Polls showed Biden had a substantial lead. Why didn't Americans wake up to a winner?

It's coming down to the upper Midwest — again.

Despite large leads in national polling before Election Day, Democratic nominee Joe Biden is still in a tight race with President Donald Trump on Wednesday morning, with a murky picture of who is winning and when Americans will know.

National polling averages all the way up to Tuesday morning, however, painted a clear picture: Biden was ahead — and ahead with a larger lead than what 2016 Democratic nominee Hillary Clinton had before Election Day.

So the polls are wrong, right? It's not that simple.

Live election results: 2020 presidential election results and electoral map

While Biden may have had a clear advantage in national preelection polling, his lead in the key battleground states that win a candidate the presidency in the Electoral College was a bit slimmer. However, the preelection polls appear to have shown him ahead by slightly bigger margins than are coming in in states like Michigan and Wisconsin.

"Just as in 2016, it seems as though the polls have underestimated support for Donald Trump. Biden was the winner nationally but the state level estimates were statistically biased in underestimating Trump's support," said Michael Traugott, a research professor at University of Michigan's Center for Political Studies.

On the national level, RealClearPolitics, the polling aggregator, showed Biden with an average of a more than 7 point advantage. FiveThirtyEight showed Biden ahead on average by more than 8 points across the U.S.

However, many votes are still being counted in the battleground states due to the influx of early voting during the COVID-19 pandemic, making it unclear how the results compare to polling.

"I think everyone needs to take a deep breath," Joshua Dyck, director for the Center for Public Opinion at UMass Lowell, said early Wednesday.

While there appears to have been a national polling error, Dyck noted, it's premature to characterize polling errors as votes are still being counted.

"We knew all along that absentee votes would be counted last in Michigan, Wisconsin and Pennsylvania and that they skew heavily Democratic," he said.

Traugott said more answers on what happened won't come until there's "a detailed post election review."

Trump attacked the polls Wednesday morning tweeting that, "The 'pollsters' got it completely & historically wrong!" Twitter later flagged the tweet as "disputed and might be misleading."

However, the data are less clear in reality.

Wisconsin showed bigger leads for Biden leading up to Election Day. Before Tuesday, his advantage was above 6.5 points per RealClearPolitics and above 8 by FiveThirtyEight's measure in Wisconsin. As of late Wednesday afternoon, he was holding on by just around 1 percentage point with 99% of the votes in, and news organizations called the race for him. In Michigan, preelection margins were similar but a bit closer. Biden's lead had grown to just over 1 percentage point as of 5 p.m. ET and news organizations had also called the race for him.

Pennsylvania, with its 20 electoral votes, was much closer in preelection polling, however. RealClearPolitics had Biden up just over 1% and FiveThirtyEight had his lead below 5. Yet a clear picture of where the votes fell there will take time. About 86% of the vote had been counted as of 5 p.m., showing Trump up, but many mail-in votes that tend to skew Democrat were still being counted.

Many political analysts eyed Pennsylvania as the key state in 2020.

Nate Silver of FiveThirtyEight explained in an article Monday, with the headline, "I’m Here To Remind You That Trump Can Still Win," that, "without Pennsylvania, Biden does have some paths to victory, but there’s no one alternative state he can feel especially secure about."

One alternative Democrats had eyed was Florida, where Biden held slim leads in the polls. RealClearPolitics's average showed a less than 1-point Biden lead and FiveThirtyEight showed a larger 2.5-point margin. But Trump claimed victory there as Biden's support in Miami-Dade waned compared to 2016.

"I think the Miami-Dade surprise result is likely driving the polling error in Florida, which appears to be about a 5 or 6 point error from the polling average to the result," Dyck said Tuesday night.

Another possibility Democrats eyed comes in Georgia and North Carolina. RealClearPolitics and FiveThirtyEight were divided in terms of who led in an average of the polling there. Voters are still waiting for about 6% of the vote to come in in North Carolina and 4% in Georgia as of 5 p.m. Wednesday, with Trump just ahead in both. The margins between the two candidates in the state-level polling before Election Day in both were razor thin.

The Electoral College, explained: Why and how it can pick a president who got fewer votes

Down ballot, the picture early Wednesday also remained unclear. Polling led to some forecasters giving Democrats high chances of flipping the Senate and making inroads in the House.

"Tuesday was an abject disaster for Democrats in Washington," read POLITICO's Playbook newsletter Wednesday morning, pointing to muddled results in Congressional races.

"While Democrats had multiple avenues to flipping the Senate, virtually all of them appear to have fallen or are set to fall short, with Republicans, in all likelihood, still holding onto the Senate majority," wrote Amy Walter and Jessica Taylor in the Cook Political Report.

There were 35 Senate seats in the election but only about 14 were truly in play. The Cook Political Report rated 12 Republican-held seats as competitive, while just two Democratic-held seats were in that category.

Democrats won two seats held by Republicans: in Colorado and in Arizona. But Republicans held off liberal challengers in Iowa, Montana and South Carolina. And they flipped a Democratic seat in Alabama.

In Iowa, Democrats eyed Republican Sen. Joni Ernst's seat, but the incumbent ultimately held on, projections showed. According to FiveThirtyEight's forecast, the race was a toss-up, and RealClearPolitics' polling average gave her around a 1.5-point lead. An Emerson College poll Sunday had Democratic challenger Theresa Greenfield up 3 points while a Selzer & Co. poll Saturday pegged Ernst as a 4-point favorite.

In Maine, Democratic challenger Sara Gideon conceded to Republican incumbent Sen. Susan Collins by midday. FiveThirtyEight also called the race a toss-up, but all preelection polling showed Gideon ahead.

In the House, Republicans unseated freshmen incumbents in South Florida, New Mexico, Oklahoma and South Carolina. Vulnerable seats in Texas were also successfully defending, and the GOP claimed its biggest prize by knocking off 15-term Democratic Rep. Collin Peterson of Minnesota.

Some pundits and social media users were quick to decry all polls as inaccurate late Tuesday into Wednesday morning while vote counts in close states were still coming in.

Polling ahead of 2016 showed a similar scenario. Clinton was ahead in the national polls and was in a much tighter race on the state level with Trump. But Trump ultimately took the Electoral College.

Many pollsters point to errors in state level polling in Wisconsin, Michigan and Pennsylvania as an explanation for the 2016 results. In reality, the national polls were not off. Clinton's lead before Election Day nationally was more than 3 points, according to RealClearPolitics' average, and she won the popular vote but just over 2%.

But a report from the American Association for Public Opinion Research came to the conclusion that state-level polling in 2016 "clearly underestimated Trump’s support in the Upper Midwest."

Can Electoral College really tie at 269 to 269? Yes. It's not likely but here's how it works.

Many pollsters and forecasters point to the quality of the data available and the level of high-quality polling data in these states as to why the results were off.

"The people who conducted the polling hadn't done enough research and just didn't have the track record," said David Paleologos, director of the Suffolk University Political Research Center, which partners with USA TODAY on polling.

One reason for the Trump underestimation, according to the American Association for Public Opinion Research: In their methodologies, pollsters did not adjust for the fact that their polls over-represented college graduates. Because voters with higher education were more likely to support Clinton and those voters were over-represented in some of the polling, it translated to an over-estimation of support for Clinton.

In some cases, the methodologies in these polls have bee adjusted and there were simply more high-quality polls being done in these states before Election Day.

"It's not in the polling community's interest to have back to back set-backs in the 'Blue Wall' states," Paleologos added.

Contributing: Lita Nadebah Beck, Christal Hayes, Joey Garrison, Deirdre Shesgreen, Ledyard King and Rebecca Morin

This article originally appeared on USA TODAY: Polls showed Biden over Trump. There's still no winner. Here's why