www.washingtonpost.com/opinions/2020/11/02/henry-olsen-2020-president-congress-election-predictions/?arc404=true
From the article...
Where my projection could go wrong
No forecast is without the possibility of error, and Trump fans certainly believe I will be wrong. They place their hopes in the idea that the polls are wrong, as they believe the polls were in 2016, and that there are millions of “shy Trump voters” who either lie to pollsters or aren’t answering surveys at all.
We can’t discount that possibility entirely, especially after polls in countries such as Australia and Britain proved to be massively off in recent elections. Nonetheless, we are not likely to see a large enough polling error to rescue Trump’s reelection.
Polling error tends to occur in one of two circumstances: when undecided voters break late in one direction, and when certain demographics are underrepresented in the polling samples. U.S. pollsters certainly underrepresented Whites without a college degree in their 2016 samples, which is why some polls were off. Regional difficulties in contacting certain voter groups lead to consistent and predictable errors, as the Cook Political Report’s Dave Wasserman recently showed. But that is not usually the cause of significant and consequential national polling errors.
The late break of undecided and third-party voters toward Trump, however, was the bigger reason Trump surprised the pundit class in 2016. Most, but not all, analysts failed to account for this possibility. This, more than any other concern, is the biggest reason for significant polling errors on the rare occasions they occur. There’s evidence that a similar factor was at play in the 2019 Australian and 2015 and 2017 British elections, as undecided voters moved sharply in one direction on the eve of the vote.
This is unlikely to be a factor on Tuesday because there are almost no undecided votes left to shift. Polls have shown for months that most voters have decided whom to support and are not persuadable. Those that are persuadable still lean heavily in one direction or another. That’s the major reason the polls have been so stable all year and why they have been uniformly against Trump. Even if every single undecided voter unexpectedly backed Trump, he would still be outside the range of victory in nearly every national poll.
That means that for Trump to win, millions of people who have rarely, if ever, voted would need to turn out and back him. These people are so politically disconnected they likely would not answer polls, and many would never have registered to vote before. That’s what GOP-leaning polling firms such as the Trafalgar Group and Susquehanna contend they have uncovered, which leads them to produce polls that show Trump much closer nationally and in swing states than the more conventional polling averages. There’s some evidence in registration data that this might happen, but we can’t know for sure whether they are correct until Election Day.