Voting intentions polls in the election that brought Joe Biden to the presidency of the United States gave the right result, but they also showed the biggest mistakes in 40 years. The analysis, published six months after the Democrat took office, was carried out by the American Association for Public Opinion Research (AAPOR).
According to the team that analyzed 2,858 polls over the period, the main mistake was to overestimate Biden’s victory and underestimate the defeat of his opponent, then President Donald Trump. They overstated the margin between Biden and Trump in the national popular vote of 3.9 percentage points and 4.3 points in state polls for the Electoral College.
Overall, the biggest mistake was pointing to a possible defeat for Trump. The polls underestimated Republican support in almost all states by an average of 3.3 percentage points, while increasing support for Biden by 1 point.
In the races for the Senate and state governments, a similar pattern was observed: the margin of Democratic candidates was exaggerated and that of Republicans minimized. Only 66% of the polls obtained the correct names of senators.
“There was a systematic error in terms of seeking excessive Democratic support in all scenarios,” Josh Clinton, a Vanderbilt University professor who coordinated the 19-researcher task force, told the Washington Post.
In November, a month before American Electoral College confirmed Biden’s victory, it was possible to observe considerable errors in the polls, which painted a more favorable scenario for the Democratic candidate than it was. actually confirmed.
The polls in the state of Michigan, for example, were down 7.2 points – they found Biden to have an 8.1 percentage point advantage when at this point he was just 0, 9 point. Similar errors were seen in Wisconsin, Ohio, and Pennsylvania.
These differences, however, were already known to Americans and gained notoriety in the 2016 election, when polls threw Hillary Clinton into the White House, but Trump was chosen.
At the time, the industry sought to figure out what was wrong and came to the conclusion that the main reason was an error in the sample of whites without higher education, a group that categorically supported Trump and was under -represented in polls.
But it seems that this is not the factor that led to the failures of the 2020 surveys. The working group team that led the study did not come to firm conclusions about the origin of the methodological errors. He said, however, that it is possible to rule out some of the possibilities. One of them is the failure of responses regarding education levels, which predominated in the 2016 elections.
It is also not plausible to attribute the error to an alleged number of voters who decided to vote only on the eve of the election, since, according to the survey report, in 2020, most Americans had already made up their minds long before they filled out the ballots. .
The document then suggests two possibilities. The first is that Republican citizens who agreed to answer the intention to vote polls had different electoral options than those who declined to answer.
This would be justified by the decline in Americans’ confidence in institutions, largely driven by the speeches and actions of Trump, who tried to discredit professional media and polling intentions.
The second, the most valued by the team, is the role played by new voters. There were 22 million more certified votes in the 2020 election than in the 2016 election – in the United States, unlike in Brazil, voting is not compulsory.
Professor Josh Clinton told the Post that 2022 will be a good thermometer for understanding the weight of the Trump factor in poll errors. In legislative elections for the House of Representatives and the Senate, also known as midterm elections, the electorate tends to differ from the presidential election.
If Trump isn’t on any ticket, there’s a chance the poll errors will resolve themselves “on their own,” Clinton said. In a scenario where this does not happen, research institutes will be left with the diagnosis that they have lost the ability to engage with certain specific groups of voters.