"We further attempted to experiment with novel weights not used by other pollsters, such as weighing on social trust, socioeconomic attitudes, and religious attitudes as measured by the General Social Survey, correcting for polling's tendency to oversample atheists and people with high social trust. We also weighed by Facebook usage, correcting for polls tending to oversample online voters."
Not sure if this is genius or stupid. We'll see come January
They might be right that polls are incorrectly oversampling such voters (I have never seen a regular political poll even break down religious affiliation to the point of identifying atheists, who are too small a group to make any real difference anyway, but "high social trust" definitely seems relevant), but those topics are just too nebulous to establish meaningful weights for. You can't definitively assert that 38% of voters, or 21% of voters, or 83% of voters, will be "high social trust" voters in the way that you can talk about race or gender or education in relatively definite terms. We know with pretty good certainty what the actual demographics of the population are independent of polling and can make reasonable assumptions about turnout, but we don't actually know how many people, or how many voters, are "high social trust".