fwiw the single biggest difference between the poll aggregation methods is how strong their priors are. 538 has much weaker priors than any other method, which leads to some oddities. The biggest I see offhand is having an almost 1% chance for new jersey to be the tipping point state. I really struggle to come up with a world in which a state that every other poll aggregate other than 538 has as >99% blue to be the tipping point state. (
http://www.nytimes.com/interactive/2016/upshot/presidential-polls-forecast.html?_r=0 if you scroll down on here you can see each aggregator's predictions for each state). You'll notice 538 expects traditionally blue states to be less blue, and (most) traditionally red states to be less red. All of these 80-85%s etc are one of the primary reasons 538 has been reporting the race as way more even. From a statistical standpoint it's pretty unclear whether stronger or weaker priors is better here, because the sample size of elections w/ anything close to modern polling to train models on is relatively small. However its worth noting that a lot of things 538 has done this election cycle have tended a lot towards clickbaity, horse race journalism (look at the existence of their nowcast). The weaker priors seem very likely to be an intentional attempt towards this, given they are so much weaker than every other aggregator, and they result in way more dramatic swings after current events, which nate silver himself has in the past very publicly criticized. I don't remember who I saw this from, but a quote of "I'd love to see 2008 nate silver debate 2016 nate silver" has really stuck w/ me regarding 538's coverage of this election.
tl;dr 538 has different methodology from everyone else, which is why they have way less certainty, and its probably in an attempt at clickbait