Yesterday, should American Public Opinion Research Association A report on voting in the 2020 general election was released. This is ugly. In my opinion, this is the key to discovery.
- The 2020 polls showed an unusually serious voting error: this is the highest level of universal suffrage in the country in 40 years, and the highest level of state-level vote estimates for presidents, senators, and governors in at least 20 years. 5 Among the polls conducted in the last two weeks, the average error in either direction in the national general election poll was 4.5 points, and the state-level presidential poll was 5.1 points.
- Voting errors are more likely to allow Biden to beat Trump. In the polls conducted in the last two weeks before the election, the average signature error in the national poll was too favorable for Biden, 3.9 percentage points higher in the national poll and 4.3 percentage points higher in the statewide presidential poll.
In short, as a group, the polls are very wrong, and they are all wrong in the same direction. When I served as the inspector of the inspector general, this was the clue we referred to.
Through politics Poll: “Impossible” to explain why the 2020 polls were wrong:
But unlike in 2016, when pollsters can identify factors such as the educational gap as the reason for their underestimation of Donald Trump and provide specific suggestions to solve the problem, the authors of the new report of the American Public Opinion Research Association cannot point out the exact problem. Face it now. Instead, they insisted on denying the idea that they had made the same mistakes as before, while pointing out new reasons that might have caused inaccuracies.
Josh Clinton, a professor at Vanderbilt University and chair of the association’s 2020 election working group, said: “We can rule out some things, but it is difficult to determine what happened.” “Based on what we know about the polls, Our understanding of politics, we have some good prime suspects for what might happen.”
From political campaigns to the news media, these “primary suspects” hardly comfort pollsters and those who rely on them. Most likely-if it is far from certain-the culprit causing the unbalanced poll results is that key populations do not answer the polls first.
For more than a decade, the decline in response rate has been a major concern of opinion polls. But the politicization of voting in the Trump era—including feedback loops from the former president, who falsely denounced poll results he didn’t like as “fake” or deliberately inhibited Republican voters’ enthusiasm for answering the polls—seems to be Distorted the results, some Republicans refused to participate in the investigation.
Several observations. For about 30 years, the response rate of opinion polls has been declining. There is no evidence in the study that Republicans refuse to participate in the investigation more than Democrats. But accusing Trump is still the solution.
This makes the question of the polls more difficult to solve than the diagnosis four years ago, which focused on adjusting the survey to explain Trump’s popularity among voters who did not earn a college degree and his holdings of a college degree. Corresponding weakness in the person.
Clinton said: “For the working group, the Republicans who participated in our polls may be different from those who support Republican candidates who did not participate in our polls. This seems reasonable.” “But how do you prove it?”
These are some great analytical jiu-jitsu. What this guy said was, “We were wrong four years ago because we underestimated the number of uneducated rubles that would vote for Trump, and we solved this problem. Therefore, our 2020 results are better than 2016. Bad. But we’re really sure that Republicans won’t do anything wrong with the hard science of opinion polls.”
According to the report, the national opinion polls of the presidential campaign conducted in the last two weeks of the election dropped by an average of 4.5 percentage points, while the state polls fell by only 5 percentage points. Most of the mistakes are in one direction: Judging from the vote gap, the national poll is too favorable for the current President Joe Biden by 3.9 percentage points, while the state poll is too favorable for Biden by 4.3 percentage points.
Most of the mistakes come from underestimating Trump’s support, not overestimating Biden’s support. Comparing the final election results with the poll numbers of each candidate, Trump’s approval rating was underestimated by an average of 3.3 percentage points, while Biden’s approval rating was overestimated by one point—will seem to be a solid worship. The lead has become a closer, but still decisive, competition.
This is not just the Trump effect. The gap in polls between the Senate and governor elections is even greater: an average of 6 points.
“In the same state, voting errors in senator elections are often greater than in presidential elections,” the AAPOR report reads. “Whether the candidate is running for president, senator or governor, overall, the profit margins of the polls indicate that, relative to the final certified vote, the Democratic candidate will do better, while the Republican candidate will do Worse.”
No method performs better than the others. According to the report, there are only “nuances” whether they conduct polls over the phone, the Internet, or using a mixed method (including text and smartphone apps), or whether they contact voters at random or from outside the list of registered voters. The report said: “Every interview method and every sampling method exaggerates the gap between the Democratic Republican Party and the final certified votes.”
On Fivethirtyeight.com, the website of opinion survey expert Nate Silver, they conducted an analysis last month (The death of voting is greatly exaggerated) And found that the polls are very good, because of course they did.
However, although the accuracy of the 2020 polls is mediocre, they are not any historical outliers. The overall average error of 6.3 points in 2019-20 is only slightly worse than the average error of 6.0 points in all opinion polls since 1998. There are also presidential terms before the periods covered by our poll ratings, such as 1948 and 1980, when polls showed a significantly greater error than in 2020.
So let’s review it. Regardless of the method used, the polls support the Democratic Party. Regardless of the model analysts use, the polls support most Democrats in the polling industry. The polls are moving in the same direction with very similar numbers. The best reason they can come up with is that Republicans do not participate because President Trump said something harsh to them, or they are interviewing a large number of Republicans who do not represent voting Republicans… Maybe they have The Bulwark, the North American Man-Boy Lincoln Association, and The Dispatch on speed dial?
I think there is a simpler explanation for what happened, and it doesn’t involve the entire polling industry’s failure to discover the key part of Republican/conservative voters. The polls are conducted to promote media reports that the United States is shocked by Trump and the Republican Party, and the Democratic Party will win a big victory. The polls provided material for “news” stories and talk shows. The idea is to create an atmosphere where Biden’s victory is inevitable. What they didn’t consider was the anger of the conservative base and the degree to which millions of Americans believed President Trump to stand up for them and vote to return their loyalty.
The real question looming in 2022 is will it happen again? Biden has minimal control over the House of Representatives and Senate. The Democratic Party needs a lot of things to maintain Congress. Will pollsters go back to their original conditions and start showing us the polls that show the Democratic Party is ruthlessly heading for victory? Or will they try to make accurate predictions? I know what I am betting on.
