The methodology they describe seems like it would be prone to the multiple comparisons problem. Time to dredge up the actual study and see what's going on.
> Seven clusters, all in rural areas, had a low risk of an association between autism and family lineage.
> “We’re really not sure why some rural areas seemed to have what might be called a protective effect,” Richards-Steed says. “It’s certainly possible that parents and grandparents living in urban areas had different environmental exposures or experiences.”
> “What we can say, based on our findings, is what we are being exposed to now is probably not just affecting us or even our children but maybe even our children’s children.”
> “Evidence shows our environment has a deterministic effect on our growth and development, which includes the germline cells we carry for the next generation,” VanDerslice says.
Some of this sounds like low stats fluctuations to me. In a town of 100, a single family of people has an outsized impact on the apparent prevalence. Cities have enough population to accurately measure it, while towns will fluctuate wildly about the mean. Small towns should have both the strongest and weakes associations if stats dominated, which sounds like it's the case.
