An old quip goes something like this: why did God create election forecasters? To make weather forecasters look good. But is that fair? Or can we trust the polls for the US Presidential election?
What should we look for when reading ‘the polls’?
It’s important to say that ‘the polls’ are not one homogenous blob. Dozens of polls come out each week, from different organisations, in different areas, with different methodologies, and different results. It would not be hard to find a poll that tells you whatever you want to hear.
Therefore, instead of focusing on individual polls, polling averages are a good place to look. They come with a much smaller margin of error since the imperfections of individual samples are blended together.
These polling averages are particularly important for swing states. These are the handful of states that are a toss-up each election. They effectively decide the election, since the winner of a state takes all of its Electoral College votes, and the Electoral College vote is what ultimately determines the Presidency. The Electoral College winner is usually the same as the national popular vote winner – but not always, as Hillary Clinton found in 2016.
What are the polls saying?
With all this in mind, there is a clear trend among the polls: Joe Biden is up, and he is up quite comfortably. Nationally, the Real Clear Politics polling average puts him 8 points up on Donald Trump; FiveThirtyEight puts this figure at 9.4 points. In the six swing states of Florida, Pennsylvania, Michigan, Wisconsin, North Carolina and Arizona, FiveThirtyEight’s polling average has Biden’s lead ranging from 2 points to 8 points; Real Clear Politics reckons the average gap is 3.8 points. These data fuel FiveThirtyEight’s election model which gives Biden an 87% chance of victory; The Economist’s model stretches to 95%.
In short, if the polls are accurate, Biden is set to win the popular vote by the largest margin this century, comparable only to Obama’s victory in 2008, and the Electoral College too.
However, it is in fashion to disparage polls. 48% of Americans say they do not believe most polls (according to a poll – yes, the irony is noted!). The one thing uniting Democrats and Republicans, voters and politicians alike, is that they all say that the ‘silent majority’ is on their side, and will reveal itself not in polls but on election day. But where does this mistrust come from?
What did the polls say in 2016?
Hubris and partisanship are both possible explanations, but another is the polling from 2016. If we rewind the clock to the eve of the election in 2016, RealClearPolitics showed Clinton with a national 3.2 point advantage and a 71% chance of winning. The New York Times put this figure at 85%. Of the swing states, her RealClearPolitics lead was 3.8 points in Florida, 10.7 points in Michigan, 3 points in North Carolina, 8.3 points in Pennsylvania and 6.7 points in Wisconsin. On election night, she lost them all – and the Presidency. The argument is, therefore, that the polls failed spectacularly to predict a Trump victory last time, so they should not be trusted this time.
How has this been put right?
After an election cycle, pollsters naturally analyse and improve their models for the next. In the UK, after the polls failed to predict Brexit in 2016 and a hung parliament in 2017, they were accurate in predicting a sizeable Conservative majority in 2019. Similarly, in the US, after the shock of 2016, they had, according to FiveThirtyEight, “one of their best election cycles, ever” in the 2018 midterms – it was the most accurate cycle since 2003-04.
Pollsters are particularly doing more work to weight samples by education. In 2016, there was a clear ‘diploma gap’ – Clinton had a 9% lead among college graduates and 21% among those with higher degrees, while Trump had an 8% lead among those who were high school graduates or less. However, this was not properly represented in the polls, in part because graduates are more likely to spend time doing polls.
Methodology has also been questioned. A poll that uses the traditional method of calling landlines will inevitably reach a different group of people than a poll that is conducted by calling mobiles. As a result, Suffolk’s Political Research Centre, for example, has increased the proportion of people they reach by mobile from 80% in 2016 to 88% in 2020. Online polling, which is still in its infancy, is also better understood by pollsters now.
Another thing that tripped up the polls in 2016 was social-desirability bias. Some voters were reluctant to admit their alignment with Trump to pollsters, for fear of how they would be perceived. This effect could be weaker this time around given the extent to which four years of exposure to Trump has normalised him and his politics. However, it could also be stronger – four years of Trump’s unadulterated behaviour might increase the shyness of Trump supporters, or the tribal state of politics could compel anti-Trump Republicans to be shy.
Is there an alternative explanation?
If the prevailing explanation of the shock of 2016 is about the failure of polling, there is another one: the polls were misread.
Quite simply, the polls in 2016 were off, but not that off. In the middle of October, the RealClearPolitics polling average had Clinton winning the popular vote by 5.2 points, and she actually won by 2.1 points. That 3.1 point difference is only 0.4 points higher than in 2012, and only 1.1 points above the average since 1968. This small error was then compounded by Trump’s efficient distribution of votes and the resulting Electoral College tally.
Furthermore, the race tightened in the closing days. The polls could not have predicted that the head of the FBI would announce he was re-opening the investigation into Clinton’s emails, and the effect it would have on voters; nor that the 15% of undecided voters would, in the end, swing to Trump – by the very definition they were undecided.
This is not to say that Trump’s victory in 2016 was not a shock, just that the polls were not to blame for the shock. Nate Silver, a polling analyst, defended the polls, instead criticising “a pervasive groupthink among media elites, an unhealthy obsession with the insider’s view of politics, a lack of analytical rigor, a failure to appreciate uncertainty, a sluggishness to self-correct when new evidence contradicts pre-existing beliefs, and a narrow viewpoint that lacks perspective from the longer arc of American history.” Chris Jackson, another pollster, agrees. “2016 was a failure of analysis and reporting. Important nuances about how they should be read were lost in the way they were presented to the public.”
Of course, pollsters have a vested interest in defending their own performance. But they are right – polls should not be used to singlehandedly predict the election.
What can’t the polls tell us?
One of the big questions for every poll is turnout. People change their mind about whether they will vote, or are reluctant to tell pollsters that they will not. Sharp, local upticks in Covid cases could exacerbate this problem by making people feel unsafe going to the polling booth. This could, though, be mitigated by the historic increase in postal voting, and these voters will also be able to confirm to pollsters that they have already voted.
However, there is also uncertainty about the counting of these postal ballots. How many postal votes will be rejected, and will this impact the race? The evidence so far is mixed. In North Carolina, for example, 1.3% of postal ballots have been rejected, which is half that of 2016, but the absolute number is a lot bigger since more people are voting by postal ballot. Furthermore, African-Americans, who vote overwhelmingly Democrat, have had their ballots disproportionately rejected – they make up 17% of postal ballots, but 42% of rejections.
Plus, there is still the possibility of another ‘October surprise’ in the dying days of the campaign. How would the race change if Biden tests positive for Covid? Or if Trump proclaims that he has a vaccine? The polls cannot help us with these questions.
Finally, the election result may well be disputed. If Trump does not win, what happens if he does not concede? Equally, counting the votes is likely to last for days, so what happens if Trump declares victory on election night, but Biden then pulls ahead as these late results come in? Again, the polls cannot help us here.
Can we trust the polls?
In short, the polls do give us a useful indicator as to the state of the election. They were not actually that far off in 2016, and they have since tried to correct what went wrong. However, there are numerous unknowns that the polls cannot help us with, new errors could emerge, and even if Trump has only, for argument’s sake, a 5% chance of winning, that is still a real possibility.
You might use the polls to bet £20 on Biden, but you would not bet your house on him.
Election forecasters and weather forecasts might agree on this one though – the next few weeks could be stormy.
Image Source: Tasnim News