Cereal correlation
56% of women who consumed the most calories before conception gave birth to boys, compared with 45% of those who consumed the least. Of 132 individual foods tracked, breakfast cereal was the most significantly linked with baby boys.
That’s from a study by researchers at Exeter and Oxford, as per Melinda Beck’s write-up in the Wall Street Journal.
This made me wonder. After all, my brothers and I turned out to be, well, boys, and my mother never ever ate breakfast cereal. However, as a social scientist I recognize that correlational results admit of exceptions.
They also, as Beck notes, occasion doubt. One of the first things one is supposed to learn about “statistical significance” is that some results that appear to be real, aren’t. If you’re operating at the .05 significance level, then in the long run somewhere around 5% of the relationships that you accept as non-random should really be random. Those are “Type 1” errors — or are they “Type 2” errors? I could never keep those straight. (Andrew, though not not a frequentist, presumably can straighten me out here.) Anyway, these breakfast cereal results apparently have produced a minor kerfuffle among epidemiologists and statisticians, which you can read about in the Beck article. If nothing else, this episode provides a nice example for those who are teaching intro methods courses, and it also gives “Monkey Cage” readers an opportunity to admire the excellent pun I devised for the headline of this item. (Please hold your applause.)
Comments
Type I error is basically a false positive, Type II a false negative. So what you described would be a type I error
Posted by: BillCinSD | January 29, 2009 04:26 PM
Applause not held….I like that pun.
Posted by: Matt Jarvis | January 29, 2009 05:19 PM
I see an evolutionary biology explanation. If you can’t deliver good prenatal nutrition, your sons are probably going to be small and possibly have low-IQ, reproductively they are likely to be failures. Small and slow girls are much more likely to have offspring then boys of similar talents. That suggests a selection effect for mothers who’s bodies can use nutrition as an input in determining gender by aborting more male fetuses when times or tough or otherwise interfering with the process.
Posted by: OneEyedMan | January 29, 2009 06:12 PM
BillCinSD: I think the source of my long-standing confusion is that a “false positive” refers to the improper rejection of the null hypothesis. “rejection” and “positive” just don’t load on the same factor in my mind.
Matt: Thank you. I agree.
OneEyedMan: Interesting thought. Let me suggest, mostly lightheartedly, that if it had come out the other way, one could formulate an evolutionary biology explanation for that, too. In this instance, given that the researchers tested the effects of 123 different foods, I’m willing to buy into the Type 1 error interpretation.
Posted by: Lee Sigelman | January 29, 2009 07:54 PM
Damn, do I hate this kind of study. See here for a longer discussion of the statistical issues. The short story: a 7 percentage point effect would be huge, essentially unprecedented.
And, Lee, you’re exactly right that evolutionary biology could explain the opposite “finding” also. (Another point we discuss in our article.)
Posted by: Andrew![[TypeKey Profile Page]](http://www.themonkeycage.org/nav-commenters.gif)
|
January 29, 2009 09:41 PM