President Trump has repeatedly decried voting by mail as rife with “tremendous potential for voter fraud,” encouraging Republicans to “fight very hard” against it. Nevertheless, given the pandemic, many more voters than usual are expected to cast their ballots by mail in the 2020 general election, with those mail-in ballots disproportionately cast by Democrats. In some states, ballots cast on Election Day will be easier to count immediately, while mail-in ballots are tabulated later. If in-person ballots are disproportionately cast by Republicans and mail-in ballots disproportionately cast by Democrats, some states’ results may undergo a “blue shift” — shifting further toward the Democratic candidate over time. Our research finds nine states most likely to undergo a “blue shift,” which in some cases might be quite significant.

Unprecedented numbers of Americans are likely to vote by mail

For the past six months, the COVID-19 Consortium for Understanding the Public’s Policy Preferences Across States has used PureSpectrum to administer monthly or twice-monthly surveys of about 20,000 Americans in all 50 states on their attitudes and behaviors related to the coronavirus pandemic. Respondents were recruited online with demographic quotas for each state; results were reweighted to state and national benchmarks by race, gender, age, education and geography (urban, suburban or rural residence).

In our August and September waves, almost 40 percent of likely voters said they are very likely to vote by mail this year. An additional 19 percent said they are somewhat likely to do so. In 2016, roughly 21 percent of all ballots were cast by mail. Of likely voters who identify as Democrats, 51 percent said they are very likely to vote by mail this year, compared with just 26 percent of Republican likely voters.

Our polling suggests that Democrats are more worried about the pandemic and more likely to want to avoid in-person voting, much as other studies have found.

This could make for a complicated election night

The possibility that a greater proportion of Republicans’ votes will be counted on election night has prompted some observers to worry about a “red mirage,” in which Trump appears to be ahead on election night in enough states to seem to have 270 electoral votes — but after all the ballots are counted, Joe Biden wins.

Combining our August and September surveys, we find Biden leading Trump among likely voters by 10 percentage points, 50 percent to 40 percent. But among voters who say they are very unlikely to vote by mail, Trump beats Biden, 68 percent to 23 percent. If we leave out those who say they’re very likely to vote by mail, Trump wins by 50 percent to 39 percent.

What might that mean for each candidate? If the pandemic surges in critical Republican strongholds over the next few weeks, likely Trump voters might stay home. But for procedural reasons, votes by mail may be less likely to be counted — a particular risk for those who usually vote in person — which might disproportionately hurt Democrats.

Either way, Biden’s margin in any given state will almost certainly improve as more mail-in ballots are counted. That won’t be true in all states; some, like Florida, start processing and counting mailed-in ballots well before Election Day. But 10 states do not begin processing mailed-in ballots before Election Day. Eighteen others (including D.C.) accept mailed-in ballots after Election Day if appropriately postmarked. We estimate that most states could be counting at least 6.5 times as many by-mail votes than in 2016. Given all this, we can expect some delays in definitive results.

What does that mean?

We found nine states where a “blue shift” is potentially consequential: Alaska, Michigan, Nevada, New Hampshire, North Carolina, Ohio, Pennsylvania, Texas and Wisconsin. These nine meet two criteria. First, they are competitive. Second, they don’t process all mailed-in ballots before Election Day, or they accept postmarked ballots after Election Day.

We then used our survey data for each of these states to see the likely election night results — leaving out voters who say they’re somewhat or very likely to vote by mail — and compare that with the results once we do include the likely by-mail voters. That gives us an estimate of the likely “blue shift” in each state, as you can see in the figure below.

For example, in Alaska you can see a 16 percentage point difference in vote margins between likely voters who are unlikely to vote by mail and likely voters as a whole. This means that the difference between Trump's and Biden's vote totals on election night and once all ballots are counted could be as much as 16 percentage points.

That’s the smallest such difference among our nine states. We found the largest in Nevada, where the difference is more than 60 percentage points. In the three key battlegrounds of Michigan, Wisconsin and Pennsylvania — the states that tipped the election in 2016 — these differences are 55, 43 and 36 points, respectively.

For instance, if the shift in Wisconsin is about half that ceiling of 43 points (because they have counted about half of the mailed-in ballots), then Trump could have a 10-point lead on election night — but a roughly 10-point victory by Biden in the days after.

It is important to keep these possible shifts in mind, especially for Pennsylvania and Wisconsin, which do not process any mail-in ballots before Election Day.

To be clear, the actual margins will probably shift less than what our chart shows. Not everyone who says they are likely to vote by mail will do so. For instance, concerns about the reliability of the U.S. Postal Service combined with the president’s repeated challenges to mail-in voting may prompt some who intended to vote by mail to instead opt for early in-person voting. Many mail-in ballots, in turn, will be counted on Election Day. But we expect a big shift.

Major news outlets’ decision desks will be aware of this likely shift, while many viewers will not. Given the tensions surrounding this election, journalists, pundits and voters alike may wish to keep in mind that if a candidate declares victory on election night, that lead may evaporate as more votes are counted.

David Lazer (@davidlazer) is university distinguished professor of political science and computer sciences at Northeastern University, where he co-directs the NULab for Texts, Maps, and Networks, and is also a visiting scholar at the Institute for Quantitative Social Science at Harvard University.

Jonathan Green (@_Jon_Green) is a postdoctoral research fellow at Northeastern University, where he works in the Network Science Institute.

Matthew A. Baum (@matthewabaum) is the Marvin Kalb professor of global communications and a professor of public policy and government at Harvard University’s John F. Kennedy School of Government.

Alexi Quintana Mathé is a PhD student in network science at Northeastern University.

Katherine Ognyanova (@Ognyanova) is an assistant professor at the School of Communication & Information at Rutgers University.

Adina Gitomer (@AdinaGitomer) is a PhD student in network science at Northeastern University.

James Druckman is the Payson S. Wild professor of political science and faculty fellow at the Institute for Policy Research at Northwestern University.

Matthew Simonson (@mattysimonson) is a PhD student in network science at Northeastern University and a dissertation fellow at the Harry Frank Guggenheim Foundation.

Hanyu Chwe (@hanyuchwe) is a PhD student in network science at Northeastern University.

Roy H. Perlis (@royperlis) is a professor of psychiatry at Harvard Medical School.

Jennifer Lin (@jennifer_lin16) is a PhD student in political science at Northwestern University.

Mauricio Santillana (@MauSantillana) is an assistant professor of medicine at Harvard Medical School.