Many 2016 polls may prove incorrect because so many Americans are undecided or else say they’re backing third parties, says Nate Silver, founder of the fivethirtyeight.com polling firm.

“Undecideds are MUCH higher than normal,” Silver said in a Tuesday morning tweet. “So risk of a polling error — in either direction — is higher than usual.”

One out of eight voters, or 12.5 percent, are either undecided or declaring support for other candidates during the last few days of the election.

That is up from one out of 32 voters in 2012, one out of 27 voters in 2004 and one out of 10 voters in 2000, when Texas Gov. George W. Bush barely beat Vice-President Al Gore.

So Silver is cautioning people to be ready for a major failure by the pollsters in predicting Americans’ last-minute choices. Clinton is likely to win, said Silver, but:

3) Basically, these 3 cases are equally likely

a—Solid Clinton win

b—Epic Clinton blowout

c—Close call, Trump *probably* wins Elect. Coll.

If the polls understate Trump’s support by three percent by a uniform margin of three percent, Clinton still wins, although just barely, Silver said.

However, in a prior tweet, Silver noted that Trump is doing much better than Gov. Mitt Romney in the midwestern states, where he may be able to pick up wins that would give him 270 Electoral votes.

Other evidence, described by Breitbart here and here, suggest that the 2016 polls are less reliable predictors than prior years’ of the final presidential vote. Also, African-American support for Clinton may be sharply down from support for President Barack Obama in 2016.

Silver’s warnings about the stability of the poll has sparked much criticism from pro-Clinton commentators, including the writers at the The Onion satire website.