"... and in at least 20 years for state-level polls, according to a study conducted by the American Association for Public Opinion Research (AAPOR). The AAPOR task force examined 2,858 polls, including 529 national presidential race polls and 1,572 state-level presidential polls. They found that the surveys overstated the margin between President Biden and former president Donald Trump by 3.9 points in the national popular vote and 4.3 percentage points in state polls. Polls understated the support for Trump in nearly every state and by an average of 3.3 percentage points overall. Polls in Senate and gubernatorial races suffered from the same problem.... 'Identifying conclusively why polls overstated the Democratic-Republican margin relative to the certified vote appears to be impossible with the available data,' the report states."
From "2020 presidential polls suffered worst performance in decades, report says" by Dan Balz at WaPo.
Posts
7 comments:
Temujin writes:
"That is actual voter suppression at work, and in front of the entire nation. That is what an organized voter suppression looks like. Add in the piling on headlines and constant repeating of this misinformation by the same people who spent the previous four years offering up misinformation on any number of subjects, and you have what might be called a 'conspiracy'. But not by me.
"I'm still of the belief that there is no way that 81 million people could have voted for Joe Biden. Not in this lifetime, not in any lifetime would that happen. What strikes me is how people who are seemingly intelligent are not only buying it, but suppressing any thought otherwise."
Michelle writes:
"Never mind "voter fraud" per se. Just ask yourself whether you're
inclined to vote for a candidate who is losing badly, both in your own
state and nationwide. Can't you imagine a lot of people's response being
more or less "F this," and apathy? This is the sort of election
manipulation that doesn't rise to the level of "fraud," but might be
even more pernicious. And it relies on our trust in the people giving us
the polling numbers. If it turns out afterwards that the polls were off
-- all in one direction -- by almost four percent, I suppose we
all just shrug and move on to the next election, yes?
"It wasn't all that long ago when there were concerns about "fraud" b/c
the state of Florida was called for a candidate while the polls were
still open in the Panhandle. This is massively more severe, and more
complex."
Sean writes:
This admittance of failure, while welcomed, is anything but a surprise. I seem to recall this same news appearing in 2016 with polling experts weighing in on how they might have missed the "shy" Trumper. The big "surprise!" when the polling certainty of a Hillary victory gave way to a modest amount of crow being eaten by various polling firms and media conglomerates. They then spent about 5 minutes in thoughtful self reflection and went right back to failing.
Evidently, the institutions that conduct Polls never learned what biases caused nearly every one of them to be painfully wrong in 2016 and never excised them from their polling and data analytics practices. The situation is analogous to certain aspects of statistical forecasting for business, a topic I've wrestled with professionally for decades. Instead of forecasting what consumer product or category the public is going to "buy", Public Opinion Polling is predicting what candidate the public is going to "buy". They both collect objective and subjective data, employ algorithms, assess statistical confidence intervals, and try to avoid the myriad pitfalls of bias.
The words "unusual magnitude" mean the error levels are next to impossible statistically. Normal statistical variation cannot explain this which would typically point to some consistent and systemic bias. That doesn't mean that every pollster was trying to get a particular result favorable to Democrats. Let me explain - you'd normally find the problem in one of 3 areas - People, Process, and Technology. People from a homogeneous group might construct polls with similar assumptions, wording, and questions. Polling firms may use similar processes within the industry that are blind to certain demographic groups. Algorithm selection may not be well aligned to teasing out accurate sample to population attributes. Possibly they have problems with all three and the effect is sustained and consistent across races and geographies.
All this is to say it's been 4 years, hundreds of elections, thousands of polls and they haven't figured out what their collective problem is. They've been over-forecasting the "Democrat" product and under-forecasting the "Republican" product with apparently no prospect of determining root cause to this day. It's a wonder how a group of firms which are so terrible at their jobs have a market for their flawed projections. I'd have been fired for less.
I'm certain there is intelligent life out there in Public Polling land, but institutional rot has ruined so many formerly respected professions that I suspect it's the case here and GroupThink rules the day. They probably don't have anyone in leadership who would look at some of these poll results and say "that can't be right, do it again".
Assistant Village Idiot writes:
"Yes, I'm sure that after getting 2016 wrong in exactly the same way, underestimating Trump votes, it was impossible to correct for that. I can't think of anything that would have improved matters, gosh darn it. I guess we'll just have to go on forever with inaccurate polls."
Steven writes:
""Polls understated the support for Trump in nearly every state and by an average of 3.3 percentage points overall". I can't help but wonder if these polls were slanted on purpose. Let's face it, some voters won't bother voting if they feel their vote won't make a difference, especially if they believe that their candidate is behind and probably won't win. I do think the media used these polls hoping to discourage Trump voters from voting. I really don't think stating that is far fetched at all. Call me cynical but the media made me that way."
I'll add:
That makes sense to me. That's what I worry is happening. The pollsters ought to be motivated to prove that guess is wrong (whether it's wrong or right).
Alex writes:
"Political polling in this country is an absolute disgrace. When it's not incompetent, it's used by grifters like Luntz to shape the narrative. Prior to the election the narrative was that there was a "Blue Wave" coming. I fully believe that they expected it because of what their pollsters were telling them, and the pollsters were telling them this because the establishment needed it in order to claim that Trump, and Trumpism, was repudiated by the voters. No one wanted to hear that Trumpism was popular with a lot of the country.
"On election night, as it became clear that not only was there no blue wave, but that Trump's coattails were going to gain seats in the House and possibly hold onto the Senate, the Dems panicked and pulled out all the stops to at least salvage a win of the White House. Hence the sudden halting of counting, hence the massive numbers of suspicious ballots in key counties. Hence the fact that Trump lost despite winning many bellwether counties, and somehow losing GA even as he won FL, or gaining tens of millions more votes. But in doing so they burned through a lot of what little institutional credibility remained."
Colorado Dude writes:
"The American Association for Public Opinion Research convened a Task Force on 2020 Pre-Election Polling. Their report says they could not explain why Biden's advantage in polls was greater than the number of votes he received compared to Trump's votes. The report does offer this theory on why 2020 polls overstated Biden's lead over Trump. “If the voters most supportive of Trump were least likely to participate in polls then the polling error may be explained as follows: Self-identified Republicans who choose to respond to polls are more likely to support Democrats and those who choose not to respond to polls are more likely to support Republicans. Even if the correct percentage of self-identified Republicans were polled, differences in the Republicans who did and did not respond could produce the observed polling error.”
"The Harvard Harris poll of February 2021 may point to an explanation of why Trump voters were “undercounted” in 2020 polls. That poll asked respondents whether “cancel culture” was a “threat to our freedom?” 80% of Republicans said it was; under half of Democrats saw a threat. Given the media bashing of Trump, fearful Americans might well decline to participate in a poll."
Post a Comment