Yes, but they completely missed the point of the data. Popular vote is relatively easy to predict, but Trump's campaign reached really deep into quiet conservatives who wouldn't vote otherwise.

Those people are still not going to respond to polls, and since they live in low-population areas, they won't affect the popular vote much, but they hit the electoral college hard.

The Dem voters in the primaries are a predictable population, and polling for primaries is relatively easier.

This is something I've seen very few people actually remark on or take into consideration: the people who wouldn't have come to the election polls in the first race certainly won't be responding to the straw polls being conducted now and the models can't really reflect that. Or at least they don't currently seem to.

They have roughly accurately predicted other elections in the middle of which there have been literally hundreds.

But, as said upthread, they predicted downballot elections in 2016 pretty well too. So unless there's been an improvement in downballot predictions - not just good absolute performance - why should we expect any difference in 2020?

>but Trump's campaign reached really deep into quiet conservatives who wouldn't vote otherwise.

There's really not any evidence of this. In the rust belt states he won that turned the tide of the election, there was less turnout on both sides than previous elections. It's just that Democrat voting turnout was down even more than Republican voting turnout.

You know the polls also accurately predicted the gubernatorial elections (and other statewide elections) in Wisconsin (D+1), Pennsylvania (D+19), and Michigan (D+8.5) right?