America is Headed for Record Turnout. Here’s What That Looks Like.

U.S. general election turnout is likely to break through 150 million votes for the first time in history. Based on our turnout modeling for the general election, we at Echelon Insights think it will be 157 million.

Or, 157,408,480 to be precise.

If this proves correct, it would represent an increase of 18.6 million votes over 2016 and a voting eligible population (VEP) turnout rate of 65.8%, topping the modern high of 62.2% set in 2008.

Two years ago, midterm turnout smashed through every prediction with 118 million votes cast, an increase of 25 million over 2014 and the highest midterm turnout rate since 1918. Because past voting behavior tends to predict future behavior, we think 2018’s election — conducted in a similar political climate as today — is a good indicator that turnout will be very high in 2020.

To predict higher turnout, you can look at long lines for early voting, the increased use of absentee voting in a pandemic, high primary turnout, and polling showing extremely high motivation to vote compared to years past. Anecdotal data may well indicate historic turnout, but none of this is included explicitly in the model. Rather, voters’ past behavior — specifically their turnout in recent elections — is what’s driving our model to predict higher turnout this year.

To arrive at our estimate, we generated turnout probabilities for every registered voter in the country — and also added in an adjustment for those likely to register between now and Election Day. The sum of these probabilities is our turnout forecast.

The Demographics of the 2020 Electorate

This year’s electorate will also be a more diverse, with nonwhites making up a share of the vote 2 points greater than 2016. That’s good news for Joe Biden, but it’s balanced by another piece of good news for Donald Trump: there should be more members of his demographic base — White voters without college degrees — than most pollsters think, even after their post-2016 adjustments to weight on college education.

Our model expects 67,471,364 white voters without college degrees nationally will cast ballots, 42.9% of the total vote, and votes from 44,991,861 white voters with degrees, or 28.5%.

Black voter turnout will reach 19,554,123, or 12.4%, while Hispanic turnout will rise to 14,778,867, or 9.4%. Asian Americans and Pacific Islanders will make up 4.0% of the total vote with 6,356,092.

More than 6 in 10 voters won’t have college degrees. We expect votes from 98,247,420 Americans without a four-year college degree (or 62.4%), and 59,141,059 votes from those with degrees (or 37.6%).

Our full state by state breakdown, with raw totals and percentages, is in this table below. Collectively, the swing states will again feature a more Trump-friendly demographic mix, with a 45.3–27.4 split between white voters without degrees and those with, an 18 point gap. In non-swing states, the split is 41.2–29.4, a 12 point gap. This is why, if the race were within 2 or 3 points, we’d likely see a split between the Electoral College and the popular vote like we did in 2016.

Demographic Highlights

• Among swing states, Nevada will have the highest rate of voters without college degrees, at 71.2%, followed by Iowa at 69.1%, Michigan at 69.0%, and Wisconsin at 67.7%. The lowest rates among swing states are New Hampshire at 60.1% and Minnesota at 60.3%. Among all states (plus D.C.), the rate of voters without degrees will range from 78.0% in West Virginia and 77.3% in Mississippi to 37.3% in D.C., 51.7% in Massachusetts, and 52.6% in Colorado.
• In the swing states, Iowa will have the highest rate of White Non-College voters — a Trump-friendly demographic — at 62.5%, followed by Wisconsin at 59.2%. White voters without degrees are less prevalent in Texas (34.4%), Georgia (36.4%), and Florida (39.6%) — but this is mostly because of these states’ more diverse populations. In raw numbers, Florida’s 4,409,754 total of White Non-College voters is within 150,000 of the country’s highest number, in California — even as California will cast more than 50% more ballots.
• Comparing the gap between White Non-College and White College voters can be instructive as polling has found this to be a key political divide in the Trump era (and there is no similar educational divide among nonwhites). In October polls, Trump leads White Non-College voters by 19 points while Biden leads among White College voters by 21 points. In every swing state, White Non-College voters are more numerous than White College voters, but there are considerable differences in margin across the Sunbelt battlegrounds that explain their diverging political paths. White Non-College voters outnumber White College voters by 25 points in Nevada, 17.8 points in Arizona, 15.7 points in North Carolina, 15.1 points in Florida, 11.7 points in Georgia, and 8.8 points in Texas. Having more voters in Biden-friendly demographics can help explain why Texas and Georgia have so quickly moved towards battleground status while the opposite being true of Nevada explains why it’s trended more Republican relative to the rest of the country.
• The swing states have a wide range in terms of the diversity of their electorates. The Black vote will range from 31.9% in Georgia to 1.3% in New Hampshire. The Hispanic vote is concentrated in four states: Texas (20.3%), Arizona (17.8%), Florida (16.4%), and Nevada (14.9%); in no other state are they more than 4% of the electorate. Nevada has the highest rate of Asian voters in the battlegrounds, at 6.6%.

The Race for Congress

While the Presidential battlegrounds tend to be less college educated than the country as a whole, the House battlegrounds are primarily in the suburbs and are more educated. In conjunction with our state and national estimates, we built estimates for turnout and voter demographics in all 435 House districts. Below are estimates for the 20 most competitive districts, as measured by the most “Tossup” ratings from House race handicappers.

While the country as a whole has a gap of 14 points between White Non-College and White College voters, and the Presidential battlegrounds have a gap of 18 points, the gap in the most contested House races is just 9 points — 40.2% to 31.2%.

Democrats are on offense in a set of Republican-held suburban seats that are very college educated. White College voters outnumber White Non-College in five of the 20 districts: Missouri’s 2nd district in the St. Louis suburbs, Georgia’s 7th district in the Atlanta suburbs, Nebraska’s 2nd district in and around Omaha, Texas’s 22nd district outside of Houston, and Indiana’s 5th district in the Indianapolis suburbs. These are the kinds of seats Democrats picked up in the 2018 midterms.

Also in this general category is Arizona’s 6th district in the Phoenix area, though White Non-College voters slightly outnumber White College voters there.

Republicans are going on offense in a handful of districts where White Non-College voters are the dominant share of all voters. These include Minnesota’s 7th district and New York’s 22nd district. Trump won these districts in landslides, though they have Democratic House members.

Another set of districts feature large minority populations, including heavily Hispanic seats like California’s 21st and 25th districts (the latter which the GOP is defending having won it back in a special election this year), and Florida’s heavily Cuban-American 26th district. Black voters make up the highest share in Georgia’s 7th district, as 26.6% of the expected electorate.

How We Built This

Understanding the right demographic mix of voters matters, for polling and otherwise. Not getting it right led to many of the state polling failures we saw in 2016.

But getting the right numbers is harder than it seems, especially in localized areas. The Census publishes statistics on the adult population in Congressional districts, but not registered voters, so understanding the right demographic mix in these areas can be harder. Going into this election, we revamped how we built these estimates to compensate for the shortcomings in Census data.

The voting and registration supplement to the U.S. Census Bureau’s Current Population Survey (CPS) is the most widely used source for data on the demographic makeup of the electorate. Every two years, it asks about voting in the election and voter registration. Its national sample (143,050 in 2018) is very robust. It’s historically been much better at estimating demographics than exit polls, which until 2018 dramatically overstated the share of college educated and young voters.

But the CPS isn’t perfect. It overestimates voter turnout, with its respondents more likely to report having voted. Because of this bias and the order in which it asks questions — voting in the election first, and voter registration status second to those who didn’t vote — its universe of registered voters who didn’t vote is much smaller than it is in real life. Other studies have also shown that those who over-report voting also report having a college degree, leading to a universe of reported voters that’s too college educated. And while the CPS’s sample is quite large, state samples sizes tend to look more like election surveys, with high error bars within subgroups. It does not break out data by Congressional districts.

To arrive at our state and district estimates, we start with a much larger government survey — the Census’s American Community Survey — based on 11 million responses over a rolling 5-year period.

Starting with the ACS, we:

• Get estimates for the racial and educational characteristics of the adult population within a given area, be that a state, Congressional District, county, or Census tract.
• From there, we estimate what these numbers are for the Citizen Voting Age Population (CVAP) — essentially, citizen adults.
• We adjust these estimates to reflect the non-institutionalized population, excluding those in prisons or other institutions who cannot vote.
• Since the most recent ACS 5 year release is 2014–18, reflecting on average what was happening in 2016, we adjust for population changes within specific groups all the way up to 2020.
• Then, to develop our estimates of registered voters, we take the group voter registration rate at the state level from the 2018 CPS and apply it. In contrast to its estimate of voters who cast a ballot, the CPS’s estimate of voter registration and the age breakdown of voters lines up pretty well with state voter files.
• Because we don’t use the vote question from the CPS to calculate our likely electorate, our last step is to model the population of those likely to vote in the 2020 election using individual-level variables contained in state voter files. We train a model on the registered voter population (where voter file data are independent variables and Census-derived estimates are the dependent variable), and predict what the Census would likely say about likely voters in our turnout universe.

Our demographic estimates for the 2020 electorate land in between those of two respected groups of demographers. We come in higher on whites without a college degree than the States of Change project; we estimate a 42.9% to 28.5% gap among whites by education while they show a 41.8% to 29.8% divide for 2020. (States of Change more directly relies on the CPS, which is also what most mainstream pollsters use.) We’ll likely come in lower than Catalist, a Democratic data warehouse, which in 2016 showed a 48–27 split in these two groups, whereas our method would have shown 46–28.

How Might This Be Wrong?

We’re putting this turnout forecast out there with the full knowledge that we might be wrong.

Our voter turnout forecast is on the high end of expectations. A critic might argue that it’s harder to expand the electorate in a Presidential election year than it is in a midterm year, since it’s harder to convince people without any track record of casting ballots to vote.

Still, there are many reasons why our baseline expectations should be higher than 2016’s 138.8 million. One is simple population growth, which would add about 5 million votes without any increase in the turnout rate, for a total of around 144 million. Simply matching 2008’s turnout rate would get us to 149 million votes. The electorate is also getting older on average, meaning there are more voters in groups with high turnout and fewer voters in groups with low turnout, so turnout rates should drift upwards all else being equal.

While political science tells us that extensive mail balloting should increase turnout, there are other reasons why the pandemic could reduce it, including mail ballots getting rejected at higher rates or dampened in-person voter registration.

Still, with most analysts missing the turnout surge we saw in 2018 by a lot, we should take seriously the possibility of records being shattered again in 2020. If you’re involved in a campaign, it’s certainly better to be prepared for a historically high turnout than bet on lower turnout and be surprised.

If we’re wrong on the demographic shares of the vote in 2020, we think it’s more likely that we have too many college-educated voters than too few, though any difference here is likely to be small. This goes back to likely vote over-reporting issues with the CPS. There’s an almost implausible gap in the share of unregistered citizens by education — just 6% for White College voters and 21% for White Non-College voters. While there’s likely to be some gap, White Non-College voters tend to be older, and age tends to cancel out any reduced participation based on education (at least when looking at turnout as a share of registered voters). So, we suspect there are likely to be more White Non-College registered voters than these numbers show.

We hope this analysis can contribute to a more realistic understanding of the American electorate — which has improved substantially since 2016’s upset. Demography isn’t destiny, though. It doesn’t predict how different groups in the population will vote. But it is a starting point for sound political analysis and campaign planning in 2022 and beyond. Watch this space for more granular voter demographics as we analyze the results of this election and look towards the next cycle of redistricting.