So marginalrevolution.com linked to a Mother Jones article on the widely-trumpeted drop in obesity among 2 – 5 year olds. Baffling, because there are no corresponding drops for other age groups, and all the efforts have been targeted at older ages — healthy school lunches, “Let’s Move,” etc. (You’re not going to send a 4 year old to aerobics class, but they’re simply naturally move around all the time anyway.) And a commenter after my own heart, a Bart Torvik, said this:
My wife is an epidemiologist who works with folks that are involved with the underlying study and they are flabbergasted at the media coverage of this. Short version: there is a statistically significant drop, but the confidence intervals around both the current number and the previous number are very large (b/c of small sample sizes in that cohort) so the “43%” figure being reported so breathlessly is essentially meaningless.
And a later commenter, highly_adequate, says this:
Beyond all the other problems, the authors of this study themselves acknowledge that even the p value quoted for this result is highly unreliable because it is does not take into account the fact that there were, in effect, multiple tests, so that the p value should be considerably higher than the .03 quoted. This would likewise of course entail that the confidence intervals be far wider.
If nothing else, an article in which people understand statistics is a great way to start the day!
Ad-Free & Subscriber Log-In