Skip to main content
‘Demographics Aren’t Destiny’ And Four Other Things This Election Taught Me

The 2016 election is in the books. Donald Trump won; Hillary Clinton lost. But it will take a while — weeks and months — to sift through the results, so be wary of any stories drawing sweeping conclusions about the country (it’s still the same nation that elected President Obama twice). That said, there are a few lessons we can learn from the results and a few myths we can hopefully now put to bed. These are my five big takeaways from the election.

1. Demographics aren’t destiny

The country is getting more diverse. That’s indisputable. But some analysts had argued that increasing racial and ethnic diversity meant that Democrats would have a durable, structural advantage in presidential elections. That was never true, and the results in 2016 show why. Trump was able to win, in large part, because he won over a lot of northern white voters without a college degree — in states like Wisconsin, Michigan and Pennsylvania, for example. Many of these voters had cast ballots for Obama twice. Trump’s more populist message likely helped him outperform recent GOP nominees with these voters.

Political parties, in other words, are dynamic — their coalitions change. Some people, including me, were surprised that it was Trump who was able to attract these voters to the GOP. But no one should have been surprised that the country’s growing diversity didn’t guarantee Democratic victory. Only two years ago, in the 2014 midterm elections, Republicans were able to win big nationally among an electorate that was just as diverse as it was in 2008, when Democrats scored a blowout victory. A lot of Democrats dismissed that win as merely a product of a whiter, older midterm electorate. They shouldn’t have.

And, no, Democrats won’t be safe even as the electorate becomes more diverse. Republicans could do even better with white voters. In some Southern states, for instance, GOP candidates win close to 90 percent of white voters. Who’s to say that won’t happen in the Midwest? Alternatively, Republicans could improve their standing with nonwhite voters. In heavily Latino Texas, for instance, Republicans have long done better with Latino voters than Republicans have done nationally.

2. There is no blue wall

Before the election, many analysts were advising people to ignore the tightening national polls because Democrats had a lock on 270 electoral votes. Democrats had won at least 251 electoral votes in every election from 1992 to 2012, and with Colorado and Virginia in Clinton’s column too, she supposedly had it won. It didn’t quite work out that way. Indeed, instead of the Electoral College swinging a close election to Clinton, the exact opposite happened: Trump won the Electoral College even as he lost the national popular vote.

The blue wall is — and always was — a myth. Democrats won the Electoral College in four of the six elections from 1992 to 2012 because they won the popular vote by about 4 percentage points or more in those four cycles. Of course you’re going to win the Electoral College when you’re winning nationally by those margins. The last time the national vote margin was less than 2 percentage points was in 2000, which was when the Republicans won the Electoral College even as they lost the popular vote.

Now, maybe people thought things had changed since 2000. In 2012, Obama won by more than 5 percentage points in states worth 272 electoral votes — even as he won nationally by a little less than 4 points. The problem is that there were signs that Obama’s map wasn’t going to hold for Clinton as early as two years ago. Iowa, which had voted more Democratic than the nation in every presidential election from 1984 to 2012 (with the exception of 2000), looked like it was shifting out of the Democratic column during the 2014 midterms. The reason was that white voters with a college degree in Iowa were abandoning the Democratic Party. Two years later, Trump easily carried Iowa, along with many non-college-educated white voters elsewhere in the Midwest.

Republicans should be careful, though. History has shown that just because a majority of electoral votes are more Democratic or Republican than the nation as a whole in one election doesn’t mean they will be in the next election. Which party holds the Electoral College edge has mostly been random and dependent on the coalition each has built. Next time, the Democratic candidate may build a coalition that’s more optimally distributed to win the Electoral College. So there’s no blue wall, but there probably isn’t a red wall either.

3. Party identification is a hell of a drug

Trump won only about 45 percent of the vote in the Republican primary, and he espoused a number of policy positions that are out of line with conservative orthodoxy. Four of the past five Republican presidential nominees declined to endorse him. You might have thought, therefore, that Trump would struggle with Republican voters. It didn’t happen.

Trump won 90 percent of self-identified Republican voters, a higher percentage than Clinton won of Democratic voters. Trump did suffer a small penalty among Republican voters — House Republicans won 94 percent of self-identified Republican voters — but the fact that Trump lost so little ground despite all his conservative heresies and flaws tells you how strong partisanship is these days. With few exceptions, Republicans vote for the Republican and Democrats vote for the Democrat — pretty much no matter what.

Party identification is far more important than it was 30 or 40 years ago. Back in 1980, for example, Ronald Reagan was able to win nearly 30 percent of self-identified Democrats. As late as 1988, George H.W. Bush carried 17 percent of Democrats on his way to an 8-point victory in the popular vote. On Tuesday, even Iowa Sen. Chuck Grassley — a household name in Iowa who won re-election to his Senate seat by 25 percentage points — could only match Bush in winning a little less than 20 percent of Democrats.

Another way to look at the importance of party identification is the total lack of split-ticket voting. This year was the first time ever in which every state that voted for a Democrat for Senate also voted for a Democrat for president, and every state that voted for a Republican for Senate also voted for a Republican for president. The correlation between the percentage of the vote that Democrats running for Senate and the Democrat running for the presidency received in individual states was also a record. It seems that once someone decides to vote for one party in federal elections, they’re more likely than ever to vote for the same party for all levels of federal government.

4. Bigger sample sizes matter

One of the reasons our model was more bullish on Trump’s chances than other models were is because it was built using data that dates back to 1972. Forecasters have been fortunate that state polls have been fairly accurate over the past couple of years, but that hasn’t always been the case, and a forecaster looking at data only over the past few cycles would have thought the chance of a Trump win was considerably lower than it actually was.

The polls have often been off by more than 3 percentage points, as the state polls were in 2016. The national polls were at least that far off in 1980, 1996 and 2000, for example. In 1980, they were off by just over 7 points. Given that our past modeling has shown that these errors are often correlated across states, the amount by which the polls were off in 2016 was not unusual. No one should have expected the polls to be perfect.

Small sample sizes can also be super misleading. That’s exactly what made my predictions (and others’) in the primaries so bad. The idea that a candidate like Trump, who had so little support from the party establishment (as measured by endorsements), couldn’t win a primary in the modern era was based off of data dating back only to 1980. That’s only nine election cycles’ worth of results. If our primaries model had looked back further, it would have included the information that neither George McGovern or Jimmy Carter had many endorsements when they won the Democratic nomination in 1972 and 1976, respectively, and that context may have made Trump’s path to the nomination seem more plausible. I, like many others, dismissed that possibility for a multitude of reasons that, clearly, were not as sound as we initially thought.

Going forward, modelers need to recognize that including more data from previous eras is a good thing, not a bad thing. Uncertainty isn’t the enemy. In fact, expressing that uncertainty is one of the most important functions that a model has.

5. Early voting isn’t predictive (except maybe in Nevada)

In the run-up to Election Day, reports of a decidedly pro-Clinton early vote in Florida and North Carolina were everywhere. Those reports came at the same time that Trump was climbing in the polls in both states. In the end, Trump carried both Florida and North Carolina by over 150,000 votes each.

This was not the first time that Democrats mistakenly pointed to early voting as a sign that their candidates were doing well. Back in 2014, the party did the same thing in Colorado, Georgia and Iowa, citing the early vote as proof that Democratic candidates were going to do well once all the votes were counted. In the end, they lost all three of those Senate races, just as the polling predicted they would.

The fact is that if polling is done correctly, it takes early voting into account. Most polls ask respondents if they have voted early, and if they have, they are marked as likely voters. Polls also have the advantage of asking everyone, not just early voters, who they are going to vote for. That’s important because people who vote early are a self-selected group whose political opinions may differ from those of people who vote on Election Day.

That’s why FiveThirtyEight doesn’t put early voting into our models. It just isn’t predictive in most cases. The exception to this rule may be Nevada. That’s because Nevada kept its voting laws the same through a number of election cycles and it releases party-registration data on its early voters, so we can make apples-to-apples comparisons to prior years.

The polls are often systematically off, underestimating one party across the board, but it’s really hard to predict in which direction they’ll be off. In 2012, polls underestimated Democrats. But in 2014 and 2016, they underestimated Republicans. It could be that early voting will look good for Republicans in 2018, but that the polls will underestimate Democrats again.

Harry Enten is a senior political writer and analyst for FiveThirtyEight.