Who’s polling who?
To the surprise of precisely no one, Sen. Bernie Sanders has been announced as the winner of TIME magazine’s TIME 100 reader poll.
The online poll has become an annual target for activists, trolling pranksters, and fans of Korean R&B. While the digitally active base of Sanders propelled the septuagenarian politician to the top of the poll this year, previous winners include Indian anti-corruptionist Arvind Kejriwal, 4chan founder Christopher “moot” Poole, and K-pop sensation Rain. On at least two occasions—once in 2009 and again in 2012—hackers have manipulated the results so the first name of each candidate spells out an obscene message or inside joke.
The struggles of the TIME 100 poll are emblematic of these sorts of online polls. Typically meant as a form of reader engagement more than any scientific measurement, news outlets are often shocked—shocked, I tell you—to find out their survey is easily brigaded or, in the case of the TIME 100, straight up manipulated. Overall, they have become a silly, embarrassing practice in self-congratulation and data collection for both the media entities who offer them and the users who enjoy them.
The struggles of the TIME 100 poll are emblematic of these sorts of online polls.
More than simply a nuisance, however, they represent a deeper problem in measuring people’s opinions and behavior in a way that is both representative and useful.
The easiest example of this might be the post-debate polls offered by websites and news outlets to allow people to vote for which presidential candidate they feel “won” the debate. The Drudge Report, for example, has held such a survey of its largely conservative readers after each of the 12 GOP debates. Thus far, Donald Trump has walked away cleanly with 10 out of 11 victories (Trump did not attend the Jan. 28 debate held by Fox News) in the Drudge poll. His “victory” is touted proudly by Drudge’s fellow pro-Trump outlets like Breitbart, Infowars, and even Trump himself as true a measure of public opinion as an actual poll.
One might consider The Drudge Report as a representative sample of conservatives—the site, managed solely by journalist Matt Drudge, has quietly remained one of the most influential news publications and one that regularly sways the national conversation. Despite it’s stoic design philosophy and abject bias, The Drudge Report collects nearly 18 million unique monthly visitors. If any media outlet—especially a digitally based one—stands a chance of accurately tracking online opinions of Republicans, it’s likely Drudge.
Except, as in the Time 100 poll, Drudge’s polls are too easily bombarded by people whose ideological interest is unlike that of his GOP rivals, a portion of Trump’s base is as youthful and tech-savvy as the same troops that show up to support Sanders. A poll done by Harvard University’s Kennedy School—an actual polling outlet with representative samples and weights against bias—finds Trump winning among millennials against his rivals by similar margins Sanders has topped Clinton.
Hence why Sanders and Trump have handily won most debate polls held by any outlet. And more than the people voting in them take their result seriously. According to Matt Drudge himself, one supporter of Sen. Ted Cruz (or at least one opponent to Donald Trump) might have created a voting bot to manipulate the results of the poll in favor of Cruz. Trump has seriously offered his performance in these polls as true measures of his support in response to perceived slights or bias.
As an actual measure of data, these polls are absolutely 100 percent meaningless. They don’t accurately measure opinions of voters. They don’t even accurately measure opinions of the visitors to the sites hosting them. All they can tell anyone is how people voted in them. Polls, to be considered even close to accurate, must include a litany of measures to prevent the kind of bias seen in the Drudge poll or the flooding and hacking seen in the TIME poll. The National Council on Public Polls offers no less than 20 considerations for journalists and consumers when reading polls, from who funded the poll and the sample population size down to the sampling errors and order of the questions asked. Online polls fulfill almost zero of these obligations.
Their ineffectiveness and vulnerability is one reason they are a poor option for actual pollsters in an age when the public opinion poll could very well be dying. According to historian Jill Lepore’s damning New Yorker piece about the state of modern polling, pollsters are struggling to get an increasingly-mobile world to respond to polls (mostly performed on landline phones). “The promise of this work is that the sample is exquisitely representative,” writes Lepore. “But the lower the response rate the harder and more expensive it becomes to realize that promise, which requires both calling many more people and trying to correct for ‘non-response bias’ by giving greater weight to the answers of people from demographic groups that are less likely to respond. ”
Nate Silver, perhaps the world’s first and only “celebrity pollster,” agrees. “It’s quite challenging to conduct a poll now when most people are not picking up a random stranger’s telephone call answering a survey,” he said. Online polls might solve the sample size problem—hundreds of thousands of people vote in each Drudge poll—but they lack the randomness to serve as an accurate representation of any meaningful population. “Online polls don’t use probability, which means there’s no way—unless you’re the NSA—of randomly picking someone online,” Silver says.
So why do them? Why would media outlets continue a practice that is functionally meaningless? The first and most obvious is traffic—giving supporters of candidates a chance to actually support their candidate in a way that makes them feel useful and heard can be a huge draw, much more than simply an article about that candidate could do. Another purpose, however, is the sheer amount of specified data such surveys allow outlets and other companies to collect. After all, marketers and data analysts use surveys all the time to gather measures of how certain populations might respond to a particular product or ad campaign.
Perhaps more than any other media outlet, Buzzfeed has this down to a science. Each of those quizzes promising to reveal which Harry Potter character you are or what kind of grandparent you’ll be is harnessed by Buzzfeed’s impressive data analytics arm to give a micromanaged view of its audience to marketers. They don’t just make data collection easy—they make it fun.
There’s little reason to suspect polls like that of the TIME 100 aren’t used for similar purposes. So not only do such surveys attract hordes of activated users, they encourage those users to hand over information about their positions or stances a news site might otherwise struggle to glean.
Online polls are useless when projecting onto any other kind of audience, no matter the overlap. But as measures of activation and devotion for a candidate, position, or even a pop star, they might just tell marketers all they need to hear.
Gillian Branstetter is a social commentator with a focus on the intersection of technology, security, and politics. Her work has appeared in the Washington Post, Business Insider, Salon, the Week, and xoJane. She attended Pennsylvania State University. Follow her on Twitter @GillBranstetter.
Pure, uncut internet. Straight to your inbox.