Voters cast ballots in early voting at the Potomac Community Recreation Center in Potomac, Md. (AFP Photo/Brendan Smialowski/Getty Images)

Depending on whom you believe, analysis of early voting is either a good or bad way to tell who is likely to prevail on Tuesday. There are some cases when it seems pretty clearly predictive, such as in Nevada, where Democrats have built up a big lead thanks to strong support from Hispanic voters. In other cases, it's not immediately clear what the data show.

One statistic off the top: More than 40 million people have already voted. That's nearly a third of the total vote in 2012.

Let's walk through some examples of what that early vote looks like. Catalist, a progressive data firm, provided us with early vote data from about midweek, as well as comparative data from 2008 and 2012. Midweek is actually fairly old at this point, but some clear patterns emerge that are worth looking at.

Racial composition of the vote


One thing that we can tell from the Catalist data is the shift in black participation in early voting. In Georgia, North Carolina and Ohio, blacks make up a much smaller percentage of the early vote than they did four years ago. In Ohio and North Carolina, that shift is due in part to changes in early voting laws, with Ohio cutting a week during which voters could both register and vote. North Carolina's shifts have been well-documented in the press, with a federal judge stating that the change in the law was directly meant to cut black voter -- and, therefore, Democratic -- participation. It seems to have worked. Democrats were much more of the early vote electorate four years ago.

The flip side of that decrease among black voters in North Carolina and Georgia is a big increase in the density of the white vote. In Wisconsin, as well, more of the early vote electorate is white than in 2012. That suggests an advantage for Republican presidential nominee Donald Trump, who polls better with white voters than nonwhite voters, particularly in more Republican states.

In Nevada (as we mentioned above) and Arizona, the density of Hispanic voters has increased since 2012.

Catalist doesn't have data available for 2012 in Florida, but Daniel Smith, political science professor at the University of Florida, does. By his estimates, the early vote among Hispanics in the state is up 152 percent over the same time in 2012. What's more, Steven Schale, who ran President Obama's 2008 operation in the state, estimates that a majority of those Hispanic voters are low-propensity voters (meaning they've never voted or voted in one of the last three general elections). If that's the case, those voters wouldn't be well-represented in polling that focuses on likely voters.

Geography

Sometimes the geography of the early vote can tell you a lot, like in Virginia. The black-and-white maps below were made by Catalist. We've added maps of the 2012 results to give a sense of what geography is important for which races.

In Virginia, you'll notice, the increase in early voting ballots is heavily centered in the northeastern part of the state -- the Democratic suburbs of Washington that are heavily responsible for the political shift in the state.


In other states, the meaning isn't clear. Michigan saw a drop in black turnout as a percentage of the early vote this year, too. You can see the slight contraction since 2012 in the dark-blue county at the lower-right of the map. That's Detroit. Trump feels as though Michigan is within his grasp, which isn't reflected in the polls. But Democratic nominee Hillary Clinton swung by the state on Friday, suggesting that her campaign wants to bolster turnout.


Wisconsin has seen a big boost in early voting across the state, which is in part a function of changes to early-voting laws. Such broad gains makes geographical analysis trickier.


Gender

For the most part, the gender composition of the early vote in states is on par with 2012. There are four exceptions, though.

In Michigan, North Carolina and Virginia, the early vote has drawn a greater number of women -- by 1.6, 2.2 and 3 percentage points, respectively. (By way of explaining the math, the electorate in Michigan is 0.8 percentage points more female now than in 2012, and therefore 0.8 percent less male.) In Wisconsin, the electorate is 4 percentage points more heavily male.

What does that mean in vote totals? In North Carolina, a lot: 182,000 more women have voted so far this year than voted in 2012, compared with 103,000 more men. Women break more heavily for Clinton than do men, which seems like a good sign for her. In Wisconsin, there were still more ballots cast by women than men. It's just that the margin was narrower. (There were 150,000 more women who voted early so far this year and 133,000 more men.)

We don't know who those voters were specifically or how they voted, which is a good reminder that we should still consider these numbers with caution. But they look the way we'd expect based on polling: Good news in Virginia for Clinton and good news in Ohio for Trump, for example. As the losing candidate always says, though, the only poll that matters is on Tuesday -- when all of the votes are counted.