Early Voting a Poor Predictor of Final Results

Early Voting a Poor Predictor of Final Results
AP Photo/Elise Amendola
Story Stream
recent articles

As we wind down this election, we begin to hear the familiar chime: The election is over and/or can be predicted because of what we see in the early vote. Don’t buy it. While we might be able to make some broad projections based upon early voting – maybe – we’re more likely to substitute our own judgments and arbitrary intuitions for actual results.

There are (at least) three reasons this it true. The first is theoretical. We can think of an election result as accurately represented by the following equation: The Democrats’ share of the vote is equal to the Democrats’ share of the early vote times the number of early votes, plus the Democrats’ share of the Election Day vote times the number of Election Day votes.

Even if you don’t have a math degree, you should be able to intuit the gist of the problem: We are missing two of the four variables for the equation, and guessing at a third. All we really know is the number of early votes cast.

Now we might be able to get a sense of how Democrats are performing in the early vote by looking at African-American turnout or overall Democratic turnout (in states with partisan registration), but we can’t know how independents are voting. We might make assumptions about this by looking at public polling, but then what value are we adding beyond what the public polling says? Plus we’re incorporating the error margins of public polling into our estimates, which will be even greater for demographic subsamples.

The real problem with this, however – and this is true with a lot of early voting analysis – is that for any of this to work we have to assume that the early vote is somehow representative of the Election Day vote in order to fill in the second half of the equation. The problem is, it isn’t. Research suggests that the early vote tends to be comprised of more partisan, higher propensity voters. In the most recent elections, they have often skewed Democratic, most likely as a side effect of increased Democratic emphasis on early voting (compare this with Donald Trump, who has been telling his supporters to vote on Election Day).

So basically, we’re left without really knowing how the early voting electorate is voting, without knowing how the Election Day electorate is likely to vote, and without knowing the size of the Election Day electorate. More importantly, we don’t know the effect to which campaign strategy is creating the appearance of a participation surge by merely cannibalizing Election Day voters by mobilizing voters who would have voted on Election Day anyway. This is a problem.

Second, predictions from early voting have a decidedly mixed track record. As University of Denver political scientist Seth Masket has suggested, the relationship between early voting results and Election Day results is pretty weak. That isn’t to say it is non-existent, but we should probably expect some type of relationship between the partisan split of early votes and the state as a whole (that is, we’d expect more Democrats to vote early and on Election Day in Maryland than, say, Utah.) A contradictory report has been published at the Monkey Cage blog, but that finds a relationship utilizing proprietary Catalist data, not the publicly available data from which everyone is presently extrapolating.

But if predictions from 2012 were iffy, predictions from 2010 and 2014 were awful. In 2010, analysts saw huge Democratic advantages in turnout in places such as Ohio and Iowa and thought that perhaps there was no enthusiasm gap in the election. In 2014, it was widely assumed that early vote totals were good news for Democrats in states including North Carolina and Iowa; Thom Tillis ended up winning in North Carolina on the back of strong Election Day turnout, while the 2014 Iowa Senate race was decidedly not close (as early vote analysts had suggested); Joni Ernst won by almost 10 points.

Third, given the lack of scientific rigor in a lot of these projections, analysts can easily fall into the trap of filling that gap with their own assumptions. As I wrote in 2014:

Humans are remarkably adept at discovering and using patterns. We don’t like chaos, and this is part of what has allowed us to advance as a species. Yet our minds aren’t precisely fine-tuned to patterns; we’re overly sensitive, and so we see dragons in clouds, a man’s face on the moon, and images of Mary in a grilled cheese sandwich. If I gave you a page with 15 dots and challenged you to fill in the gaps with what you saw, you’d probably come back with a picture of a Dimetrodon (or at least, that’s what I’d be inclined to draw) or some such; you wouldn’t likely return the page and tell me it is just random noise.

We do the same thing with data. We do it in very obviously bad ways – there was a cottage industry of predicting presidential elections based on the winner of the final Redskins football game from 1932 to 2004 (there’s actually a statistically significant correlation between the margins of those games and the margin of presidential elections during this time).

But it’s most dangerous when we have good reason to believe there has to be a pattern.

To see an example of this, consider North Carolina, where early voting patterns are frequently cited as good news for Democrats. You can certainly make that argument. But consider the counterargument, using these data from Dr. Michael Bitzer. Early voting turnout is up somewhat from 2012, but does that suggest heightened turnout or not? The trend in North Carolina has been toward voters shifting from Election Day voting to early voting for quite some time. Perhaps voters have continued to shift from Election Day voting, and overall turnout is off quite a bit.

Dig deeper into the data. Remember that in 2012, Romney narrowly won North Carolina. As Bitzer's data clearly shows, Democratic participation is down, Republican participation is up, and unaffiliated participation is significantly up. Now, this could mean that unaffiliated voters in urban and suburban areas are coming out of the woodwork to reject Trump. Or it could mean that Trump’s supporters are turning out en masse. Which solution you are attracted to probably depends on your prior beliefs, and any additional information you hunt down to test your conclusion is probably motivated, at least in part, by your earlier beliefs as well.

It's worth noting in Bitzer's data that the black percentage of the early voting electorate in North Carolina is down significantly from 2012 as well, to the tune of some eight percentage points. Incidentally, we’ve seen similar effects in other states that track early voting by race, such as Georgia, Florida and Louisiana. David Wasserman has offered similar analysis from Virginia (which has limited early voting).

If African-American participation rates revert to their historic mean of trailing their white counterparts by six points, a lot of the polls are going to be off, including in places like Ohio, Michigan and Wisconsin. But all of this is meaningless if African-American voters were uniquely motivated to vote for Barack Obama in 2008 and 2012, and so voted early in those years, but still opt to vote on Election Day of 2016 despite lower enthusiasm.

So there are multiple ways to interpret this. You could believe that African-American voters are just going to show up on Election Day, or you could look at a surge in white unaffiliated voters and see a huge vote materializing for Trump. But what you decide is likely affected heavily by your prior views.

Finally, I have read a lot about how early voters are locked in, or banked, and so they are unable to respond to late-breaking news in the election by deciding not to vote or by changing their mind. There is some truth here, and that is clearly part of campaign strategy. The problem is that, as we noted above, early voters tend to be people who have made up their minds already. It is true that both campaigns have large numbers of voters banked, but the remaining persuadable voters and voters who haven’t decided whether to vote are most likely disproportionately in the Election Day pool already.

As I noted above, I will admit that extreme results may allow for some predictions; journalist Jon Ralson, the best analyst of Nevada politics there is, saw the Democratic rout in the state coming in 2014 because of Republicans’ unusually strong showing in early voting, though even he hedged about Election Day turnout and understated the extent of the wave. But overall, there is no need to engage in risky tealeaf reading from early voting when polls have a much better – and longer – track record. We will find out how this election is going to turn out soon enough.

Sean Trende is senior elections analyst for RealClearPolitics. He is a co-author of the 2014 Almanac of American Politics and author of The Lost Majority. He can be reached at strende@realclearpolitics.com. Follow him on Twitter @SeanTrende.

Show commentsHide Comments