiguous blocks of time. I ask the students lots of questions and frequently. The answers they give provide direction for the course. Students come to the board and work out various matters with themselves, guided by me. I have students collect data (which fits a given paradigm) on any subject which interests them. This a wonderful way to maintain interest, but it limits the use of canned examples, the use of which would free up time.
I also have to spend a lot of time walking people through basic computer tasks, especially R. As a final exam, each student presents a talk on their subject as to an audience presumably unaware of statistics (many use data from their workplace, which they later show their bosses, reportedly to good effect). They must describe their interest, the data, the pertinent questions, show pictures, explain how they quantified their uncertainty, and finally detail how they would check all when new data arises.
Could this work on line? I’m skeptical, but intrigued. Places like Phoenix University push thousands of students through their pipes, and not all the classes are recipe-like, so maybe it can be done. All ideas welcomed.
The last difficulty is “credit.” Some courses earn credit as normal classes. But the free courses universities and Coursera offer come with nothing except a pat on the back or perhaps a letter stating that the student made it through. Of course, I love this trend away from formal “certification” and towards actual love of learning. Seems to work best for recipe courses, though, whose students actually want to bake a cake.
The “free” part doesn’t hurt when attracting students: a fine plan for behemoth institutions; wouldn’t work well for little guys like me.
Test of latex. Should be pretty: , \Pr(x|e) = 0.5[\latex]