Okay, everyone calm down (including the Trump supporters). Consider the following from the poll write up:
Recall a previous Upshot feature where
different pollsters were given the same Florida dataset and calculated the margin. The two pollsters weighing on party registration predicted 2-4 points higher votes for Trump relative to the one who didn't, holding likely voter screens equal.
The very good writeup also mentions two possible poll biases toward Trump:
- Oversampling of Republicans who support Trump rather than staying home, since weighing was done on party registration.
- The usual likely voter screen critique. Specifically, the voter screen may have worked against minorities sampled in the poll: if Latino or Black voters don't respond that they're very likely to vote, the Cuban-Americans crowd out other votes.