US Election: Why did the polls get it wrong – again?
The biggest loser of the US election was neither the Democratic Party nor Republican Party, it was the polling industry.
Pre-election polls once again predicted a comfortable win for the Democratic nominee Joe Biden, but the contest is proving to be much closer than experts predicted.
One of the most egregious errors in forecasting predicted that Biden would defeat Donald Trump by 17 percentage points in the state of Wisconsin. At the time of writing, with 98% of votes tallied, Biden leads Trump in the state by just 0.6% of the vote.
Even if Biden were to win the election, the scale of his victory in the Electoral College will be far smaller than what many pollsters predicted.
US elections are characterised by an abundance of data on voters’ preferences, so why did the polls get it wrong?
News coverage in both the US and the UK, countries that occupy a disproportionate amount of attention in the English-speaking world, rely heavily on public opinion polling to predict the outcome of competitive elections.
But the credibility of election polling took a knock in 2016.
Polling data gave Democratic candidate Hillary Clinton a substantial lead over Trump, with the New York Times infamously suggesting that Clinton had a 91% chance of winning an election she eventually lost.
During the Brexit vote in the UK, most polls suggested that the Remain campaign, supported by both the government and the opposition, would win. The surprising victory of the Leave campaign triggered the prime minister’s resignation, a leadership contest within the governing party and two early elections in three years.
But if polls are simply indications of voting patterns and useful talking points on news programmes, why does it matter if their results are inaccurate?
An over-reliance on projected outcomes can lead to voter complacency.
One of the major surprises of the 2016 US elections was a surge in turnout among voters in rural areas, mostly supporting Trump, while turnout among urban and educated voters, mostly supporting Clinton, was not as high.
Not all surveys are created equally
Many polls are not representative of different voting constituencies. How and from whom pollsters collect data is significant for their accuracy.
If a poll is conducted exclusively online or on the phone, the sample is likely to be skewed towards wealthier, urban and more educated respondents.
And if a poll relies on only a few hundred respondents to predict voting patterns among millions, it is unlikely that the results will accurately reflect the wider population.
The public has no real way to interrogate these polls because survey samples and data collection methods are often not included in media coverage.
No survey can be entirely representative of millions of people, but some surveys are more robust than others.
In South Africa, polling is not as common as it is in the US. But public opinion surveys that probe citizens’ attitudes to the state of governance and trust in elections, such as the Afrobarometer and SA Reconciliation Barometer (SARB), are nationally representative.
These surveys are designed to be reflective of SA’s diverse population, across age, gender, race and population groups.
Both the SARB and Afrobarometer surveys conduct face-to-face interviews with respondents across the country in various languages.
Good things come to those who wait, and survey data is no different.
Accurate and reliable surveys take longer to complete than polls, which are often simple and short snapshots of public opinion.
Fixing the polls
The polling industry should adopt two old adages to improve their analysis: less is more and quality over quantity.
Anyone who has followed the US election would have been bombarded with an uncannily regular supply of new polling data and projections from a variety of sources.
More surveys do not necessarily yield more accurate results. Rather, pollsters should prioritise improved sampling methods. Higher quality surveys are more likely to yield better projections than a higher quantity of polls.
As COVID-19 restrictions have limited opportunities for socialising and as social media becomes increasingly polarised around a range of issues, surveys provide an empirical basis for acknowledging the views of different constituencies.
Although many polls in the US have provided inaccurate projections in two successive presidential elections, survey data remains a valuable resource for understanding wider society, particularly the views of people outside of our own bubbles.
Mikhail Moosa is the project leader for the South African Reconciliation Barometer at the Institute for Justice and Reconciliation.
Article first published on News24