> just look at voting results
"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.