Against Polling

The polling industry claims to offer live snapshots of what the public thinks. Yet the obsession with polls is reducing politics to market research.

Nate Silver is the founder and editor in chief of FiveThirtyEight. (Corbis / Getty Images)

In a UK local election in 2018, a tiny, newly formed centrist party standing in a London borough issued a briefing that cited the high percentage of voters who agreed with its policies. But the “policies” in question were just generic vagaries like “more housing” and “clean air.” The briefing’s conclusion — holding that these results demonstrated the party’s own popularity — was simply meaningless.

More recently, Tom Watson, deputy leader of Britain’s Labour Party, used an online poll promoted via social media without a trace of methodology or rigor to evidence support within Labour ranks for a second referendum on Brexit.

These deliberate and disingenuous uses of big numbers are not difficult to correct methodologically — in the first case, one could get a clearer answer by surveying attitudes towards the new party (or even its name recognition), while in the second one could draw on the multiple more detailed surveys of Labour members’ views on Brexit, presenting a nuanced picture.

These examples are hardly unique. But they are worth highlighting because they come not from charlatans who talk about being “fed up of experts,” but from precisely the sort of people who bemoan the rise of fake news, “post-truth politics,” and the supposed demise of evidence-based argument.

Bending the truth to prove people agree with you predates polling. Think of everyone who backs up their opinion with a questionable anecdote of a conversation they overheard on the bus, or those claiming to speak for “the silent majority.” But big numbers lend such assertions the weight of objectivity. And to understand the manipulation of those numbers as just a case of vandalism and blasphemy against “proper” polling is insufficient. The apotheosis of nonsense claims has been written into “the polls” for a long time, and our current broken political model has turned the worship of polling into blind evangelism.

To blaspheme, one must first understand the divine. It’s the virtually religious quality with which polling is treated that creates an environment where manipulating opinion to debase debate has become a viable strategy for the warrior-priests of our political tribes. Opinion polling, or whatever passes for it — now backed occasionally by the new world of real-time survey visualizations on social media — is ascribed the powers of a digital Delphic oracle. Those capable of interpreting the oracle enter its sanctum, receive its wisdom, and distribute it to those overseeing public life.

For a price.

Pick a Number. Any Number.

The basic techniques of opinion polling are old. Inferential statistics, which underpin the polling methods most widely in use today, have their roots in the nineteenth-century academic debate between those who understood the social world as a largely biological phenomenon and those who saw it as the product of human power relations. This branch of statistics, applied in the social sciences, produces claims about a population on the basis of samples deemed representative of it, by virtue of weighted demographic factors like age, race, and gender.

In the early twentieth century, such methods came to be used in elections. Prior to this, the political choices and social lives of those who had the vote were researched in other ways, like observation and interview or by measuring the amount of space provided for different issues in a newspaper. With the popularization of broadsheets came advertising; and with advertising, the opportunity to ask readers about their purchasing habits. Soon the giants of this industry, like Gallup, were adding questions about voting preferences to their surveys, thus heralding a new field of study for those keen to understand politics through statistics. When an American poll conducted by the Literary Digest of over two million voters wrongly called the 1936 election due to over-representing affluent Republicans in its subscriber base, the first modern major polling error was born.

Some approached opinion polling with the naïve hope that it would foster democratic representation, since politicians who knew electorate opinion would feel obliged to follow it. Qualitative ways of studying society had been under threat from the fledgling field of economics. The centuries-old apathy of elites for engaging the wider public in genuine political debate remained intact. A splinter epistemology, the economists of the eighteenth century severed ties with the social scientists they had long been associated with through the conviction that society could best be understood through statistical measurement of its relation to markets and money. In this fluid context, politics entrenched the role of the pollster as high priest.

But politics wasn’t the force motoring it. In fact, political polling was a byproduct of the 1930s advertising boom. The Literary Digest ran its political polls to increase consumer interest and circulation. Today, the big pollsters make a very small proportion of their revenue from political polling, but constant free media coverage of their political work helps keep them relevant enough to net most of their commercial clients. The way we use big numbers is not neutral. It frames politics as market research. And it is emblematic of the neoliberal period.

Do Politicians Dream of Electronic Electorates?

Pollsters’ commercial clients need to pay for market research data because people will not volunteer their time freely to canvas people on their preferences regarding ceiling fans. Market research data is sufficient for their purposes, because they seek information to tailor products to consumer preferences in one-way transactional relationships.

But politics is supposed to be different. Political parties were historically mass organizations with engaged and geographically diverse bases, not only giving leaders and strategists a potential well of opinion to tap into, but forcing them to do so. In the 1950s, the Conservatives peaked at three million members; Labour had 1.5 million. In 1980, on the eve of Margaret Thatcher’s war on Britain’s trade unions, 12.8 million workers were union members. Through the Labour Party, this provided an organic link between the collective interests of organized workforces and political power. The decline of both mass parties and organized labor has, however, undermined this deeper political engagement, replacing it with a neoliberal notion of consumer choice by atomized individuals — the transactional model of the relationship between ceiling fan vendor and buyer. Indeed, the depletion of space accessible for popular organizing cannot be separated from the financializing economic reforms that have taken hold in advanced capitalist countries around the world.

For politicians, drawing strategic conclusions from ground-level political intelligence is as much of an art as a science. Strong links between pavement politics, trade union power, and national political parties, as in the mass-organization model, can improve political representation in important qualitative ways. In the UK, Labour has recently regained some momentum in this regard under Jeremy Corbyn’s leadership, but these ties remain greatly weakened. The American parties are less reliant on a standing army of party volunteers to turn out voters, but their looser assemblages of canvassers work less well as a scaled intelligence-gathering and representation apparatus over vast territories. Parties thus seek a shortcut to political intelligence by placing their trust in big numbers via polling and focus groups. As political adviser Glenn Cullen witheringly puts it in BBC satire The Thick of It, they play “dial-an-opinion… Send me three liberals, two fucking mavericks, and a racist.”

The opinion industry is eager to sell those shortcuts and exaggerate their benefits. The polling companies of the digital age offer glossy presentations on the capabilities of their platforms and the superiority of their algorithms. Such awe-inspiring “scientific” thinking belies the real politics of big numbers that political advisors and journalists rarely have the knowledge or capacity to interrogate carefully. Even if they did, interrogating the meaning of numbers takes time and effort which a media and political landscape ruled by clickbait and fast-moving news cycles is not geared towards.

The industry’s political clients often want to believe the impressive PowerPoints. Statistics and algorithms don’t need to be argued and engaged with. They don’t shout at you in meeting halls. They promise a low-effort, cheap-representation, and high-support approach to policy.

The unspoken vision here is sinister. Political communication is not two-way; strategists decide what questions should be asked, how they will be answered, how the data will be interpreted, and what actions it will inform. The interpreters no longer limit themselves to conveying the results of surveys, or even suggesting what they could indicate. Increasingly, they allege powers of divination, the ability to predict potential scenarios in unnerving detail. The dream underpinning this does not use technology to supplement democracy; instead it imagines the automation of politics. It wants to replace the spaces in which power is negotiated, the messy realities of human existence, and instead achieve balance with the dry discipline of numerical modeling. It does not fall far from the realm of nightmare science fiction.

The implications of this potential reality pose profound threats to democracy as we know it; the vanguards of Silicon Valley technocracy have long fantasized a political world where our opinions are frozen in time and rendered utilizable on the blockchain. The state becomes obsolete, as decisions about what and who we value in society are made through calculation and the programming know-how of a dispersed (and unaccountable) programming literati. This development extends far beyond the domain of polling alone. But the problems of our current political model more often result from the belief that electronic electorates are possible and desirable.

How to Unfreeze Time

“This poll shows X percent of respondents think Y” is a sentence employed to mean: “The people believe Y and I have ironclad evidence for this.” It is then, crucially, weaponized into “therefore we should do Z.” The “push poll” tactic beloved of contemporary spin doctors — to massage a result and use it to make a news story or win an internal argument — follows from the belief that polls are numerical values of the will of the people and therefore should determine political action. Professional research bodies take a dim view of this development. The American Association for Public Opinion Research deems push polls a form of “unethical political telemarketing” that “exploits the trust people have in research organizations.”

So why do so many politicians and commentators put faith in surveys even when they lack methodological rigor? The cult and culture of polling has to be more than just a phenomenon summoned forth by the demands of political strategists for intelligence, and those smart enough to occupy the routes of supply. It is self-perpetuating and self-replicating, a living worldview which continues to multiply and spread throughout the body politic with every push poll or more legitimate statistic enlisted for illegitimate ends, informing how each new political actor makes sense of the world around them.

The relationship between politicians and polling gurus, is, however, also driven by a wider neoliberal philosophy.

First is the denial of the role of collectives, territories, and associations, from family to community, in forming views. Opinion is read as the aggregation of individual units, to be weighted by individual “identities.” Second is the assumption of the dominance of “center-ground” politics; because big numbers allow all the messy and conflicting and different worldviews that aren’t tribally partisan to be averaged and then framed — inaccurately — as “people in the center ground.” Third, the wielding of big numbers sits within neoliberalism’s technocratic assumption: that politics is not a site of conflict between different interests or values and we simply need to find the “right” policy that achieves the “best” outcome. Meritocracy helps us sort the “best” politicians who can then govern with the assistance of “experts” while occasionally “consulting” people via survey data before imposing policy. Polling is the oracle, but the religion itself is a context-less “expertise.”

Floating above the electorate, connected only by the polling data which you have outsourced to experts, is a lonely and isolating position. It makes public opinion a strange, snarling behemoth that you rely on your agents to tame. An opinion poll that claims 53 percent of people oppose Z is not merely intelligence indicating that a campaign to win the argument for Z should be mounted with strategy and guile. It makes Z seem unachievable, and anyone who argues for Z seem unelectable. When Z means something like migration or welfare – issues on which the Right has for many years controlled the argument while the center-left vacated the field – it is easy to see how the logic of politics-by-numbers acts to constrain progressive interventions. And the status quo is amenable to the vested interests of those who control many of the levers of political discourse. Or as a lyric in Bright Eyes’s “Let’s Not Shit Ourselves” has it, “The approval rating’s high… so someone’s going to die.” If you can’t change politics through a project designed to transform not just opinion but the roots of opinion, then time itself stops. It becomes what Francis Fukuyama called (with a much darker spin) “the end of history.”

But time is now unstuck again. The Brexit referendum and the 2017 UK general election confounded pundits’ expectations not only because the pollsters got it wrong but because of the vastly inflated predictive power that had been ascribed to them. (In 2017 some polls did creep, at the last moment, toward the Conservatives losing their majority, but the parliamentary press gallery’s structural inability to take Jeremy Corbyn’s Labour seriously outweighed even their faith in pollsters.) Whether the polls were “wrong” or not it was clear that they did not serve as reliable indicators of what would happen when an effective campaign was applied in real life. Lessons have not been learned; the Daily Telegraph recently commissioned a poll that asked voters how they would most likely vote in a general election if each of the Tory leadership candidates were to become prime minister. Readers would be forgiven for assuming that such an exercise has any predictive capacity whatsoever.

Polls Apart

Subordinating politics to polling has taken us further into a universe which looks very different to the objective, rules-bound reality that the digital technology industry and neoliberalism promised. This poses a challenge for those who do want to transform politics, do have richer theories of how opinions are formed and changed, and do believe in movements and communities as a valuable source of political intelligence. The Left should appreciate the limits of opinion polling but develop an understanding of the specific contexts when its methods can be useful.

This proves challenging today. It’s been suggested that the number one threat to the methods traditionally employed in electoral polling is that fewer people than ever before pick up the phone when a pollster calls. Professor Dan Cassino suggests low response rates for polls for the 2016 US election may have been disproportionately associated with groups of people who were skeptical of the polling industry. This “self-screening” undermines the random sample on which public opinion’s statistical models are based. New models developed for analyzing public opinion surveys using large, non-random samples conducted via digital channels may provide some relief, but these also throw up significant statistical challenges. The old problems of working with categorical data (such as demographic information provided by a census) remain, and the non-probabilistic nature of the samples themselves render whole branches of statistics futile.

Yet these new approaches at least attempt to account for the limits of polling in the digital age. It is up to the Left to find a place for them, within a broader project of rebuilding structures of representation and channels of communication between political institutions and communities. Socialists should push for care and rigor in the use of numbers rather than being tempted to join everyone else in the gutter of Twitter polls. Progressive statisticians and social scientists have an important role to play in improving the models and methods of academic public opinion research, and their applied use. A better politics of polling must ensure that our analytical capabilities are unshackled from the realm of market research and placed in the service of better understanding the world, so that we may change it.