@mattblaze The thing I keep bumping up against discussing this online with folks is that they misinterpret the behavior of large populations: they think that a large population should cause every sampling to drift towards the mean, and that's the opposite of how it works. Especially when the sampling is non-random and includes modifying factors like "99% of the folks in this one town worship at the same place."
"You mean to tell me you think it's more likely that everyone in this town decided to vote against a candidate while voting for another candidate down-ticket in the same party, and that's no indication of fraud?"
"You're asking me if I think it possible that happened once in 100,000 polling locations? Um..... Yes?"