the original study is http://ehp03.niehs.nih.gov/article/info%3Adoi%2F10.1289%2Fehp.1002835
HT Russel Roberts at Cafe Hayek
Briefly, “Researchers interviewed and examined 304 children with autism and, as a control, 259 typically developing children in the Los Angeles, San Francisco and Sacramento metropolitan areas. Researchers found that children whose families lived within 1,000 feet from a freeway at birth — about 10% of the children in the study — were twice as likely to have autism as those who lived farther from a highway.”
So let’s do a little math, 10% is about 30, half that (expected from the control) is 15. The standard error on 30 is a little over 5, but there is also an error on the control. Combined error is probably between 6 and 7. They report the confidence interval as an effect size (as a retio) of 1.04-3.45.
I give them credit for giving us enough information to see that the lower edge of the effect size is just barely above 1.0. That is, it is just barely statistically significant. Moreover, the measured effect is about the smallest that could be measured. If I had a stats package handy, I could do a retrospective power analysis, see Andrew Gelman on this subject http://www.stat.columbia.edu/~cook/movabletype/archives/2010/12/the_truth_wears.html but this seems to be an underpowered experiement. The “theory” is really more a stab, without a working model for either the polutants or the expected effect. This is exactly the kind of study likely to measure an effect that wears off over time.