2020 Turnout: Figures and Maps

[Antonio the Sixth]

Hey guys!

So, as promised 6 months or so ago, I've compiled quite a bit of data from the 2020 election, and now I finally have the time to present it. One of the things I was most interested in was the record turnout we had last year, how it was distributed geographically, and how it compared to previous cycles. There are some really interesting patterns.

Overall, 61.54% of the Voting Age Population cast a vote this election, or 66.27% of the VEP. I prefer to use VAP figures, because they're more consistent and rely on fewer estimates and assumptions, and are also less dependent on policy differences across states. However, I do have VEP data as well if anyone is curious.

This is an increase of nearly 7 points from 2016, when VAP turnout was just 54.85%, and represents the highest turnout rate since at least the 1960s (and possibly much earlier). The previous record since 1980 was in 2008, with a 56.95% turnout, but this new figure blows it out of the water. Pretty impressive given how much was made of the enthusiasm back then. The maps below show this exceptional turnout by state. The one on top shows the overall VAP turnout % by state, while the bottom one shows change (in percentage points) from 2016.



Overall, turnout across the US continue to show a North-South divide. Upper New England and the Upper Midwest, as well as Mondana and Colorado, were the only areas to reach a turnout over 70%. Meanwhile, in a long contiguous swath of land ranging from Tennessee to California, turnout remained below 60%, cratering to a low of 51.3% in Texas. It is frankly sad that, even in a year like this, an increasingly purple state like Texas can barely get half of its adults to vote (although in fairness, quite a few of these adults are probably noncitizens). A state like NY also still stands out by its low turnout for a Northeastern state, likely due to its abysmal election laws.

Still, the bottom map shows that turnout improvements have been across the board: every state has seen a higher share of its VAP vote this year than in 2016. There is a massive difference in the extent of this increase, ranging from a low of 2.5 points in Oklahoma to astonishing bounces of almost 11 points in Utah and over 13 points in Hawaii. In general, turnout jumps seem to have been particularly large in the West, likely Western states were among those to most dramatically expand vote by mail this cycle. In California, which adopted universal VBM, turnout jumped from 47% to 57%, making the state pull its weight in national voting more than it previously had. Texas and Georgia also had significant turnout increases of about 8 points, befitting their newly prominent role in national politics. So did the perpetual swing state of Michigan. In the Northeast, New Jersey and Vermont also stand out, having significantly expanded voting themselves. Meanwhile, the Deep South and the lower Midwest saw limited to no extensions in voting abilities, and therefore limited increases in turnout.


Breaking it down by candidate, Joe Biden captured 31.55% of the VAP, more than any Democratic candidate since at least 1980 (and probably since LBJ in 1964). This is up from Hillary's 26.34% and surpasses even Obama's 30.1% in 2008. By the standards of the recent decades, it represents an overwhelming mandate. T***p meanwhile got 28.81%, up from 25.19% in 2016 and more than any Republican candidate since Reagan in 1984 (although Bush in 2004 came close with 28.16%. With such a figure, he would have won the popular vote against Hillary in 2016 and even Obama in 2012. The following maps show the share of the VAP that each candidate received by state (Biden in red and T***p in blue, as it should be):



Both of those maps are fairly remarkable by historical standards. Biden got over 20% in all but a few highly rural states, while T***p did so everywhere but in California, Hawaii and DC. The two coasts and much of the Upper Midwest saw Biden win more than a third of the VAP, and states like Vermont came close to giving him an outright majority. Meanwhile, T***p cemented his dominance over the rural West and much of the Midwest and Appalachia (though interestingly, in most of the South he can't quite crack 35%. Both of those results are significant improvement over the respective 2016 performances, shown in the maps below. As you can see, the contrast is remarkable, especially in seeing how many states gave Hillary Clinton less than 20% of their VAP:



Overall, the increases in both parties VAP showing can be visualized in the map below. Once again, it is remarkable that both candidates improved from 2016 in every single state.



Once again, the West stands out as the epicenter of turnout growth on both sides of the aisle, and especially Utah (although that's largely an artifact of McMullin's candidacy in 2016). Biden's rebound in Vermont really stands out, as he increased his VAP share 12 points from Clinton's. Colorado, the Pacific Northwest, and even Hawaii also stand out massively, with gains of 8 or 9 points. Biden also improved significantly in swing states like MI, MN and AZ, which were key to his victory, and of course in his native state of Delaware. On the other hand, he barely improved in much of the Deep South and Lower Midwest, as well as in the perpetually-infuriating Florida. T***p's gains from 2016 were more evenly spread, which makes sense since it was the same candidate running as the incumbent. Aside from Utah, he made gains of more than 5 points in neighboring Mountain West states of ID, AZ and MT (the former likely a product of a Mormon swing), as well as in Alaska and Hawaii. His 4-point gains in Florida and Texas, likely fueled by low-propensity Hispanic voters, helped him fend off Biden there, and similar gains kept him competitive in WI and PA. Meanwhile, his VAP shares stagnated in much of New England, the DC area, but also Colorado and parts of the Deep South.

There might be more to come in this thread, if there is any interest.

[Biden his time]

-SNIP-

Great stuff man
On a related note, anybody got any idea when we might expect to see turnout figures put onto uselectionatlas.org?

[Antonio the Sixth]

-SNIP-

Great stuff man
On a related note, anybody got any idea when we might expect to see turnout figures put onto uselectionatlas.org?

I wouldn't count on it. Dave seems to have completely abandoned including turnout data in his election results for quite a while. Luckily that's where I come in.

[Antonio the Sixth]

There might be more to come in this thread, if there is any interest.

Not sure if there's a ton of interest, but here's one final set of maps.

So, one thing that you can do once you have candidates' shares of the VAP is to also calculate a candidate's margin of victory as a share of the VAP, rather than as a share of votes cast. This has the effect of inflating the margin of victory in high-turnout states, while deflating it in low-turnout states. In other words, winning a state by 10 points is more impressive when that state had 70% turnout (7-point win in VAP terms) than when it had 40% turnout (4-point win in VAP terms). Once you have candidates' VAP margins, you can compare how those margins change over time: the VAP swing. And once you have VAP swings, you can compare them to the national VAP swing, and calculate the VAP trend.

The maps below show those VAP trends for elections from 2004 to 2020 (click to enlarge):


The first thing you'll notice, of course, is that most of these maps look almost identical to the regular trend maps you find on Atlas (although note that the scale is based on increments of 2.5 points rather than 5). When there are differences, though, they reveal an interesting picture. The maps that differ most from their regular counterparts are 2012 and 2020, and they differ in diametrically opposite directions. In 2012, turnout dropped by about 3 points nationwide, and the VAP-adjusted trend map shows Obama doing better in conservative Southern states such as Texas, Arkansas and Tennessee. Indeed, the entire former Confederacy is a contiguous block of pink. Meanwhile, the 2020 VAP trend map, corresponding to a massive increase in turnout, is almost completely unrecognizable from its standard counterpart. The standard Atlas trend map saw little decisive trend, but in general a blunting of partisan differences, with many middle-Amercan states trending Democratic while states like California and NY trended Republican. These shifts are almost reversed in this map: almost all D-trending states are states Biden won (the only exceptions being NE, KS and AK), while more conservative rural states appear to continue their move to the right. West Virginia, which made big news on Atlas by having a slight trend back to Biden, instead look to have become even more T***pish in this election somehow.

So, clearly, the effect of using this VAP adjustment is clear: when turnout goes down, a VAP-adjusted trend metric will be more friendly to the minority party in a given state, whereas when turnout goes up, VAP-adjusted trend tends to show polarization, with each party doing better in their respective strongholds. Which metric is more useful, I leave for you to decide, as I'm genuinely unsure myself. On the other hand, there's a good reason to take turnout into account: winning by 10 points under 70% turnout is a lot more impressive than winning under 40% turnout. On the other hand, the VAP-adjusted metric would tend to suggest that winning by 4 points under 70% turnout is equivalent to winning by 7 points under 40% turnout, which is not necessarily obvious. I'm interested if there's any debate on this.

[Biden his time]

There might be more to come in this thread, if there is any interest.

Not sure if there's a ton of interest, but here's one final set of maps.

So, one thing that you can do once you have candidates' shares of the VAP is to also calculate a candidate's margin of victory as a share of the VAP, rather than as a share of votes cast. This has the effect of inflating the margin of victory in high-turnout states, while deflating it in low-turnout states. In other words, winning a state by 10 points is more impressive when that state had 70% turnout (7-point win in VAP terms) than when it had 40% turnout (4-point win in VAP terms). Once you have candidates' VAP margins, you can compare how those margins change over time: the VAP swing. And once you have VAP swings, you can compare them to the national VAP swing, and calculate the VAP trend.

The maps below show those VAP trends for elections from 2004 to 2020 (click to enlarge):

This is actually a really interesting way of looking at it that I'd never have thought about
Great thinking

I'd be especially curious to see a county map of Texas with these trends
Expecting a split between the Houston MSA and Dallas-Fort Worth MSA honestly, where Houston trended right and Dallas-Fort Worth trended left.

Very interesting dichotomy

[Antonio the Sixth]

I'd be especially curious to see a county map of Texas with these trends
Expecting a split between the Houston MSA and Dallas-Fort Worth MSA honestly, where Houston trended right and Dallas-Fort Worth trended left.

I'd love to see that map too (or a county map of the whole country for every election, really!). Unfortunately, that would require 1. county-level VAP estimates, which McDonald doesn't provide (those should be available from census data, but I haven't yet found the right dataset to use) and 2. lots and lots of time. Neither obstacle is unsurmountable, though, especially with some help!

Thanks for the bump btw. I'm a bit sad this thread didn't get more interest given that it's probably the single most unique aspect of the 2020 election compared to the previous decade.

[Skill and Chance]

There might be more to come in this thread, if there is any interest.

Not sure if there's a ton of interest, but here's one final set of maps.

So, one thing that you can do once you have candidates' shares of the VAP is to also calculate a candidate's margin of victory as a share of the VAP, rather than as a share of votes cast. This has the effect of inflating the margin of victory in high-turnout states, while deflating it in low-turnout states. In other words, winning a state by 10 points is more impressive when that state had 70% turnout (7-point win in VAP terms) than when it had 40% turnout (4-point win in VAP terms). Once you have candidates' VAP margins, you can compare how those margins change over time: the VAP swing. And once you have VAP swings, you can compare them to the national VAP swing, and calculate the VAP trend.

The maps below show those VAP trends for elections from 2004 to 2020 (click to enlarge):

This is actually a really interesting way of looking at it that I'd never have thought about
Great thinking

I'd be especially curious to see a county map of Texas with these trends
Expecting a split between the Houston MSA and Dallas-Fort Worth MSA honestly, where Houston trended right and Dallas-Fort Worth trended left.

Very interesting dichotomy

2016 really sticks out as weird when you do trends this way.  2020 looks more like a return to some long run normal.

[Biden his time]

county-level VAP estimates, which McDonald doesn't provide (those should be available from census data

Well, luckily the 2020 census does provide it!

You may find this table interesting

[Antonio the Sixth]

county-level VAP estimates, which McDonald doesn't provide (those should be available from census data

Well, luckily the 2020 census does provide it!

You may find this table interesting

Excellent! Yeah, this is very good to have, though I'd have to check if they match McDonald's estimates. I think he explicitly tries to model the VAP on November of the relevant year, whereas the census is based on the count in April. So there might be some interpolation required. Nothing too hard with an Excel spreadsheet, though!

[Biden his time]

county-level VAP estimates, which McDonald doesn't provide (those should be available from census data

Well, luckily the 2020 census does provide it!

You may find this table interesting

Excellent! Yeah, this is very good to have, though I'd have to check if they match McDonald's estimates. I think he explicitly tries to model the VAP on November of the relevant year, whereas the census is based on the count in April. So there might be some interpolation required. Nothing too hard with an Excel spreadsheet, though!

Yep

In fact, I'd reckon while it doesn't match McDonald's estimates, it'd be more accurate.
This is of no fault of McDonald, as he made his turnout, VAP, and VEP estimates back when 2020 Census data wasn't available using the 2018 and 2019 estimates, but there were massive overperformances in many Northeastern and Midwestern states this census, coupled with large underperformances in Arizona and the Carolinas. NY had 800K more people than expected, NJ overperformed by 400K, PA and IL both overperformed by 200K, and AZ underperformed by 250K.

These aren't really small numbers, many hundreds of thousands of extra people, making up significant portions of the population, and when put in the right and wrong places, could result in perceivable swings, especially on the county level, where errors were even greater.

Interpolation, of course, can fix the other problems

[Antonio the Sixth]

county-level VAP estimates, which McDonald doesn't provide (those should be available from census data

Well, luckily the 2020 census does provide it!

You may find this table interesting

Excellent! Yeah, this is very good to have, though I'd have to check if they match McDonald's estimates. I think he explicitly tries to model the VAP on November of the relevant year, whereas the census is based on the count in April. So there might be some interpolation required. Nothing too hard with an Excel spreadsheet, though!

Yep

In fact, I'd reckon while it doesn't match McDonald's estimates, it'd be more accurate.
This is of no fault of McDonald, as he made his turnout, VAP, and VEP estimates back when 2020 Census data wasn't available using the 2018 and 2019 estimates, but there were massive overperformances in many Northeastern and Midwestern states this census, coupled with large underperformances in Arizona and the Carolinas. NY had 800K more people than expected, NJ overperformed by 400K, PA and IL both overperformed by 200K, and AZ underperformed by 250K.

These aren't really small numbers, many hundreds of thousands of extra people, making up significant portions of the population, and when put in the right and wrong places, could result in perceivable swings, especially on the county level, where errors were even greater.

Interpolation, of course, can fix the other problems

That's a very good point, yeah. I expect McDonald himself to update his turnout figures based on the new census data soon enough. I might want to wait for that before compiling the 2020 estimates. Estimates for previous elections should be good to go, though!