Let’s rewind in time back to the summer of 2015 — when Donald Trump declared his candidacy to become the President of the United States. I recall many folks including myself thinking, “Donald Trump for President? This must be a joke, there’s virtually no chance that somebody like Trump can become president. For Trump to even become the Republican nominee, he’d have to defeat not only a number of other Republicans, but he’d also have to overcome scrutiny from Fox News, which doesn’t think all that favorably of him, and there are just too many hurdles to clear for a Trump nomination to happen.” And then, as we’re all aware, Donald Trump jumped all of those hurdles and became the Republican nominee for the U.S Presidency, with only one more hurdle in his path, Hillary Clinton. And then, as recent as the very evening of the general election, many like me believed, “It’s improbable that Trump will win the White House. Just about all of the polls are showing Clinton holding a formidable lead, and too many states would have to turn in Trump’s favor for him to win.” Well, here we sit today, and many of us are surprised by the fact President-Elect Trump will occupy the Oval Office come January.
So now we’re asking ourselves, what happened, and how did Trump overcome these inconceivable odds? Earlier this year, FocusVision conducted a study about how people felt “emotionally” about Trump and Clinton, and overall, the electorate was feeling pretty negative about both candidates. Clinton scored higher than Trump on positive emotions such as joy and trust, and lower than Trump on negative emotions such as anger, disgust, fear, and sadness. The results of our study surely could’ve been regarded by some as a precursor to a Clinton victory, given that people clearly felt emotionally less negative about Clinton than Trump. Our study was fielded back in August, so we don’t know if the emotional intensity towards the candidates shifted since then, given that emotions are best tracked over time. But perhaps the emotional scores we saw were, in fact, predictive of the outcome in this year’s election. While Clinton did not win the Electoral College, she’s projected to win the popular vote by one or two percentage points.
Nevertheless, the lurking question remains: how did the polls at the state level get things so wrong? The answer is, just like with any other quantitative research study, obtaining a sample that mirrors the population proves to be very difficult. The ugly truth is that all polls and quantitative research studies have at least some level of sampling error. So what exactly was missed that led to sampling errors in the case of this election? While it may take months to fully understand what happened, there are some factors that can be identified.
- Voter turnout once again proved to be unpredictable. Projections were skewed toward Clinton because fewer Democrats than expected made it into the voting booth.
- Today’s poll-based models attempt to reduce error by combining a number of different polls. In states with fewer polls available, there is more room for error.
- Trump outperformed in states with higher numbers of white voters without college degrees. State polls underestimated Trump’s support within that demographic.
- While this last point is hard to prove from a quantitative standpoint, some pollsters are suggesting that female Trump supporters may have been uncomfortable expressing their position. It would be interesting to know if projections would have been different if pollsters employed implicit measurement techniques.
In today’s world, people expect spot-on perfect accuracy from polls, which seems to be an unrealistic expectation. In the 2012 election, some pollsters correctly predicted the winner in all 50 states, which is in part why many, including those within the media, put so much faith in the polls during the 2016 election cycle. The reality is that a polling error of two to three percentage points is not all that uncommon; in fact, the national polls have historically missed the target by nearly two percentage points. In this year’s election, two percentage points turned out to be the difference between winning and losing. To provide an idea of just how big of a difference two percentage points can make, Nate Silver of FiveThirtyEight asked, “What would have happened if just 1 out of every 100 voters shifted from Trump to Clinton? That would have produced a net shift of 2 percentage points in Clinton’s direction. Michigan, Wisconsin, Pennsylvania and Florida would have flipped back to Clinton, giving her a total of 307 electoral votes.”
This year’s election showcases the difficulties researchers face when trying to match their sample to a target population. It’s hard to think through all the numerous variables that can make up a study’s sample, and if just one small variable is missed (such as white voters without college degrees in Wisconsin), the numbers can be thrown off. We can still trust polls, but we do need to readjust our expectations and keep in mind that all polls contain error and a small percentage is normal. As we saw in this election, political research often requires pinpoint accuracy in order to predict the outcome of an election. While pinpoint accuracy may be an unrealistic goal, research in all fields will suffer if a sample design is poorly crafted. If a political campaign, company or any organization is to going to make sound decisions, it’s imperative that the right people are interviewed.