Something the last three (and maybe more) elections have in common...

[Crumpets]

In both most elections there is a discussion in September/October about the discrepancy between registered voters and likely voters in polls. In 2008 and 2012, likely voters, the narrative went, were more likely to vote Republican than registered voters as a whole, and thus the election came down to Obama's ability to drive up turnout among his base. In 2016, the narrative was flipped and Hillary was seen to do better among likely voters.

But in all three elections, the candidate who did worse among likely voters than registered voters as a whole ended up carrying the election. So perhaps the faults in the polls isn't so much one of Democratic or Republican bias, but a likely voter screen that's more tuned for midterm elections than general elections, and polls should continue using registered voters up until election day when it comes to presidential elections.