Some economists are causing tons of trouble by using economic-style research design to show autism rates are linked with cable television and the rain. Certainly tv watchers in rainy city washington are going to get less sunlight, so that adds more credence to her argument.
If you sample 100 uncorrelated variables looking for significance at 95% confidence, you should find about 5 which are significant.
Data mining to produce hypotheses is fraught with danger. You should declare your hypothesis and then go to the data, otherwise your confidence measurement is bogus.
Their study:
http://www.johnson.cornell.edu/faculty/profiles/Waldman/AUTI...