Do vaccines give Somalis autism? Can diabetes give you Alzheimer’s? Does losing make you win? Anyone scanning the science news articles this week would know the answers to these questions.
First, Freakonomics has a discussion of a recent paper showing that NCAA basketball teams are more likely to win if they are 1 point behind at halftime than if they are 1 point ahead. It seems that when people are slightly behind in a game at halftime, they work even harder in the second half relative to people who are way behind, slightly ahead or way ahead.
Second, the New York Times (byline: Donald McNeil Jr.) discusses the abnormally high rate of autism among Somali immigrants in Minneapolis. The article gives several explanatory hypotheses (including a statistical fluke), but a lot of time is spent on the “possibility” that these cases of autism are caused by vaccinations. The fact that the article doesn’t mention that this is simply absurd is glaring (though it does mention “some children” had autistic tendancies before being vaccinated). More interesting is that many of these kids appear to have had seizures, something which is mentioned only in passing.
Finally, Amanda Schaffer at Slate discusses the possible relationship between insulin and Alzheimer’s (Diabetes of the Brain: Is Alzheimer’s disease actually a form of diabetes?).
PDF 
Coglanlab,
I have to agree that the NCAA data provides an interesting factoid, but the Freakonomics blog is not very convincing. Looking at the data points without the bias of the two different fit curves above and below zero points difference, the only conclusion I would be willing to draw is that a one-point difference at halftime has zero predictive value in the final score.
In fact, I would question why the those curve do not meet at the origin, given that the zero-difference outcome has to be 50-50 since the winner and loser were in identical positions at the half, and given that the other points are anti-symmetric about zero.
The fact that +/-1 data points are very close to 50-50 is to be expected. A one-point differential at halftime is hardly significant psychologically. Adding in the error bars, and I doubt there’s any significant difference at all.
What I find more surprising is that a mere 2 point lead at the half seems to be enough to produce a 59-41 chance of winning. A 3-point differential produces a 64-36 chance. But a 4 point lead gives only a 62-38 chance.
From that, I conclude that the error bars in predictions are probably +/-2 or 3 percent, making the 49-51 difference at 1 point indistinguishable from 50-50.
It looks like a statistical fluke to me, which as you note is also the most likely hypothesis for the cluster of autistic Somali kids.
But I must admit that it is fun to hypothesize some deeper psychological truth hidden in the NCAA data.
Go, Pitt! (and Robert Morris, too)
Pittsburgher Fred Bortz
Childrens Science Books
and
Science Book Reviews
Hi Fred,
Without significance testing, it wouldn’t be very convincing! Freakonomics helpfully links to the academic paper, which does include p-values. The difference is statistically significant.
Which is not the same thing as important. A nice way of thinking about effect size here is, if you were to start using this information to bet on NCAA basketball games, how many games would you have to bet on to have a decent chance of making money?
I actually don’t have time to do the math right now because, ironically, I’m studying for a stats exam. So, as many a lazy textbook author has written, I leave this as an exercise for the reader.