A group of five leading Democratic poll companies took a deep dive into errors that left polls in 2020 for the second consecutive presidential election far behind – and admitted that the data was often “ blatantly ” wrong in the Republican states.
The assessment comes five months after an election in which pollsters let Democrat Joe Biden lead Donald Trump by 10 percentage points in the run-up to the election, for a race in which Biden’s actual margin was 4 percentage points.
As per consequence, a series of pollsters had Democrats dominating the main House and Senate races – they missed how Democrats saw their majority in the House shrink. Democrats took over the senate only after winning two second elections in Georgia.
A group of five Democratic polling stations looked deeply back at what went wrong in 2020, after polls underestimated support for Donald Trump and failed to foresee that “ low inclination ” voters who supported the Republican would show up on election day.
After taking the unusual step of sharing data to identify structural flaws, pollsters say that by 2020, “ our industry saw major errors and fell short of our own expectations. ”
There was more data in Republican states
“We saw that in more democratic states and districts, and in some closely divided states like Georgia and Arizona, the data was pretty good. But in more Republican areas, the data was often wrong, sometimes even blatant, ‘they wrote in the US rating
The group of top pollsters, including the lead pollster for the Biden campaign’s ALG survey, identifies several factors contributing to the industry-wide scent.
Polls struggled to reach low-tendency voters who supported President Donald Trump, seen here on election night at the White House
The pollsters shared data and conducted their own analysis to find out why they missed the mark
Primary inquiries overestimate Biden’s margin and underestimate Trump’s. In some pre-election elections, Biden had a 10-point lead. Biden beat Trump 51 to 47 and won by about 7 million votes in the popular vote
Pollers again struggled to predict which voters would turn out to be. They say they were bothered by Trump supporters who were skeptical of pollsters
Primaries consistently put Joe Biden in the lead, with some pollsters showing him a 10-point lead in the popular vote. The final result, a 51 to 47 percent victory for Biden, was tighter.
Trump supporters were skeptical of institutions – pollsters included
Some of the difficulties are based on changes in the country’s technological habits: more people are using mobile phones, losing landlines and refusing to retrieve unknown numbers. But pollsters have ways to account for such factors – as well as adjusting turnout models to tweak for the electorate they think will actually show up.
The group is also thinking of polls specific to Trump supporters, who may be more skeptical of institutions like the government, the media, and pollsters. Pollsters say Trump himself may have exacerbated the situation.
What is ‘bias for social desirability’ and why does it affect the polls?
Social desirability bias is the name some pollsters have given to the effect that Trump supporters keep quiet when asked who they vote for.
It simply means that people are more likely to give the answer they think the person asking them wants to hear than what they really think.
While this could be true for any question, in Trump’s case it is exacerbated because of the abuse Trump supporters are often exposed to, pollsters say.
Robert Cahaly, an interviewer for the Trafalgar Group, said the abuse has worsened between 2016 and this year.
In 2016, the worst thing about Trump voters was said to be ‘regrettable’, he said.
Now ‘people [are] be beaten up for wearing the wrong hat, people are harassed for having a sticker on their car. People just don’t want to say anything. ‘
Fellow pollster Arie Kapteyn said this effect can be measured by asking people who they vote for and then asking who they think other people vote for.
Asking people who they will vote for in 2020 provided Biden with a solid 10-point lead, he said. But if you ask them who other people will vote for, cut that in half.
This is because it gives people the freedom to speculate about how popular they think a candidate is, without involving themselves.
“Trump may have helped make this a problem for pollsters by attracting suspicious voters and making his most ardent supporters even more suspicious of other people, the media and perhaps even the polls themselves,” they write. “That, in turn, had made his supporters less likely to answer polls.”
The review is attended by ALG Research – which conducted the survey for the Biden campaign, Garin-Hart-Yang Research, GBAO Strategies, Global Strategy Group and Normington Petts.
“Trump went after the polls,” said a pollster Politics“He was very open to those who listened to his suspicion in opinion polls or media.”
Another said the 2020 release was an ‘Oh, s —‘ moment for all of us, adding, ‘I think we all got to the point of having to put our ego aside quickly. We have to do this right. ‘
Models were wrong for ‘low-tendency voters’ – underestimationing white rural turnout – The pollsters also solved the problem by educating voters with low inclination, those who are unlikely to show up, even if they tell a pollster they have a preference.
They found that their models “consistently overestimate Democratic turnout relative to Republican turnout” for such low voters. Among such voters, “the Republican share of the electorate exceeded expectations at four times the rate of the Democratic share,” they conclude.
“This turnout error has, in some places at least, again underestimated the relative turnout among rural and white non-university voters, who are over-represented among low-propensity Republicans,” they write.
A voter poll amid a global pandemic would be challenging under any circumstances. But pollsters believe the pandemic could have worked to skew the data further.
“Perhaps voters with a more progressive attitude towards COVID-19 were not only more likely to wear masks and stay at home, but also more likely to answer our polls, while conservatives remained more difficult to reach,” they write.
Late break not considered controlling
The authors cite theories of a similar polling error in 2016, which may have missed a late break among voters that gave Trump a surprise victory over Hillary Clinton. But they note the stability of the polls in 2020, saying it probably wasn’t the driver this time around.
How to avoid making the same mistakes again
The pollsters have suggested a number of possible solutions, but don’t try to propose a detailed solution that would fix the poll if Donald Trump walked again in 2024, as he could. One idea is to go back to door-to-door investigations from an earlier era. They also talk about paying for research topics to encourage participation, and about developing ‘multimodal research tools’.
They point out that the error factor may be in part specific to Trump, so they warn against over-correction against someone they say will “ hopefully ” never be on the ballot again.