The first article in an occasional series about prediction.
Though it seems ages ago, this month, Senator Ted Cruz ended his presidential campaign in the wake of a big loss to Donald Trump in the Indiana Republican primary. Coincidentally (or not), May 3rd’s New York Times was happy to predict such an outcome on the day of voting, on the front page, above the fold. Just a few weeks earlier, the same newspaper proclaimed with near certainty that there would be a convention floor fight for the nomination. A few months before that, nearly everyone was agreed a Donald Trump candidacy seemed laughable.
Going back to April of 2015, before many of the candidates had even officially entered the presidential race, the Times had already decided that Hillary Clinton had sewn up the primary. “It may seem too early for such bold pronouncements, with nine months until the Iowa caucuses. But presidential primaries, like presidential general elections, have underlying fundamentals that help determine from the start who will win.”
Hillary Clinton is indeed likely to win the Democratic primary, but at the point of the Times’s “bold pronouncement” her main opponent was not even on anyone’s radar screen. CNN polled Bernie Sanders’s support at 3% once he officially tossed his hat in the ring on April 30th of last year, about two weeks after that article appeared. As this article is being posted, Sanders has won 20 of 44 states.
But consider one implication of these sometimes incorrect polls and prognostications. They suggest that your vote—indeed everyone’s vote—is irrelevant, and that only particular candidates should even bother to run for high office. If you were on the fence about schlepping down to the precinct hall, the subtext is that you should stay in your PJs. This sentiment is one of the subtler, and more sinister, effects of one our newest technologically-driven obsessions: prediction.
So what’s wrong with prediction? Isn’t it good to be prepared? Like many things prediction, in moderation, seems harmless enough, even beneficial. Why not grab an umbrella for later if the forecast calls for rain? Why not plan hiring for your business based on the economic outlook in the next quarter?
What I would argue is that, due to our advances in data-gathering and modeling, helped along by media invested in generating sales and clicks, we have become a society obsessed with predictions. We overestimate their accuracy, overvalue their importance, and underrate the impact of the prediction itself on our behavior and experience. It’s at this point that forecasting begins to do harm.
Outside of sports or awards shows, where the prediction game is mostly for entertainment purposes, there seem to be five main areas where prediction and mass culture overlap: economics, politics, security, health care, and the weather. I would suggest that in most of these areas, our prediction obsession is hurting us more than it is helping, sometimes to the point of undermining freedom and democracy. Democracy, in particular, is at stake in one our favorite predictive pastimes and the one I’m examining here: election polling.
People have been handicapping political races for eons. The 1824 presidential election was the first recorded instance of exit polling being used (PBS.org). George Gallup began his public opinion research in 1935. But, as the Pew Research Center describes, even through the 1960s most polls were conducted with in-person interviews by just two major organizations, a process that might take a month, laughably slow in today’s media climate. Widespread adoption of cheap telephone service (and increased computing power) made polling a lot easier. By the late 70s and early 80s media companies like CBS and the Associated Press began launching their own polling services; it’s been runaway escalation ever since.
So political prediction is not new. What is new is the extreme data-crunching that tracks county-by-county voter preferences, the meteoric rise of quantitative analysts like Nate Silver of 538.org, and the belief that our data in this age of information is (or soon will be) nearly infallible. This has resulted not only in an increasing number of polls but an increasing reliance on polls as the salient measure of a political campaign, by both the media and the candidates themselves. That’s despite a long record of spectacular failures.
Political prediction has long been a risky business. A victorious Harry Truman holding up a copy of The Chicago Tribune with the infamous headline “Dewey Defeats Truman” following the 1948 election is perhaps the single most iconic moment of the pollsters getting it wrong. But it’s hardly the only one.
In 1994, it seemed clear Republicans would do pretty well in the first midterm elections of the Clinton presidency, with experts predicting a gain of 20 seats or more, according to the New York Times. But that failed to prepare for the absolute shock the country experienced when they flipped 54 House seats and 9 Senate seats, taking all of Congress for the first time in 40 years, in what became known as the “Republican Revolution.”
Of course, you need look no further than this year to see what will certainly rank as one of the prognosticators’ biggest failures ever: the inability to imagine Donald Trump winning much more than a state or two, let alone a major party nomination. The media has done much hand-wringing over its blindness to this outcome, but they did none over offering one of the wealthiest men in America a near endless supply of free media.
Nor has that failure modified their behavior. On May 5th, two days after Trump cleared the field, and before Hillary Clinton had even won her party’s nomination, The Boston Globe’s front page was filled with a giant graphic of the new polling between Clinton and Trump—six months before voters participate in the only poll that really matters, the election.
Ironically, technology may have inadvertently made polling less accurate than it used to be. That’s because pollsters are legally prohibited from using auto-dialers on cell phones, unlike landlines. With response rates dropping precipitously, this means pollsters have to hand-dial extraordinary numbers of people, which is prohibitively expensive for most operations. During the 2008 and 2012 elections, Nate Silver vaulted into the upper echelons of political punditry with a big-data approach to the presidential vote that turned out to be incredibly accurate. Nonetheless, our new statistical guru completely whiffed on Trump, demonstrating the fallibility of even massive number crunching when your data is unreliable.
Political prediction, when it is inaccurate, is at the very least a waste of time. But even if it is accurate, what then? Outside of the candidates themselves, their families, and the prospective cabinet (who might need to make moving or employment arrangements), knowing the outcome of an election in advance benefits no one. That might be fine if there was also no harm. But is there?
I would argue there is harm. In the first place, as studies of other cultures tell us, and quantum physics echoes from its weird subatomic world, you can’t observe something without changing it. In the case of polling, there are at least four dangers that flow from this. The first is probably the most disturbing.
Polls Affect Opinion
There have a been a number of studies examining how and to what degree polls can affect opinions. One of the more recent is a paper by David Rothschild and Neil Malhotra that was published in 2014 in Research and Politics. They had subjects record their support for three policy proposals and then exposed them to “polls” about those issues (really just random numbers between 20 and 80 percent). After a few minutes of distracting questions, subjects were asked to re-register their support. Even after one exposure and a mere five to ten minutes of distraction, the fake polls changed individual opinion. When averaged, a 60 percent change in fake polls (the full gamut between 20 and 80 percent) corresponded to an 8 percent change in individual support.
Over time, and with repetition, Rothschild and Malhotra believe that “polls, by directly influencing individual-level support for policies, can be self-fulfilling prophecies and produce opinion cascades.” In particular, those people in the center of the distribution, that is to say, those who are 50% in favor of any given policy, move towards the extremes after exposure to polling.
That was a policy study, but the effect has also shown up in candidate studies, such as “Effects of Poll Reports on Voter Preferences,” led by UCLA professor Albert Mehrabian. That one, while small (123 participants), had the benefit of being conducted during an actual campaign for president. Early in the campaign, Republican voters in California were exposed to bogus poll numbers showing either Bob Dole or Steve Forbes leading by 47% to 19%, and then asked who they would support. Mehrabian found, unsurprisingly, “Voters favored Dole over Forbes to a greater extent when they were given poll data showing Dole leading Forbes,” than the reverse. The so-called bandwagon effect had created a positive and significant change in which “6% of the variance in votes was due to conformity to peer-group preferences (the bogus polls).”
You can see opinion cascades in this year’s Republican primaries, where Donald Trump was consistently unable to draw more than half the vote for 33 consecutive contests until April 19th, when he suddenly won the next six states with large outright majorities. Would that have happened if all states voted simultaneously? It strains the imagination.
It’s not that talking with your friends and neighbors about political issues is a bad idea; it’s a great idea. But that’s different from being constantly exposed to polls. We need space to think and to form our own opinions. Polling, rather than giving us the counsel of a few trusted advisers at a few specific times of our choosing, leads to constant social pressure and a mob mentality. My guess is you’d vote for someone who aligns with your values if you knew nothing about their chances. But what if they polled at 1%? Theoretically, it shouldn’t matter. Your values are your values, after all. But these external cues have snowballing effects, and people will tell you you’re “wasting your vote”—as if voting for anyone other than the winner is a waste.
Why vote at all, when the pundits are telling you the outcome is a foregone conclusion? There may be many reasons why voter turnout is so abysmally low in the US, but making everyone believe they already know the results beforehand cannot be helping. According to the Pew Research Center, turnout for our last presidential election was 53.6% of the voting age population, putting us 31st out of 34 countries listed. The primaries are even worse. This year (with data through March 6th) has seen only 29% of voting age population turn out. That’s a near-record high.
The way US primaries are conducted themselves is a terrible idea, from the perspective of allowing voters to make independent choices. Each state’s result influences the next state’s voters, as well as media coverage and donations in that state. During the general election, news organizations deliberately wait to call states for particular candidates until everyone in that state has voted. The three-hour delay between East and West Coasts does create a minor issue, but it’s hardly the problem faced by, for example, the District of Columbia’s Democratic primary. That will be held on June 14th, a week after all the other delegates have been awarded and almost surely after the nominee is known. This is particularly problematic because different states have very different populations, insuring our nominees are heavily tilted towards the desires of the early-voting states. Consider that while at least 17 candidates ran for the Republican nomination this cycle, 11 states, including New York and Pennsylvania, only got the opportunity to vote for three, and nine states, including California and New Jersey, will have one choice: Donald Trump.
The goal of the modern campaign, it seems, is not to win the day with a better message or a better vision for the entire country, but rather to know precisely which winnable micro-area needs a few extra canvassers, or how many delegates accrue from which caucuses. President Obama was praised in many quarters for his savvy number-crunching team during the 2008 primaries, when he managed to vanquish Hilary Clinton in part by winning smaller caucus states that awarded delegates proportionally.
This year, we were treated to stories about Ted Cruz’s superior efforts at stacking the delegates in his favor, so that he might triumph in the event of a brokered convention. That is hardly an inspiring reason to participate in the electoral process.
What’s interesting about this is that over time we’ve taken to using the campaign itself as a measure of a person’s fitness for office. Having great ideas, pushing policies that benefit the whole of society, being able to both lead and build consensus—these skills have very little to do with raising money, conducting polls, or eating corn dogs at state fairs. “She ran a great campaign,” or “He ran a terrible campaign,” are the types of phrases you hear pundits use to explain electoral success or failure. Not “I guess 55% of voters didn’t want John Kerry to be president,” which might be a more accurate, if tautological, summation of the 2004 race, for example. One might argue that Obama ran a terrible campaign in 2012, yet he still won re-election. Where does that leave the analysts?
Superficial Media Coverage
Election polling was of questionable news value even in the horse-and-buggy era. In the information age, it’s become a gaudy and superficial replacement for substantive reporting, one that is nearly impossible to avoid. The difference is not only the increase in communication speed and computing power that turn polls into instant feedback loops, but also in the media’s obsession with the polls and the sheer number of polls that are conducted.
After fielding criticism on this point, the New York Times public editor Margaret Sullivan decided to look into the matter. She discovered that in a mere two weeks this past February, 180 of a stunning 234 political stories were about the horse race.
Not only do such stories divert attention from issues, they also divert resources from covering other—undoubtedly more important—issues. Imagine the hours devoted to those 180 stories instead being used to delve into, say, a deep examination of policies that both liberals and conservatives agree on and might be able to achieve together. The mind reels.
Suppression of Minority Viewpoints
It may be that a stump speech is repeated 100 times and therefore no longer newsworthy, that there is really nothing further to say about Hillary Clinton’s positions on issues. But there is certainly plenty to say about the other candidates running—like the Libertarian Party’s Gary Johnson, the Green Party’s Jill Stein, or even former Democratic candidate Lawrence Lessig—all of whom the media has taken pains to ignore. Instead, they use the polls as a measure for whom to cover and, ominously, as a hurdle to get into debates. Too many candidates running in the Republican primaries resulted in the famous “undercard” debates, with all the low-polling candidates shunted off to a electoral version of the kids’ table. The untaken solution to this problem was always quite simple: assign the candidates to the separate debates randomly, and hold each one in prime time by itself.
Eliminating the low-polling candidates from debates ensures a feedback loop of declining support. How can someone raise their standing in the polls without being able to reach the audience that a debate provides? It may very well be that many of these candidates are terrible (Lincoln Chaffee comes to mind), but this is not for the media to decide. Getting on the presidential ballot in most or all of the states is no mean feat; anyone who does it deserves an opportunity to at least make her or his case to the public. They may also offer interesting viewpoints otherwise unavailable in our pre-packaged age. Excluding these people only furthers the nation’s alienation from the political process, making it seemed rigged.
Ultimately it must be asked what service, excluding entertainment, is provided to the public by prediction in the political arena? Even presuming it were accurate, I honestly can think of none; the results of the actual election will give you all the information you need. Yet the harms seem evident, significant, and damaging to democracy. And the prediction is often inaccurate to boot.
What is the point of a political race in the first place? If we’re serious about it, it is an opportunity to learn about candidates offering competing visions for governance and make an informed choice about which of these candidates best serves the interests of the general population. Even in a loss, a candidate can shift the political debate or offer novel solutions that may later influence policy. Ross Perot’s concern with deficit reduction likely contributed to the balanced budgets of the 1990s, even though he only garnered 18.9% of the vote in the 1992 presidential election.
Even famed Republican pollster Frank Luntz has complained about the ultimate failure of polls to genuinely help his clients understand the feeling of the electorate. “I agree with your theme,” he replied, when I told him about this article. “The explosion of polls is why I love to moderate 30-people dial focus groups. It tells me why and gives me a level of intensity that polls don’t do.”
Luntz may be genuinely interested in reading the electorate, but in political contests, the media use polls to predict winners and losers. This desire to know the outcome of something before it happens is a curious one. In the first place, it’s literally impossible; any number of things—think Robert Kennedy’s assassination—could change results in an instant.
I’m guessing the drive to predict stems from an inherent anxiety about not knowing. The real truth is the future is murky and that makes some people very uncomfortable. Nonetheless, it’s in this state of not-knowing that we can learn and experience the present. That is the ultimate casualty of the prediction game. By attempting to divine what might happen tomorrow, we lose the ability to appreciate what is actually happening right now.
In the case of politics, the repercussions are fairly serious. We use polling to ask endless questions about who we think is “winning” instead of learning candidates’ positions. We use polling to mock and marginalize legitimate points of view. We use polling to bar entry to televised debates, by setting a threshold candidates must meet before they can appear, a vicious circle designed to keep current party structures intact, by what is supposed to be an impartial media. We use polling, wittingly or not, to manipulate political opinion. And, worst of all, we use polling indirectly to discourage voters, by telling them the outcome is a foregone conclusion.
If this all sounds rather undemocratic, it is. And although it’s long been a sport to dismiss those the majority considers to be the lunatic fringe, it does no service to anyone to eliminate viewpoints that come from outside of the two major parties, particularly at a time when those parties are held in historically low esteem. Listening to the arguments from low-polling candidates does not mean you need to vote for them, but not listening gives you a stunted view of the world and precludes the opportunity for growth or change.
In short, polling does for politics what the economy has done to Americans over the last 30 years: it helps consolidate power with the few that already have it and marginalize everyone else. Given that, I’m not sure that we should be devoting our time to make our polling ever more accurate. Rather, we would do well to ask ourselves why we are polling to begin with. If we’re candid, I don’t think we will be flattered by the answer.