What You Need To Know About Early Polls And How To Read Them
The early 2016 presidential polls are flying, which means the complaining about polls is in full swing, too.
"You guys should know by now that the Monmouth University poll was created just to aggravate me," Chris Christie told The Washington Post recently about a New Jersey-based poll that showed him with 2 percent support. "There couldn't be a less objective pollster about Chris Christie in America."
It's not just on the crowded GOP side. Democrats have pushed back against a Quinnipiac poll showing Hillary Clinton with high unfavorability ratings in key swing states.
Polls always have weaknesses. Early polls can be worse. And they're all spun every which way by every pundit — and politician — with vocal cords. If you're baffled by what to believe in polls, here's a helpful rundown of what you need to know without being misled.
1. Polls are not predictive
Write it on a Post-it and stick it on your computer monitor or iPhone or wherever you read polling stories. Scribble it on your hand. Tattoo it on your body somewhere: The polls you read about in the news right now can't — and aren't intended to — tell you who will be elected president in 2016.
Consider this 2011 blog post from George Washington University political scientist (and early polling critic) John Sides, in which he found that the further out from an election a poll is, the less likely respondents are to choose the eventual nominee.
At this point in 2011, Texas Gov. Rick Perry was winning in some Republican polls. Clinton and Republican Rudy Giuliani were leading in midsummer 2007 (though Fred Thompson and Mitt Romney were putting in a strong showing on the GOP side, as were John Edwards and some guy named Barack Obama for the Democrats). In 2003, Howard Dean looked dominant in midyear polls.
That might sound surprising now, but think back also to Obama's effective grass-roots campaigning in 2008 or Dean's criticisms of the 2004 Iowa caucuses, followed by a third-place finish and the shriek heard 'round the world. All sorts of things could (and will) happen between now and Election Day. Polls are also not the only things that can indicate what's happening in a campaign or point to the potential nominee — fundraising, establishment party support and endorsements, for example, all play huge roles.
But this is important: Pollsters aren't trying to tell you the future.
"The notion that a poll will tell you what's going to happen in 2016 is a ludicrous notion," said Peter Brown, assistant director of the Quinnipiac University Polling Institute. "Polls will tell you what Republican opinion is today. That's what they do."
Polls are a snapshot, not a crystal ball. So, for example, if Bobby Jindal (currently polling at around 1 percent) ends up being the GOP nominee in 2016, it won't mean that the current polls got it wrong. It will mean that he wasn't registering with voters early on and then something happened to boost his stock.
But this brings us to another point.
2. What's in a name? A lot, particularly when it comes to early polls
"I think [early polls] measure name recognition," said Clifford Zukin, professor of public policy at Rutgers University and member of the executive council of the American Association for Public Opinion Research. "And that's something, but is name recognition really the way you want to choose a presidential candidate?"
It's true that early polls are largely measuring who's best known. The Washington Post's Philip Bump showed earlier this year a marked correlation between support and name recognition. So when Clinton or Donald Trump leads in a particular poll right now, it doesn't necessarily mean the respondents were all — or even largely — making a well-informed decision on policies and leadership styles; it could mean Clinton and Trump both have really, really high name recognition.
3. Be careful with those hypothetical matchups
There's another aspect of early polls to watch out for: the ones that try to put nominees in head-to-head matchups.
That might tell you a little bit about how the election would turn out if it were held today, but unless you're reading this on Nov. 8, 2016, the election is not today.
And by then, the Republican nominee will no longer be fighting off assaults from 16 other Republicans, and both the Republican and the Democrat will have a more clearly formed message and policy stances. Bottom line: Again, see point No. 1.
4. But polls do tell you something, like roughly where the candidates stand — and that's important
"Polls are not very useful right now except for telling you about tiers of candidates," Zukin, the Rutgers professor, said. "They really tell you that Trump has more support than Christie and Rand Paul. It really tells you about tier of public visibility."
So the three GOP candidates who are regularly polling in double digits make a clear tier right now — but in many polls, they are so close to each other that it's hard to tell who exactly is leading whom in any given poll, in any given state. Likewise, the mess of lower-rated candidates is so close in polling that it's hard to pluck out who, exactly, is leading whom.
No, that doesn't tell you voters' substantive views, but it does suggest that to get people's attention, Sen. Lindsey Graham, R-S.C., has further to go than Scott Walker, for example. Or, despite his massive crowds, it suggests that Sen. Bernie Sanders, I-Vt., has enormous ground to make up to catch Clinton, who has been in a tier of her own thus far on the Democratic side.
They also tell you a bit about who could get a boost in the race, as polls help some donors figure out whom they want to give to. And those donations help candidates raise their profiles.
5. Two polls (or three or five or nine) are better than one
A single poll is a snapshot. But string enough snapshots together and you start to get a real story.
"If you're seeing a change in the number of people who are supporting Hillary Clinton, if that number has decreased, and then you go and look at what's happening with her favorable and unfavorable ratings, something has happened to change that gain that is showing up now," explains J. Ann Selzer, president of Selzer & Co., a polling firm in Des Moines, Iowa. Her survey, often conducted for the Des Moines Register, is generally considered the "gold standard" when it comes to the very-difficult-to-poll Iowa caucusgoers.
To put Selzer's remarks in other words — context matters. One poll can be a fluke (otherwise called an outlier). Even two can. But should six consecutive polls of the same population show Clinton's support falling, that isn't something you (or the Clinton camp) can ignore. It would suggest that something she's doing is upsetting the base.
Or consider the close scrutiny that Trump's poll numbers have received since his remarks about whether John McCain is a war hero. If that had been the point when his support started slipping (it didn't), it would have sent the message that that — and not, for example, his remarks about immigrants in the U.S. illegally — is what irked his base enough to go find another candidate.
But if he maintains support or continues to gain in the polls, it would also tell you that for his supporters, the McCain comments obviously didn't make them think less of him.
6. Details, details, details
Margins of error aren't just the stuff of statistics classes; they tell you when one candidate is meaningfully ahead of another one. If the pollster's reported margin of error is plus or minus 3 percentage points and Candidate A is 8 percentage points ahead of Candidate B, Candidate A is clearly winning. But if A is leading by 4 points — less than twice the margin of error — Candidate A has a lead, but there's still the possibility B could be instead.
Of course, this can all get even trickier quickly. In polls with more than two candidates, or in which the candidates aren't polling at around 50 percent, the margins of error can shrink ... sometimes substantially. (For a more detailed explanation of all this, read the National Council on Public Polls' excellent explanation of sampling error in its guide for journalists.)
It's also important to pay attention to who's being polled. If a poll of likely primary voters put Trump far ahead last week, for example, while a poll of registered and leaning Republicans put Walker ahead this week, it's not necessarily that voters changed their minds over the course of seven days. It could be that the bigger pool of Republicans and Republican leaners have different preferences from those of primary voters. (Likely voter models are a whole other bag of worms of weighting, matching and concocting.)
7. Polling, in general, is getting worse
Cellphones have made nearly everyone's lives easier — everyone except pollsters, that is. They can call your landline (if you have one) on an automatic dialer, but they have to dial your cellphone by hand. (In other words, and in polling terms — only "live caller" surveys are legally allowed to call cellphones.)
Given the fast-growing number of cellphone-only homes (and the demography of who uses them — younger voters), that makes accurate polling a much more labor-intensive and expensive process, as calling a representative sample of landlines just takes that much more work.
What's more, response rates — how many people agree to be polled — is way down from 30 years ago. All of that has made polls — early or not — less accurate with every successive election in recent years, Zukin points out.
The biggest question in polling right now is what to do about it. Online polls are the future, Zukin adds, but, for now, they're just not that good of an option. They tend to be less reliable than phone polls, because they can't get a random sample — the people who take them are a self-selecting group.
There's one added problem here: herding. If polls are getting less reliable, pollsters might start glancing at each other's answer sheets to make sure their results aren't too far off from the norm. This is called "herding" in public opinion research, and it means that polls could all be wrong together. There's more downside in having your poll being way off and wrong than in being way off and looking like everyone else's. And it's already happening, as FiveThirtyEight's Nate Silver wrote after the 2014 midterm election.
"We won't figure it out for 2016," Zukin said. "There's a better chance of not being right this year than there was four or eight years ago."
So as polling continues to grow less reliable (before it can get better), it means one more reason to look askance at the latest numbers. They're not meaningless, but you have to take them with a few hefty grains of salt.
Copyright 2020 NPR. To see more, visit https://www.npr.org.