US election polls miss shy Trump voters and failed to predict the way votes were cast (original) (raw)

At the moment, it looks like Joe Biden has won the US election with a 4-5 per cent margin nationally. This means the polls were around 5-6 per cent off on average — a larger error than in 2016.

After the 2016 polling errors, it is legitimate to wonder why pollsters got it wrong, and once again underestimated the popular support for Donald Trump.

The Trump campaign has claimed that public polling was used as a "voter suppression tactic" against Trump voters to mislead and demotivate them.

There are two points in that accusation.

First that biased polls may have penalised Trump and second that these biases were intentional.

The second point is extremely unlikely. Pollsters, in a democracy like the US, are carefully regulated and monitored.

Furthermore, it is in their interest to be accurate — their business model relies on their reputation of making the right calls.

They are highly unlikely to engage (and even less to coordinate among many pollsters) on a partisan strategy to influence public opinion.

However, the first point likely stands. Polling biases of that magnitude raise a real concern about the fairness of elections. Polls influence voting.

Research shows that positive polls do provide a "momentum effect" and increase a candidate's chance of winning.

Referring to tense political differences in the US, Joe Biden said: "It's time to put away the harsh rhetoric, lower the temperature, see each other again, listen to each other again and to make progress." (Reuters: Jonathan Ernst)

It's getting harder to poll

However, it is important to appreciate the challenge pollsters face.

Ideally, the method of polling is to survey a small sample of respondents who are representative of the population to get a fairly accurate picture of the public opinion.

In practice, this approach faces some problems. First, people who are surveyed may not feel comfortable revealing their true political positions.

It is particularly an issue when respondents support a politician who is polarising, like Trump. In the US, such respondents have been called "shy Trump voters".

Second, pollsters may struggle to gather a representative sample of the population. When everybody had a landline, polling could be done by randomly picking numbers in the phone book.

In the era of mobile phones and internet, that is not possible anymore. People who answer polling surveys on the internet may be very different from the population as a whole.

Donald Trump's legal team launched several suits in key states as the vote count dragged on. (AP: Sue Ogrocki)

Is the polling sample representative?

In 2016 and 2020, it seems that the problem has been, in each case, the failure to reach a representative sample.

The analysis of the polling errors in 2016 suggests that respondents with less education were not given enough weight.

In 2020, it seems that the issue may have been with non-respondents: people who choose not to respond to surveys, or who don't answer the phone when a pollster calls.

The support for Trump was stronger in this "low social trust" category of the population.

Some states have good reputations for counting votes quickly and accurately. (Reuters: Kevin Mohatt)

The fairness of elections requires polls that do not systematically underestimate the support for one of the contenders.

This changes what we thought we knew

The outcome of the 2020 election also invites us to revisit a few beliefs.

Popular support for Trump and the Republicans was likely higher than estimated throughout his presidency.

While Trump lost, the Republicans retained the Senate and gained seats in the House of Representatives.

An Edison Exit poll indicated that Trump increased his vote shares in 2020 among all race and gender categories, except white men.

In particular, he received more votes from minorities like Latinos and Black Americans, as well as from women in general.

If confirmed, this result may invite Democrats to reconsider the roots of the support for Trump and how to address it.

It also suggests, as political scientist Matt Grossman has pointed out, that the 2016 election loss for the Democrats may have been less about Hillary Clinton's personality, the Comey letter or foreign influence than previously thought.

Finally, this election once again showed how much social media may distort views about political opinions in a country.

People active on social media are far from representative of the whole population.

As a consequence, the topics discussed and positions that are popular on social media may be quite different from those that matter to the average voter.

The Biden campaign seemed well aware of that. They indicated that they had opted to turn off Twitter, weeks before the final election.

Lionel Page is a behavioural economist at University of Technology, Sydney.