Early vote analysis has always been a sketchy business. But after the pandemic upended historical voting patterns in 2020, it’ll be especially easy to misunderstand and misrepresent the data this year. Sure enough, the first pieces of early vote “analysis” popped up to torment the internet’s collective intelligence before we’d reached September. So let’s prepare for our favorite unhealthy obsession and talk about early voting data and a couple of the most common ways we’ll see it distorted this year! 

We’ll break the early vote (EV) into two buckets: in-person early voting and absentee/vote-by-mail (VBM). Most states have some type of widespread early voting, but the features vary. Some states release data about their early vote, from demographics to breakdowns by partisan registration. For example, Pennsylvania’s last report of the 2024 primary tells us that registered Democrats had returned 499,000 VBM ballots, and Republicans returned 175,185. Alright, now we’ve set the stage, so just how useful are these numbers?

What EV data can(’t) tell us

The most important thing to understand is this: No matter what, early voting data will always offer an incomplete picture at best. The pre-election data will always be missing election-day votes and unprocessed mail-in votes. And while the attention on the partisanship numbers is high, there isn’t a 1:1 relationship between party registration and vote intent. 

Can EV data tell us who’s going to win an election?

Broadly speaking, no. If we expect close races, that tends to be too precise of an ask when we can’t know election day turnout or margins. But there are exceptions. Jon Ralston’s Nevada predictions are famous for their impeccable track record. Ralston is an exception partially because Nevada has historically had a few features that make EV data more useful:

1) The partisan composition of the EV roughly matches vote intent.

2) There’s enough EV to be meaningful and enough pre-election data about it to draw solid conclusions.

3) Worthwhile historical comparisons¹.

Even then, Ralston’s predictions are not purely driven by EV data. In 2022, he correctly predicted a split result, with Democrats winning the senate race and Republicans winning the governorship.  Early voting data can’t tell us anything about ticket-splitting. Ralston made that prediction (and all of his) only because of his unique familiarity with the state, its elections, and its campaigns. Many states don’t have the same characteristics as Nevada, and most don’t have someone like Ralston to fill in the information gaps.

This doesn’t make early voting data useless, though. Split Ticket’s incredibly successful Georgia runoff models are EV-based. The fresh baseline from the extremely recent general election provides predictive value even without partisanship data (detailed demographic data and racially polarized voting are a helpful combination there). In the absence of an exceptionally recent reference point, we should avoid turning EV data into something it’s not. 

After Trump’s anti-VBM rhetoric in 2020, Election Day voting tends to be much more Republican than the early vote (VBM in particular). With Republicans preferring election day, the EV data will naturally tell us more about the position of Democrats. For instance, Pennsylvania’s VBM returns match close enough with results to indicate how much of a VBM “buffer” Democrats have for Election Day. And in some cases, EV data can tell us if there’s an impending disaster for Democrats. In 2022, a large Republican lead in Florida’s early vote signaled the apocalyptic result in store for Democrats that year. 

But this is where it gets tricky. When things aren’t as nightmarish as Democrats outright losing a “swing” state’s early vote (most of which is VBM!), you’d want to look at previous EV data from elections to look for swings. A four-year-old election already isn’t a great baseline, but 2020 is a uniquely poor election to follow. It was an aberration in voting behavior because of the pandemic and left enough permanent changes behind to make any prior election just as unhelpful.

Why things will be just that much worse in 2024

As a share of total votes cast, early voting spiked from an all-time high of 39% in 2016 to nearly 70% in 2020. Democrats in particular embraced mail-in voting to minimize personal contact during COVID. This year, a few things are true. There’s going to be a lot of early voting. Aside from 2020, it’ll easily be the most ever seen in a presidential election. However, without pre-vaccine pandemic concerns, the total won’t come close to 2020 either. 

How this manifests will depend on each state’s history with early voting (and its election laws). There will be reversion, but some behavioral and structural changes are permanent too. Pennsylvania first introduced no-excuse mail-in voting in 2020, making it a great example:

This table shows the share of total votes cast by mail and on election day for the past three elections.

More From Pivot Point!

On The "Vibecession"

You Made Your Bed, Now Lie in It

Where Have All The Republicans Gone?

GA Redux: Can Dems Win Again?

2024 Ratings

Now back to the article

In 2022, 23% of all votes cast were VBM. That’s considerably less than the COVID-fueled 38% in 2020, but way ahead of the 4% in 2018 (before VBM was widely available). It makes sense that behaviors will be different than during the height of the pre-vaccine pandemic. We can be confident that VBM usage will fall from 2020. But since VBM wasn’t widely available in the state before then, previous years won’t function as a decent baseline.

That won’t stop the data punditry though. Because Pennsylvania’s VBM is roughly 75% Democratic, some people argued the drop in VBM was a bad sign for Democrats in 2022. The awful takes we’re already seeing this year? Essentially the same claims, but using incomplete ballot request data instead of returns (somehow managing to make the argument worse). But John Fetterman won easily in 2022 anyway, because those Democratic voters didn’t disappear — they just returned to election day. Biden won 34% of election-day voters in 2020, and Fetterman won 44% in 2022, receiving more raw votes on election day than Biden did.

This type of pattern will repeat all over. In some states, mail-in voting may return to being nearly nonexistent. Democrats will get better margins on election day. But doesn’t mean those margins will improve enough to win, either. So here are a couple of claims you can safely ignore:

“This big decrease in early voting is bad news for Democrats!” 

“This big lead in the early vote is great news for Democrats!”

We should expect these things to happen as a baseline, so neither should qualify as good or bad news. Especially in isolation, these are hackish arguments designed more to score partisan points than to be tethered to reality. This bad-faith “analysis” is usually easier to spot because the posts tend to be clearly partisan and hyperbolic. It’s “HUGE” or a “BIG DEAL,” and the people behind them don’t even try to look all that serious. 

The return timing problem

A common method of reading early voting data is comparing numbers (usually partisan composition) to equivalent points in a previous election cycle’s early voting window. This appears to be a more serious way to consider early voting data (and in many ways it should be), but it suffers from how distorted 2020 is as a baseline. 

In 2020, Democrats didn’t just vote by mail in historic numbers. Spurred by highly politicized US Postal Service fears, they also returned those ballots historically quickly, naturally making later VBM returns disproportionately Republican.

This dynamic led to Fox’s infamously premature Arizona call for Biden (and the AP call a few hours later). They hadn’t accounted for the change in return patterns. Fox’s decision desk only expected Trump to get 44% of the outstanding vote when they made their call. Trump would instead receive about 60% of that vote, just shy of the 61% he needed to overtake Biden.

Later VBM returns were more Republican everywhere, even if it presented differently. In Arizona, where almost everyone (including Republicans) voted by mail, Trump outright won later drops. But in a mostly election-day Pennsylvania, Biden won later VBM drops by closer to 70-30 instead of the nearly 80-20 leads he had on election night.

This effect is very visible in early vote data. Just like with VBM usage overall, it’s also been unwinding after 2020, creating another snag with year-to-year comparisons. To show what I mean, let’s look at two states with a long history of VBM use: Arizona (majority VBM, almost entirely in 2020) and Colorado (a universal VBM state with negligible in-person Election Day voting). Let’s visualize this a couple of ways. First, this table shows the share of each party’s total early that was already returned one week before election day in 2020 and 2022.

Take Colorado’s 2020 numbers. That year, Democrats returned 1.025 million ballots. Over 662,000 had already been returned a week before election day, roughly 65% of the eventual total. For Republicans, it was about 50% (472,000 out of 945,000). A similar gap existed in Arizona and almost everywhere with enough mail-in voting. But in 2022, with less urgency behind Democratic returns, the return disparities either partially (like AZ) or fully closed (Colorado mailed ballots out later in 2022, making the absolute drops fairly large). We can see how this makes for some misleading comparisons by graphing out the partisanship of each state’s early vote over time.

These graphs show the net Democratic advantage in the partisan composition of the early vote during the two weeks before election day in 2020 (purple) and 2022 (green).

Fourteen days before the 2020 election, about 42% of Colorado’s returns were from Democrats and 23% from Republicans. That 19% margin narrowed to less than 5% by the day before the election. 2022’s early vote partisanship was instead nearly constant for the entire 14-day period. If we looked at the partisan composition of the early vote at the seven-days-out mark, we’d find: 

Arizona: D+6 in 2020 and D+2.5 in 2022

Colorado: D+10 in 2020 and D+4 in 2022 

With the full picture, we see that the two years were converging. The partisanship difference between 2020 and 2022 would nearly vanish in both states. But for most of the early voting period, same-day comparisons would have shown a large shift in EV partisanship that did not exist. 

2024’s EV patterns will look much more like 2022 than 2020. The reversion could easily continue too, making 2024 even less like 2020. We’ll see comparisons to 2020 anyway, often completely devoid of context. It might be horserace-esque framing like “Compared to the same point in 2020, the early vote is 3.5% redder in Arizona and 6% redder in Colorado.” It could also be a long list of comparisons, with no attached commentary: just the partisanship of the EV in 2024 for each state next to the 2020 numbers for the equivalent day. 

The latter is often justified as “letting the numbers speak for themselves” or “just sharing the data.” I think that’s a weak argument. Leaving out context that is fundamentally important to understanding data is misleading. Letting something speak for itself is entirely unhelpful if it can’t speak in complete sentences.

We’re still in a period where voting patterns are actively shifting, trying to settle into a new post-pandemic normal.  There’s so much noise that the data itself will often show things that are wild but ephemeral. Side effects of the first post-pandemic election, only making it easier for anyone who goes looking to find something to spin into good news for their preferred team. And people absolutely will go looking. There will be plenty of wild declarations, ill-advised comparisons to 2020, and premature graves dug. 

The best advice is, of course, just to be patient. But I know that’s not going to stop anyone (it certainly won’t stop me!). So instead, spare as much mental energy as possible by being ready for the ungodly amounts of bad early voting takes. We’re in for a long couple of months, so good luck out there!

_____________________

¹ But even Ralston’s task is getting more difficult. Nevada’s registered voters are becoming increasingly unaffiliated, and Nevada introducing universal VBM and dropboxes has meant a significant portion of the early vote isn’t visible until after election day.