I've kept a spreadsheet to compare the last polling estimates from fivethirtyeight.com with the actual election results. Although there may be 8 or 9 million vo

I've kept a spreadsheet to compare the last polling estimates from fivethirtyeight.com with the actual election results. Although there may be 8 or 9 million votes left to count, the percentages are probably good enough to make some comparisons. What I saw was an average polling error of 4.9 percentage points. The actual results for Trump were 4.9 points higher than the polls. And the errors were greater in states that had higher percentages of votes for Trump. For example, North Dakota: The last polls showed 37.9% for Biden and 59.6% for Trump, a 21.7% difference. The latest popular vote tally from CNN shows 31.8% to 65.1%, which are 33.3% apart. That's an 11.6% error, much larger than expected sampling errors of 3-4%. Shor's explanation was interesting. Whether the explanation is right or not, I can see the effect Shor is describing.