"Just"? How would you build a predictive model that inferred aggregate individual qualities such as
"% atheists" based on voting results? That would be a rather indirect and distorted path for estimation. There are better ways.
It's not a great way, admittedly, but there is a very high correlation between Republican voters and religiosity. Very high turnout for Republican candidates plus lots of active churches in an economically-poor area I think is a reliable indication that atheism in that area is low.
"Just"? How would you build a predictive model that inferred aggregate individual qualities such as "% atheists" based on voting results? That would be a rather indirect and distorted path for estimation. There are better ways.