Outreach opportunity

To the best of my recall, the shortest Khan Academy video is twelve minutes. Ars Technica ought to multiply their time span by four or five if they want to get good entries.

The best part of the "Manufacturing Consent" movie is when Chomsky says he makes a crappy television guest because the topics he is interested in cannot be compressed into topical television show sound bite length.

I suck at making videos and I hate my spoken voice, but here's an approximate transcript of what I would tell the kids.

Imagine that some terrible illness kills most of the men affected by it, but spares most of the women. Also assume that most affected women take a certain drug, while most men don't. And on top of that, assume that the drug is completely useless: it doesn't affect your chances of survival either way. Men die from the illness more because of some tiny physiological difference between men and women. And women take the drug more just because it's marketed toward women more.

In this situation, if you didn't get the lucky guess of counting men and women separately and instead counted them together, you'd arrive at the conclusion that the drug is pretty damn effective because taking it is very correlated with survival! So, just by splitting people into groups in clever ways, you may make apparent relationships of cause and effect appear and vanish. This is known in statistics as Simpson's paradox. It is often observed in practice, like in the famous Berkeley sex bias case of 1973 when the university as a whole was found to be biased toward admitting men, while every individual department was found to be biased toward admitting women.

Sometimes you may get around the problem by making experimental tests. If you change the cause and it affects the outcome, you can be sure you're looking at a real relationship, not a statistical illusion. But it's hard to imagine an ethical way to test the hypothesis that smoking kills. In such situations, if you can only observe real-world trends, your best bet is to use many "control variables", like counting men and women separately in my original example.

Cognitive bias has a much higher chance of being selected than those other three. Bayes' theorem, love it though I do, just isn't going to be something that grabs people's attention. Defeating death isn't science and neither is invisible purple dragons.

Quite important:

The contest is open only to legal residents of the 50 United States, the District of Columbia and Canada (except the Canadian province of Québec) who are at least 18 years old as of the date of entry.

Though in a pinch you could have a North American LWer submit your entry for you and give you credit and eventual prizes, since the rules seem to allow third-party collaboration.

Defeating death has no business being in that list.

Edit: I agree with lsparrish's caveat

Maybe I could make one but I'd need some memetic assistance. What are high school science classes interested in watching?