When we first looked at how early voting was evolving in states that allow the practice, we included an important caveat: these things shift quickly for a variety of reasons, including that we're often talking about fairly small numbers of votes. And, sure enough, our latest look at data from the United States Election Project -- including two new states -- shows a much better picture for Democrats than it did even earlier this week.

First, here's the graph. If you're new to this, dots above the diagonal line represent voting that's outperforming voter registration. Dots below the line, underperforming. The farther from the line the dot appears, the better (or worse) the effect. We've tracked the past two reports as well; they're shown on vertical lines connected to the current data.

Earlier this week, we released a version of this graph that showed red/Republican dots above the line more than blue/Democratic ones. We'd also noted a shift in how the vote in one state that looked good for Democrats -- Iowa -- had shifted negatively for the party. Now, particularly with the addition of the new states, the picture is more nuanced.

We've broken out that tight cluster in the middle to make it easier to see.

The vote in Iowa is about even, with Democrats and Republicans representing roughly similar portions ofn the early vote. The trend is largely the same: Democrats fading, Republicans rising. North Carolina has seen a dramatic turnaround, with Democrats bouncing back above the diagonal line in the newest set of numbers. In Louisiana and Oregon, the two newly added states, Democrats and Republicans are both outperforming registration -- but there are a lot more Democrats in each.

Which brings us to caveats that we should have been more explicit about earlier this week. Dr. Michael McDonald, the University of Florida associate professor of political science who does the thankless* work of compiling the data, explained to the Huffington Post four reasons that the metric we're using (namely: early turnout versus registration) might not be an effective tool for forecasting final results. Those four reasons, in short:

  1. Registration isn't representative of turnout, even in high-turnout years.
  2. Democrats and Republicans tend to differ in how they use different methods of early voting, meaning that there's some apples/oranges at play above.
  3. Even winning the early vote doesn't mean you'll win higher-turnout Election Day votes.
  4. Non-party voters tend to vote later -- meaning that "overperforming" as a function of the percentage of people who have early voted is skewed by lower turnout now in one of the three groups being tracked.

All valid points, of course. The first point, as befits the point that comes first, is perhaps the most important. Pollsters spend a lot of time considering who they expect to come to the polls so that they can weight survey results appropriately. Turnout doesn't reflect registration perfectly. But the fourth point is important, too: a big jump in turnout from non-Democrats and Republicans could shift the party's results significantly. All those yellow dots under the diagonal line could indicate apathy -- or an imminent surge.

That said, as with polling and as with many other things about political campaigns, trends are often revealing. Republicans in Iowa continue to increase the proportion of the early vote they represent while the Democratic proportion continues to decline. That doesn't mean that Joni Ernst (R) is guaranteed to win her Senate race, but it also doesn't mean that Democrat Bruce Braley's ground forces are winning the turnout battle.

* Thanks, Dr. McDonald!