In the Georgia Senate runoff elections, some of the only data we have to analyze whether Republicans or Democrats have the advantage is who’s voting early.

Early voting started Dec. 14 in the two Jan. 5 runoffs between Sen. David Perdue (R) and Jon Ossoff and Sen. Kelly Loeffler (R) and Raphael Warnock, which will decide control of the Senate. So far, at least 2.6 million voters have cast their ballots, which is already a turnout record for a statewide runoff in Georgia.

To date, Democrats seem to be in a slightly better position than they were at this point before the November election, thanks to Black turnout being a larger proportion of the early votes cast in the runoff.

But there are a number of reasons Democrats’ perceived advantage might not translate to victories on Jan. 5.

Early vote data doesn’t capture all voters

At any given moment, the early vote data is just a snapshot of who has voted so far. More people are going to vote, and partisan trends often shift significantly as the early-vote period goes on.

(There is a small exception here for states that conduct their entire election by mail — such as Washington, Oregon, Colorado, Utah and Hawaii. Nevada is also a bit of an exception because it has such high vote-by-mail participation. Given the early vote is the vote, or at least the vast majority of it, in Nevada, trends there can tell us more.)

Historically, only strong partisans, older voters, military and overseas voters voted by mail or early. But the pandemic and expansion of absentee voting has led many more Americans to vote early. These changes make it difficult to determine who else has not voted, and, especially, how many people are still going to vote.

This was tricky to analyze even before the pandemic. In the 2016 presidential election, absentee votes led many to think that Hillary Clinton was favored to win the state of Iowa. Democrats’ early vote margin in the state had shrunk relative to 2012, but because Obama had won the state by nearly 6 percentage points, the Republican Election Day vote was not expected to be big enough to change the final outcome.

However, Donald Trump ended up winning Iowa comfortably, by more than 9 percentage points. The Election Day electorate had simply been underestimated. While in 2012, 44 percent of voters had voted absentee, in 2016 that number was only 41 percent, and the Republican strength among those voters swamped the Democratic early voting advantage.

Trends might not hold throughout an election

One way to mitigate the problem of overestimating early vote data is to look at historical trends. For example, if Republicans were more likely to vote by mail in the past, then maybe we can assume that they are more likely to vote by mail now too.

The problem with that is that trends change — and often suddenly. Historically, Republicans were more likely to vote by mail in Florida. In 2016, 40.5 percent of mail votes there came from Republicans, while 38.4 percent came from Democrats. However, after the unprecedented attacks on mail-in voting by President Trump, many Republicans decided to change how they voted. In 2020 only 31.1 percent of mail ballots were cast by Republicans, and 35 percent were cast by Democrats.

This large drop in vote-by-mail among Republicans is one of the reasons political analysts initially thought that Florida was going to be closer in the 2020 presidential race than it ended up being. While a drop in GOP participation in mail-in voting was expected, the surge in Republican day-of voting was substantially larger than expected, leading to Trump comfortably winning this swing state.

We don’t know who people are actually voting for

The third reason you should be careful when trying to infer much from early voting is probably the most obvious one: While you might know voters’ partisan affiliation, you don’t actually know who people voted for.

Demographic and party information is sometimes provided by the states (and sometimes by voter file vendors), but while that can be useful, it definitely doesn’t tell the entire story. This is especially true now as we are probably witnessing a political realignment in which Democrats are losing support from Hispanic voters, and possibly Black men, while gaining support among White voters with college degrees.

Joe Biden’s narrow victory in Georgia, coupled with two runoff elections that will determine control of the Senate, have drawn national attention to the state. (The Washington Post)

Education is quickly becoming a key factor, and that’s not something collected on voter registration forms. It’s a lot harder to model. While regional differences in education and turnout can give us some insight into whether people with or without bachelor’s degrees are voting, this type of ecological inference is also fraught.

Even using party registration (as we did above with Florida) isn’t as straightforward. Party registration, party affiliation (as in, which party one might feel closest to) and presidential vote preference aren’t necessarily always aligned, which can make it hard to use them to predict the final outcome.

What does this all mean for Georgia?

In November, despite the pandemic, more Georgians voted early in person than by mail. It looks like that trend is continuing for the January runoffs, with proportionally even more Georgians voting in person than by mail.

This suggests that more people might vote on Jan. 5, the actual Election Day, as well. Since the early vote seems to favor Democrats, the Election Day vote will probably favor Republicans. The key question is how much better will Republicans do?

But even more generally, comparing the current numbers to those from October and November is problematic. Christmas and New Year’s mean there are a few days this time where early in-person voting won’t be possible. And we generally expect early voting to slow around the holidays.

This does not mean that you should discard early voting numbers, but it does mean that you should be careful when consuming them. The next time you read somewhere that Republicans need X percent of the early vote to be White so they can win, or that Democrats need Y percent of mail-in votes to come from voters under the age of 30, remember that there is a lot we don’t know about early vote data.