The 2016 issue was that those polls were __correlated__. They were NOT independent polls, but somehow all the polls made the same mistake across the board.
> Assumption No. 4: State outcomes are highly correlated with one another, so polling errors in one state are likely to be replicated in other, similar states.
> Basically, this means that you shouldn’t count on states to behave independently of one another, especially if they’re demographically similar. If Clinton loses Pennsylvania despite having a big lead in the polls there, for instance, she might also have problems in Michigan, North Carolina and other swing states. What seems like an impregnable firewall in the Electoral College may begin to collapse.
--------
Note that this post was written well before the results of 2016 were known. This was Nate Silver challenging the poll's assumptions back then. It was clear that "correlation" needs to be accounted for in people's models. That's the #1 issue regards to modern polls.
Lo and behold: Pennsylvania swung right towards Trump, and all the polls that were predicting a Pennsylvania win for Clinton were wrong. But not only that, it proved a bunch of other polls wrong in other states.
Such is the nature of Bayesian statistics: P(A, B) = P(A given B) * P(B).
P(A given B) should equal P(A) in an ideal world. But we live in a messy world with correlations: understanding B will change our understanding of other probabilities.
sure. but their post analysis says that national polling in 2016 was CLOSER than 2012. and state polls were even closer - but there weren't as many as this cycle
Right now there are a tons of Florida polls and they are basically all within the margin right now.