Political polling in the United States is neither as reliable nor as harmless as most observers suppose. Technology has made that more true than ever in recent years.
By Adam Garfinkle*
Last month, we noted three rogue factors bearing down simultaneously on the fast-approaching United States presidential election, rendering it the most accident-prone and unpredictable in American history: multipartite foreign interference; the logistical complications of voting in the time of pandemic; and a president who, uniquely in US history, has stated publicly and repeatedly that he will refuse to recognise the legitimacy of any election he does not win and will not concede defeat.
A fourth predictable rogue factor has since appeared: President Trump’s 1-2 October 2020 revelation that he tested positive for COVID-19. To this shock was soon added the spectacle of his muscling his doctors to claim that he was COVID-free, and his ongoing attempt to politicise his own medical adventures to advantage.
More Rogue Factor: Polling the Politicians
Thanks to an unbroken record of White House mendacity over the past nearly four years, we still cannot be sure if the entire episode was not a hoax designed to evoke sympathy, and turn around what are ghastly polls for the Republicans, but that seems highly unlikely given the epidemiological nightmare that is the White House right now.
Assuming it was not a hoax, we still cannot be sure if the president’s “cure”, near to singular for an overweight 74-year old male anywhere in the world, is real; or his most perilous period lies just ahead. That’s a lot not to know on the eve of a critical election.
Alas, a fifth rogue factor has emerged, though it is more a constant in recent American elections: the impact of published polls.
Few Americans realise it, but most of the polls we read or hear about in the media are not social science-valid exercises. These polls bear little resemblance to those many learned about in their undergraduate political science classes. Polling is a business in the US and in many other countries where opinion surveys have become part of the democratic ethos, and the customers are the media corporations that buy them.
Why do they buy them? Because survey research ─ another kind of polling ─ shows that people are attracted to polls as big news, at least when they can be presented in very simple ways, dressed up with eye-catching graphs or charts, and associated with pulsating news “hooks”.
Opinion Polling or Gossip?
Commercial political opinion polling is really just gossip in the plural tense at a time when politics has become more of a series of reality-TV episodes than anything like Max Weber’s “vocation”. Polls can even be made clickbait-attractive to niche audiences, such as the recent Carnegie poll entitled “How Will Indian Americans Vote?”. When those kinds of boxes can be checked, commercial polling is good and often big business.
And why in turn is that? Because, as Friedrich Nietzsche put it in Beyond Good and Evil in 1886: “Were it not for the constant counterfeiting of the world by means of numbers, men could not live.” Numbers create a false sense of concreteness, a simulacrum of certainty about reality that, for many, renders the complex simple.
After all, few have time and opportunity to learn about the various methodologies of polling and their relatives and merits and deficiencies, such as coverage distortions, mismeasurement and misweighting, non-response factoring, and others. So when they see a number that magically seems to produce clarity from confusion on a topic they care about, they cling to it as if by mental reflex.
This is unfortunate, because many commercial political polls have proved notoriously unreliable. The pollsters in Britain got Brexit so very wrong. Some years before that pollsters in Europe mispredicted the results of referenda concerning the European Union, notably in France and the Netherlands. And then, most memorably, they botched spectacularly the November 2016 US election, as well.
We know the reasons. One is conscious bias, and another is unrecognised bias that seeps into the way questions are asked and interpreted.
Bias in Polling
Conscious bias is designed to produce polling news that either mobilises or demobilises constituencies ─ donors long before the time to vote as well as voters nearer to balloting day. Unrecognised bias leads those who design and choose methodologies and sample sizes, for example, to cause the over-representation of some social groups of voters and the under-representation of others.
In 2016, pollsters, either themselves being or relying on highly educated social scientists, designed questions, selected samples, and used techniques like “opt-in” panels that over-represented the minority of college-educated people (like themselves) in a newly populist political environment.
Note that unrecognised bias can apply to serious polling efforts as well as to commercial ones. Hardly any American political scientist predicted Trump’s victory in 2016. A seminal 1964 essay by Philip Converse on the striation of public opinion and style of assimilating political information has evidently still not sunk into the typical academic pollster’s mental apparatus.
A third reason concerns technology. It is now much easier for commercial polling start-ups to surmount barriers to entry thanks to the equalising effects of powerful and easy-to-use computer data manipulation.
The new technical power at hand has caused ferment in sampling techniques, leading some to combine online surveying with robocalling and other methods, and then claim, non-transparently most of the time, huge reassuring sampling sizes and small error margins that are, to put it bluntly, bogus.
Why Polling Faults Matter
These infirmities matter today as they did in 2016, when optimistic polls helped persuade Hillary Clinton not to campaign in important swing states late in the campaign season. When polls say that Biden is so far ahead that Trump barely has any chance, it gives rise to a secondary media meme whose premise has the effect of persuading many people that their vote doesn’t matter. That has two intertwined effects:
It may, under the special circumstances of this upcoming election, reduce pro-Biden voter turnout on election day, for those unable or unwilling to vote by mail; and it may cause the redoubling of anti-Biden energies from the White House and supporters nationwide out to the far sticks.
The presumption of a Biden victory in November may also have affected Biden’s choice of Kamala Harris to be his running mate. If Biden thought he might lose, his choice would have logically been taken with a priority on who would help him win key swing states ─ like first-term Senator Tammy Duckworth of Illinois, if indeed it had to be a woman.
But if he thought he was going to win in a breeze, his choice might well have turned instead on who would be most likely to heal, or paper over for optical purposes, fissures within the Democratic Party. He believed the polls, so it’s Biden-Harris ─ a choice that risks some centrist Democratic and independent support.
How all of this will come out in the wash after 3 November we’ll all just have to wait and see. When it does, we won’t need a poll to tell us the answer.
*About the author: Adam Garfinkle is a non-resident Distinguished Fellow at the S. Rajaratnam School of International Studies (RSIS), Nanyang Technological University (NTU), Singapore and an Editorial Board member of the new magazine American Purpose. This is part of an RSIS Series.
Source: This article was originally published in RSIS Commentary, a publication of RSIS.