Were 2020 election polls wrong? FiveThirtyEight's Nate Silver explains how to interpret polling

As the ballot counting for the presidential election dragged into a third day, the country has watched voting margins narrow in key battleground states.

The shifting returns, which in states like Georgia and Pennsylvania showed President Trump with a strong lead on election night that has since slimmed, have left some wondering -- were the state polls wrong?

As of Thursday morning, ABC News has yet to project a winner in Arizona, Pennsylvania, Nevada, Georgia, North Carolina and Alaska. Both President Trump and former Vice President Joe Biden have remained steadfast in their corners, each claiming that they are on a path to victory.

The disparity between state polling before the election and the election returns has once again raised questions about whether polls delivered an accurate preview of actual voting -- and whether there’s a fundamental shortcoming with polling itself.

FiveThirtyEight editor-in-chief Nate Silver said what his team does is provide “likely outcomes” to help people understand the state election results, but that does not translate into what it means for the actual election.

“[In 2016], Trump was narrowly behind in these states [like Pennsylvania] and he narrowly won. The margins are pretty close. But, you know, the winner was wrong,” Silver told ABC News’ daily news podcast “Start Here.” “This year, it's kind of the opposite, actually. The polls will be right probably in about 48 or 49 states when states like Pennsylvania get counted eventually, but there'll be way off on the margins. Trump will narrowly lose a bunch of states that he was supposed to have clearly lost.”

Silver pointed out that FiveThirtyEight, which is a partner of ABC News, doesn’t conduct its own state polling, but “we try to prepare people to understand the chance that polls might be wrong.”

“All we're doing is taking polls and mapping that to probabilities because ... it's not obvious that if Joe Biden is up five points in Pennsylvania and eight points nationally and seven points in Wisconsin and one point in Florida, how that translates, given all the different permutations that can occur in the real world,” Silver said.

In 2016, FiveThirtyEight caught a lot of flack for forecasting that Hillary Clinton had a 70% chance of winning the White House going into election night and Trump won instead, which Silver said was people taking election poll results out of context.

The same continues to apply to the 2020 election, he added.

“Here's your reality check,” Silver said. “[Biden] basically is going to have the polls be off in the same direction by about as much as they were off in 2016, but because his lead was more robust than he's going to probably ... narrowly win probably a bunch of states that Clinton narrowly lost.”

Trump finished the 2020 election night with a strong lead in Georgia, but that lead has continued to shrink as more mail-in and absentee ballots are counted. Trump had 49.6% of the vote in that state, while Biden has 49.2%, as of Thursday morning.

The story is similar in Pennsylvania, where Trump has maintained a slight lead with 51% to Biden’s 48%, but as more votes are counted, Trump’s lead has narrowed.

Based on current poll results, Silver said Biden is “probably the favorite” to take Pennsylvania.

“President Trump basically has to win all the outstanding states at this point,” Silver said. “Trump needs to win, I think, three key states, which are Arizona, Pennsylvania and Georgia. There are some longer shot parlays involving Nevada or North Carolina, but it's those three, Pennsylvania, Arizona, Georgia, and he has to win all three.”

According to Silver, the shifting vote in Pennsylvania is “exactly what everybody expected.”

“In Pennsylvania, they count ballots that were mailed in last,” he said. “There was a huge split in who voted early versus in-person in Pennsylvania, where Democrats are overwhelmingly more likely to vote-by-mail. Republicans are the more likely to vote … in-person. So if you count the in-person votes first and then you count the mail votes the next day or the day after, then you have this blue shift.”

As was the case in 2016, Silver said people should be prepared for these state polls to have errors this time around -- “three or four points is pretty common,” he said.

“There are still a lot of votes to be counted in the popular vote ... Biden's margin will rise probably to four or five points, which is not as bad as it might seem at first glance in some states like Ohio,” Silver said. “So people should wait till the final numbers are in before they assess exactly how pollsters did.”

This report was featured in the Thursday, Nov. 5, 2020, episode of “Start Here,” ABC News’ daily news podcast.

"Start Here" offers a straightforward look at the day's top stories in 20 minutes. Listen for free every weekday on Apple Podcasts, Google Podcasts, Spotify, the ABC News app or wherever you get your podcasts.