BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

What You Need To Know About 2016 Election Polls

Following
This article is more than 7 years old.

Paying attention to the 2016 election can sometimes give you a feeling of whiplash -- with the days bringing breathless reports of poll swings, poll declines, talk of momentum and more. But despite the excitement, when you look at poll aggregators -- organizations that take a look at the available polls and average them together -- the race has been remarkable steady, with Clinton maintaining a pretty constant edge over Trump over the course of the year. You can see an example of this in a chart from Pollster, below.

But are the polls reliable? How should average voters understand polls and use them to understand the election? For answers to these questions, I turned to Dr. Sam Wang. Dr. Wang is a physicist, but he also manages the Princeton Election Consortium, which takes data from polling organizations across the country to build a statistical model of election outcomes, and has been doing so since 2004. Here's a few tidbits you can take out of the conversation to understand polls as we near the end of the election.

Individual Polls Are Often Incorrect, But The Combination Of Polls Is Reliable

The nature of polling means that any given poll is going to be off by at least a little bit, just given the math, Wang explained to me.

"If you sample say, 800 people in a race that’s really 50-50 and trying to get a feel for what voters will do, it's like tossing a coin 800 times," Wang said. "If coin is fair you’ll get about 400 heads and 400 tails. But it’d be surprising to get exactly that. You'll likely get something like 390 heads and 410 tails. It's impossible to get within better than a few percentage points, because you can’t interview everybody."

Another aspect of the sampling, too, is that pollsters have to apply their professional judgement. Since different demographic groups have different likelihoods of answering the phone, and that likelihood is different than how likely they are to vote, pollsters have to apply judgement and math to get a good snapshot of the electorate.

"They're all working to get honest results but they're applying their own judgement," says Wang.

By the same token, though, the aggregate of polls can be pretty close to the real result. To continue with the coin flipping metaphor, if you have three different sets of coin tosses that resulted in 395 heads, 405 heads and 398 heads, the median result there is 398 heads -- pretty close to the actual result of 400 in terms of the coin flip odds, and closer than the two outliers.

So the key for predicting election results is to look at multiple polls for the same area and take the median value. That said, it's important to take the median, which is the middle value of a set of numbers, and not the average -- which gives much heavier weight to outliers.

"For example, right now in Michigan there's a handful of recent polls that show Clinton ahead by a few points," said Wang. "But there's one that shows her leading by 20 points -- that's way off. If you take the median of all the polls, you have Clinton ahead by a few points, but if you average them, it shows a bigger lead than she likely has."

Pay More Attention To State Polls Than National Polls, Because The Latter Can Have Wilder Swings

Another thing that Wang cautions voters paying attention to the election about is national polls. That's both because presidential elections are decided by the electoral college, and also because national polls can produce a lot of wild swings just by their nature. For example, three days ago an ABC-Washington Post poll garnered a lot of attention on social media because a few days earlier the same tracking poll had showed Clinton with a 6-point lead. (The same poll today gives Secretary Clinton a 3-point lead.)

"About 1 in 20 times you’ll get a swing like that," said Wang. "With about 20 organizations running polls, you'll see an outlier every day -- and outliers tend to get more attention."

Because of this, Wang encourages voters who want to stay informed to the median of poll results in a few key states rather than national polling. "A few key states are absolutely necessary for Trump. If he doesn’t win those it’s over."

A Swing Towards Trump Right Now Isn't Momentum -- It's "Regression To The Mean"

Over the past few days, there have been a few reports showing that in the aggregate, Mr. Trump has gained 2-3 percentage points of support in the election. For supporters of Mr. Trump, this has been seen as a sign of "momentum" for him to overtake Secretary Clinton on Tuesday. But Wang urges caution for both Trump and Clinton supporters interpreting this swing.

"One thing that’s been apparent is that a major feature of voter opinion for last five elections -- this is the sixth -- is that voters have become entrenched," says Wang. "The movement of voter opinion has been within a narrow range. In finance and other types of statistical analysis, we call this kind of movement a 'regression to the mean.' It happened in 2008, in 2012 and it's happening this year. When things go too far in one direction, they'll start to head back to a midpoint. Clinton is now at the low end of where she’s been this season. But if the regression to the mean holds, we should see a little movement back to Clinton. But we'll see."

The Brexit Vote Results Were A Pundit Failure, Not A Polling Failure

The specter of Britain's vote to exit the European Union in June has cast some uncertainty on this election, particularly when it comes to the reliability of polling. But there are a few reasons why Britain's "Leave" vote shouldn't color our thinking about the presidential election, says Wang.

"If you look at Brexit coverage, it was really pundit failure," says Wang. "Commentators were saying there was no possible way for Leave to win. But if you look at the data -- and I did this the day before -- it showed Remain as only very slightly ahead. I said then it was too close to call and could go either way. That’s what the data was saying. It's not data’s fault that pundits said Remain was going to win."

The other thing to consider, says Wang, is that Brexit was a one-off event, but there is a lot of experience when it comes to polling U.S. presidential elections.

"Presidential state polls are the gold standard of polling data around the world," Wang said. "U.S. pollsters are constantly experimenting with new methods. I'd be very surprised if there’s a community-wide error of more than 1-2 points."

The Hardest Elections To Predict In 2016 Are For Senate Seats, Not President

While it's not over until it's over, Wang notes that Hillary Clinton is "heavily favored to win" the presidential election. What's really interesting, he said, are Senate elections, which are very close and leave the question of which party will control the Senate in 2017 very much up in the air.

"Six Senate races are within one percentage point," he said. "If I were a voter I'd be paying attention to the Senate."

Follow me on Twitter or LinkedInCheck out my websiteSend me a secure tip