RealClearPolitics has a page listing the pre-election polling for the 2020 presidential election. Some highlights, or lowlights: A Quinnipiac University Poll showing Biden leading by 11 percentage points. An Economist/YouGov poll showing Biden winning by 10 percentage points. An NBC News/Wall Street Journal (yes, the Rupert Murdoch-controlled Wall Street Journal) poll showing Biden up by 10 percentage points. Votes (or mail-in ballots, which may not be exactly the same as votes) are still being being counted, but at the moment, the race looks much closer, closer to 3 percentage points. How many people decided not to spend the time or risk catching Covid-19 to vote because they figured it wasn't even going to be close?
These polls cost money. One explanation of why news organizations or universities conduct them is to burnish their reputations, but at this point, the effect may be more to discredit the poll sponsor. There is not a lot of transparency about pollster compensation. There are some efforts, by Nate Silver, for example, to rate polls for quality based at least in part on accuracy in previous contests, but in general the poll sponsors have longstanding relations with polling companies. There's no financial penalty for being wrong, and there's not much financial reward for being right. It's like litigation in non-contingency-fee cases: the lawyers get paid the same amount whether they win or lose. Winning consistently or in a high-profile case might lead to being hired for future work, and losing consistently or in a high-profile case might discourage potential clients from hiring a person for future work, but there are so many variables in the outcome of a case other than lawyer quality that accountability is scarce. The chances are small that Quinnipiac, or NBC/Wall Street Journal, or Economist/YouGov, are going to fire their polling firms after this outing and hire someone else who got the result more accurate. Pollsters may defend themselves by mentioning "margin of sampling error," but that is de-emphasized in the news reports of the polls by the news organizations that sponsor them.
We've had this error at least twice. In 2016 the pre-election polls predicting a big Hillary Clinton victory were also wrong. It's enough to make one wonder why the error seems to consistently be in the Democratic direction. I guess it's possible that Democrats may be so confident of a victory, based on the polls, that they stay home and the effect is to hurt their candidate. But people like to vote for a winner, so I suspect that the discouraging-turnout effect is stronger for the candidate who is trailing in the polls than for the candidate who is leading in the polls. Recognizing that effect and spotlighting it may help Republicans the next time there is a close election: "Hey, if you are thinking of coming out and voting for me on Election Day, please do so despite the fake news media polls showing that I am trailing by a double digit margin. Those polls have nothing, actually, to do with accurately predicting the final result, which is why they rarely do. They are just a voter-suppression tactic by the left. Don't let them work on you by discouraging you from voting."
If Russian intelligence had spent millions to buy Twitter ads or Facebook ads with phony news stories claiming with no factual basis that Trump was winning by double digits, the FBI, congressional Democrats, and tech platform moderators would be denouncing it as a foreign plot to manipulate the outcome of the election.