By DAVID CHAUVIN //
No matter where you stand on the accuracy of the polls leading up to the 2020 Presidential Election, it’s not particularly hard to find supporting evidence.
If you believe the polls were largely accurate, you can easily say: “Hey, the polls said Joe Biden was going to win by a fairly healthy margin, and that’s exactly what happened. FiveThirtyEight, for example, correctly picked all but two states! How much more accurate do you need?”
If you feel that the polls were woefully inaccurate, you also have lots of ammunition: “Every single pre-election poll had Joe Biden winning Florida, and Trump won it outright,” you can say. “An ABC poll from a week before the election showed Biden winning Wisconsin by 15 points – he won by less than 1.”
On and on we can go. The rub is, in perfect 2020 fashion, everyone is right. And everyone is wrong.
It’s true in a larger sense that polls predicted Biden would win, and he did. It’s also true that, in many states, the predicted margins were almost laughably inaccurate.
So, what gives? And why this widespread narrative that polls can’t be trusted anymore?
Are the methodologies for polling Americans – or the polling models that ultimately make the predictions – inherently flawed? Probably not. Several of the most prominent prediction models (the aforementioned FiveThirtyEight model, the New York Times Upshot, The Economist’s trusted methods, etc.) have been expertly refined for generations, and as recently as the 2018 midterms were heralded for amazing accuracy.
And contrary to what many people believe, polling itself isn’t just cold calls over landlines anymore. Modern polling is conducted via several channels, reflecting the vast array of 21st century communications devices.
Was it the pandemic? Surely this can’t be ruled out. But with the largest turnout in American history – both President-elect Biden and President Trump received more votes than any other candidate, ever – it’s hard to blame COVID-19 for widespread polling inaccuracy. Perhaps we all should have considered, before the fact, that nothing about the pandemic is predictable, including its effects on election polls.
Is it Trump? Is there something about his supporters that makes them inherently difficult to poll? Is it the so-called “shy Trump voter,” who secretly supports the president but is hesitant to admit it socially?
There is no scientific evidence that the shy Trump voter actually exists, beyond media hyperbolizing. But there are still good arguments to be made here: Of the last three national elections – 2016, 2018, 2020 – the only one the polls nailed was the one without Trump on the ballot.
The problem is this: We’re trying to use the polls for something that they’re not.
Polls, and the prediction models that aggregate them, are powerful tools, but they’re not crystal balls. The sooner we stop expecting them to be that, the better.
Polls offer a snapshot of a moment in time, of the overarching mood of an electorate in a given period. Often, there’s much to glean from these snippets of information – the country’s priority’s, for example, or whether a candidate’s particular message is resonating. But what we can’t ascertain from polls is a clear look into the future.
Statistician Nate Silver and his ilk became quasi-celebrities after their success in predicting the 2008 Presidential election. But here’s the unspoken little secret of 2008: The election wasn’t all that hard to predict.
Barack Obama – clearly a transcendent, once-in-a-lifetime candidate – was running against a historically unpopular Republican party in an election that was quite clearly about change. What the pollsters did during that election was impressive, but it might have given us unrealistic expectations.
Polls, be they canvasses of national elections or ThoughtExchange surveys sent to your customers, give us insights into the minds of our community members. That doesn’t make them soothsayers – but for communications professionals, they’re still incredibly valuable.
David Chauvin is executive vice president of ZE Creative Communications.