Historically, data on early voting has not been very insightful in predicting final outcomes. This is why the moniker, “Democrats win the early vote while Republicans win Election Day” is a common and true refrain. Yet, just as with many other variables in 2016, it was used as a justification to fit a narrative that Donald Trump would lose.
Many smart people have fallen into the early vote trap. Republicans argued positives in 2012 meant Romney was in a stronger position than state polls showed. Democrats did the same in 2014. Trump’s camp and Republicans did not argue the early vote numbers for 2016 but the media made it seem as if the election was all but over because of it (that is until actual returns started coming in).
Not only this, but interpretations of early voting can differ by state. For example, the Washington Post reported Hispanics early voting totals in Nevada were 99 percent higher than 2012. Yet, Fox News argued it remained a close race. In states that do not register voters by partisanship, early voting numbers are a patchwork of guesses at best and usually come down to urban vs. rural and red vs. blue.
This is not to say that early voting totals do not have their uses. They do. They can give us indicators of how a party is fairing with its base and the enthusiasm of the party faithful for their nominee. But it does not provide enough statistical data to call an election before we have it.
Yet, that is exactly what we saw last year. Initially, early voting numbers cited in Mid October correlated with Trump’s sagging numbers. But once October passed and November came about with a tightening in the polls, early voting data was used in opposition to the polls. Indeed, sites such as FiveThirtyEight and Realclearpolitics were hammered for not incorporating early voting data in their tallies.
So what happened? Well, if you want to take the two highest profile cases of where early voting was off you can look at North Carolina and Florida. In North Carolina, Clinton won the early vote by 2.5 percent. The early vote comprised about 47 percent of the electorate. But, Trump won the election day vote by 16 percent and it was 53 percent of the electorate. Thus, Trump easily flipped the early vote totals.
What about Florida? Well, despite Democrats recording record early turnout, Donald Trump won the state. He lost the early vote by only 1.5 percent (compared to Romney’s 3 percent) and these votes comprised over half of the final tally. Where about 6.4 million voters voted early an additional 3.5 million opted to wait until election day. Among these voters, Trump crushed Clinton by double-digits.
In fact, the GOP’s election day surge was nothing new. In North Carolina, Clinton was running behind Obama’s margins just as in Florida. There is also evidence the Clinton campaign knew this was an issue and made last-ditch efforts to increase its lead (to no impact). Predictably, few networks noted this trend and simply focused on the top-line numbers (forgetting the mantra above).
It should be stated early voting is still fairly knew. Not every state has it and access varies. New laws and regulations can impact turnout in ways we cannot measure. For example, despite GOP voter ID laws that Democrats decry minority turnout increased in North Carolina and Florida (especially Georgia).
Additionally, early voting numbers are generally reported at once from different states and given little context. For example, early voting numbers in Florida are reported the same way as numbers from Mississippi, despite the fact one state requires registration by party and the other does not. This lack of context makes the numbers meaning even fuzzier.
It is also true that the networks tend to focus on one pattern over others. For example, this cycle, many networks noted the massive surge in Latino turnout in early voting in Florida. Less noticed was the decrease in turnout in South Florida and, later on in the night, the Midwest. The Hispanic surge fit with the narrative the media had pushed since the beginning of the campaign about how Trump was damaged with Hispanics over immigration.
It should also be mentioned that early voting is not a good measure to use to question polls. Many polls account for the early vote and weight accordingly. Additionally, more and more pollsters are breaking this down in their cross-tabs for public information.
This ultimately brings up concerns over the validity of data. There was tons of data this cycle open to interpretation. The problem was the interpretation was one-sided. Almost every time, polling data was used to argue Clinton was benefiting despite the fact she was lagging behind Obama among key constituencies. Likewise, few pollsters, nor the media, talked about how beneficial it was for Trump to have such strong support in the Midwest. Instead, it was stories about Clinton in Arizona or playing in Indiana while Trump made a desperate, last ditch appeal in Minnesota.
In this we can see the warning signs about how data can be misinterpreted. Polling has issues and the media tends to practice group-think. We have access to tons of data but not always the tools to make sense of it. As such, we should be careful what conclusions we draw from data.