Over at 538, Nate Silver has a detailed analysis of poll performance. It tells us that 2020 poll performance was mediocre, but it turns out to have been mediocre in a specific direction:
What's going on? Silver offers this explanation:
We think there’s good reason to expect that these types of mistakes in one direction or another — what we sometimes call systematic polling errors — will be more of an issue going forward. How come? The systematic errors aren’t necessarily a function of the polls themselves. Rather, they’re because in a time of intense political polarization and little ticket-splitting, race outcomes are highly correlated with one another up and down the ballot.
....The old cliche that the Electoral College is really “50 separate contests” is highly misleading in our nationalized, polarized electoral climate. Everything is connected, and for better or worse, you need some relatively fancy math to get a decent estimate of a party’s chance of winning the presidency, or the Senate.
Put another way, in the past random errors canceled each other out, so overall bias was usually small. But if every race is similar, then a single mistake just keeps accumulating in the same direction.
Another possibility is that Donald Trump is a wild card. Presidential polls were heavily biased toward Democrats in both 2016 and 2020, but not so much in most other years. Perhaps this is because Trump voters tend not to answer the phone for pollsters?
In any case, Silver says that phone polls are no longer the gold standard of polling. The tiny rate of responses makes them no more reliable than other forms of polling. No surprise there.
I would assume the "closet Trump voter" was a real thing. And that probably applied to the GOP as a whole. Saying you'd vote GOP was saying you support Trump. I know several people who claim they would never vote for Trump, but when you talk to them they are 100% on board with his general agenda (except for admitting the racist and cruel things out loud).
Maybe, but I disagree about 2016, there was no large polling bias. Go back to 2000 and it fit patterns. Dems got lucky in 2008/2012 with their performance up in the Midwest with Reagan/La follette democrats.
Reagan & La Follette: so much in common!
I don't think it's so much the "closet Trump voter" as it the out in the open Trump voter that pollsters just ignore because they don't vote in any other election. Look at the reaction a lot of the local politicians got at Trump events. No one cared. Sure, they'll vote for Joe Senator or Jane Congressperson if they're on the ballot when the voter goes to vote for Trump, but they aren't going out of their way to vote for those people when Trump isn't on the ballot. They're non-traditional voters who get sifted out of likely voter models because they aren't likely voters when Trump isn't running. They're the cult of personality people. That's why I think Hawley, and Cruz, and the other Republicans with the charisma of dead fish are in for a rude awakening if they think they can just commandeer the Trump base.
Well said.
OOOOR it could be that the Russians, who have been dry humping our election infrastructure for two elections now with the help of a rogues gallery of Trump insiders, did in fact flip votes in key states.
Naaaaaa, the far more likely explanation is that the entire field of statistics has stopped working.
Or maybe they "hacked" the polls?
The math that calculated the polls were fine.
The answers given in the polls were not.
Why would you assume that somehow the Russians were able to flip Senate races, but decided to let Biden win AZ GA MI PA WI, as well as both Senate races in the GA runoff?
It's as simple as this: POLLS ARE GARBAGE.
Get your voters to the god damn voting booth.
In order for the "POLLS ARE GARBAGE" theory to hold, we'd need to explain why they were fine more or less prior to Trump, and why they are fine worldwide when we monitor other peoples' democracies. But suddenly now Everything Is Different™! I dunno, just hard for me to buy that any more than other crackpot theory. Which brings me to...
My Russian flipping conspiracy theory basically goes like this: Whatever they had cooked up, it was heavily rat fucked by a sudden surge of mail in voting, which was a significant reason Trump went so hard against it from 20 different angles. They got the election close, but it just wasn't enough given the mail-in tallies.
Right-wing authoritarians aren't going to answer a poll to admit that they want to destroy the country. But they'll make sure to go out and vote, until their rightful authority figures get rid of voting and just use their rightful authority to punish all of the outgroup people living in their homeland.
It's not any more complicated than that.
Polling science is actually really fragile. It relies heavily on two basic assumptions:
1. Polling organizations can reach a truly representative sample of people who will vote on election day. That was somewhat easier when everyone had a landline and was listed in the phone book. In the unlisted cellphone era, much tougher.
and
2. People respond and are honest. According to what I've heard, 90% of people that pollsters either refuse to answer, or they hang up immediately upon finding out that a pollster is calling. I vote straight Dem and I do the same thing. Zero interest in taking a poll and telling ANY organization how I will vote. Pollsters work on the assumption that the 10% who will talk to them, are both truthful in their responses and also fall in the same voting distribution as the 90% who won't.
Right, but why would those polling problems result in a dramatic shift in one partisan direction? For this to hold we have to believe that the people storming the Capitol to overturn an election are too shy to express their opinions to pollsters.
My point is that it's fascinating to me how plausible this seems to everyone looking for an explanation, yet the idea that nation-state level hackers could manipulate our voting systems is simply beyond the pale. Anyone with any computer science experience knows otherwise.
I think a point is that pollsters underestimate the degree of motivated hate and antipathy people who vote for Republicans have for everyone else, and also the degree to which Trump motivated idiots, people who will almost never vote otherwise, e.g. 'very unlikely voters'.
I think Donald Trump is the wild card. Remember the "skewed polls" from 2012 and "missing white voters" from an election cycle or two before that? The premise was that there were a bunch of conservative white people who favored Republicans but (a) were not being picked up in polls and also (b) didn't end up showing up to vote.
I think they were still being missed in polls, but the orange dipshit got them fired up enough to get them actually out to vote.
Who knows what'll happen in the future w/o Trump.
I doubt it. The big polling misses were up in the midwest, which has a bunch of essentially independent voters, but you knew that in 2000. Obama got lucky in 2012 having Mitten's run against him. The type of guy who would push those voters away.
IANA political scientist, but I favor the view (behind which there is research) that few voters actually swing between parties, and that elections are won by turning out more of your marginal (sporadic) voters than the other party does. I tend to think Trump was the difference in 2016 and 2020 - and in 2018, by his absence from the ballot. His out-loud bigotry, personal attacks, vitriol and bluster - that is, his generally being a huge a$$h0le - attracted a lot of people who don’t often vote, otherwise. Therefore, the bias was not so much in the raw response data, but in pollsters’ likely-voter models.
Maybe, but I think the main difference was Clinton's refusal to spread her campaign like Obama's, who's motto was, campaign where you are weak but can improve. In Ohio, you don't campaign in Cleveland or Columbus, you campaign a county over. Enough of the black bias. Coalitions between the wwc and mwc comes first.
Joseph Rupczynski from the AC Delco parts plant thinks Antoine Carter III is a typical shiftless **** who doesn't deserve the employer health care he receives, let alone government managed Medicare4All.
Antoine Carter III thinks Joseph Rupczynski is a pretty chill coworker who handles his business, & is fun to have a beer with at Tiny's Tap on Sundays during Browns games.
Both Antoine & Joseph receive similar annual evaluations from Arturo Tomas, their line supervisor, & identical performance raises.
They have a lot in common. But in his heart, Joseph knows Antoine is no good & scheming to take whatever crumbs he may have.
Both Antoine & Joseph are in the union, but where Antoine makes a point to miss one Wednesday evening bible study a month to attend the meetings & actively participate, Joseph's attendance is perfunctory, & he remains generally mute, only likely grumbling when two of Antoine & Joseph's younger colleagues, Tang Lo (a 23 year old son of Hmong refugees who graduated the local tech high school with a 3.7 gpa & came direct to ACD after) & Merritt Hawkes (a 27 year old woman who identifies as queer, & used to work as a line cook at a local vegan hipster bar-cafe before finding her dream job as a machinist), pip up about creating a better workplace. Joseph saves his words for whispered asides to his fellow forty fortysomething ACD lifers Gary Szasz & Mike Antonelli, all of whom still can't get over that somehow the 39 year old Antoine has lasted with ACD for 17 years.
Joseph, Mike, & Gary have so much in common with Antoine, Tang, & Merritt, it's a wonder #OurRevolution hadn't come twenty years ago, during the Bill Climpton abetted Dotcom Recession when nouveau snakeoil salesmen like the Pets.com CEO made out like bandits & left the Anti-WTO crowd holding the bag.
& all this, Antoine wonders, while thinking how strange it is that Joseph will have 10,000 words to say about Bernie Kosar on Sundays, but doesn't say peep in union meetings about Brian Cross, the ACD president who is looking to fundamentally restructure retiree health insurance.
Touching and poignant. I laughed, I cried...
I suspect many people didn't want to admit they were Trump voters.
OK, maybe. But why would that affect polling on down ticket races? Were they also ashamed to admit they were voting for their standard issue GOP governor or senator?
The fascinating aspect of this to me is that so many Trump voters were sure that he was going to win, in spite of all the polls to the contrary. What were they seeing, what were they being told? I don't accept that it was just wishful thinking or that they are dumb.
Maybe the Trump rallies gave that impression? Maybe social media? I don't know. I know one very smart, very math oriented guy who told me privately, "I thought Donald was going to pull it off". I didn't probe because in many ways he's a good guy, and I didn't want to push him away. But I am mystified.
No they are not dumb, but yes it was wishful thinking. Many of these same people are convinced the Trump actually did win the election. It's all part of their general conspiratorial world view. The "elites" which somehow does not include Donald Trump, are conspiring to do terrible things. The mainstream media is full of lies. Polls can't be trusted. The only person you can trust is Trump and by extension, anyone who repeats what Trump says. Sorry, but no, these folks did not have any sort of secret insight that pollsters and people like you and me were missing.
Scroll down, my friends, for some crazed Democratic conspiracy theorists.
The fascinating aspect of this to me is that so many Trump voters were sure that he was going to win, in spite of all the polls to the contrary. What were they seeing, what were they being told? I don't accept that it was just wishful thinking or that they are dumb.
I think many of millions of Trump voters have but the flimsiest understanding of how things work in the real world, so their ignorance as an explanation doesn't strike me as fanciful in the least.
I think random phone polling is pretty much done. At best it's useful in a hybrid set-up where folks are contacted some other way, then can do the poll by phone (although I like mail polls more than anything else). The amount of weighting they have to do now is enormous on those.
2012 told me phone polling was done. Big miss against the Republicans. 2020 also had covid related issues that made it worse.
Using maths to predict the probability of who is going to win an election is maybe the single stupidest thing maths has ever been used for. My biggest criticism, but not the only one, is that theses predictions influence the outcome. Nate Silver isn't remotely as brilliant as Nate Silver will have you believe (or he is really smart but should choose a more socially useful activity to use those skills). But if Nate wants to prove how smart he really is, rather than promote how smart is on his website, he could put his predictions in some sort of escrow without publishing them prior to elections. Only after the elections are held do we get to see how accurate his inscrutable models are. But, doing that wouldn't get 538 the traffic and notorieity in the media that publishing these predictions for months in the run up of elections that Nate clearly craves.
At a minimum can we get rid of those insufferable realtime election night needles that swing wildly throughout election nights?
How much is Nate Silver's reputation staked to getting NE2 right in 2008?
According to the polling averages, Biden was supposed to win Florida, North Carolina, and be competitive in Ohio and Texas. Since none on that happened, I can only conclude that some sort or massive voter fraud occurred to give Trump millions of additional votes. I can't prove it, but the numbers don't lie. If we spent the next ten years investigating, I'm sure something nefarious would show up.
Proofreeding fail. Sorry.
No way in Florida. Biden wrote that off by October. I find it funny the Biden campaign's internal polling was so much better. I wonder why???
I know the answer, Shootie: Dominion Voting Systems.
There is another possible explanation for the poll errors; massive Republican cheating.
Yup.
The polling misses, more in margin of defeat than who won, in Maine & South Carolina Senate, where polling showed Sara Gideon & Susan Cornpop a toss-up & had Jaime Harrison down at most five to HAMBISCUITS!, are more sus to me than any of El Jefe's performances (Florida, North Carolina*).
*Shoutout to my Tar Heel peeps, who managed to, in the voting booth, end up abjuring Cal Cunningham** as morally too depraved to serve in Washington given some 10th grade level sexting, while voting to reinstall El Jefe Maximo de Maralago as president, even knowing he raw rawdogs pornstars. The dissonance is real. & it's spectacular.
**In the end, a terrible candidate. & for the good that the Democrat Party in 2028 & 2032 got out ahead of the New John Edwards.
But that suffers the same defect that rightwing theories of vote fraud do-- or rather the exact mirror opposite. If there was massive GOP cheating delivering down ticket races to the GOP how did Biden win? It would be stunningly stupid to let the Democrats have the White House while pulling out the stops to make sure a couple of GOP senators and governors got reelected, just as stupid as the Democrats cheating to achieve Biden's victory but not bothering with Congress and state houses.
Trump did better than he should have, so the obvious answer is that in the presidential race the Republicans weren't able to cheat enough in enough states.
Yup.
Maybe the GQP was hoping the Rona & vaccine issue would continue to be bollocked -- even more, in fact, since the GQP would have every interest to do so at Congressional level -- with the president, a Democrat, & therefore his party, taking the heat in the next midterm.
Given the results of the past two elections, I'm finding myself to be a post election poll kinda guy, maybe a week after polls close kinda polling appreciation guy. But then the past few cycles have taught me that pre-election poling of a nearly evenly split electorate where fealty counts more than platform and balloons, is of dubious value, except as an adrenalin rush.
Given the results of the past two elections, I'm finding myself to be a post election poll kinda guy, maybe a week after polls close kinda polling appreciation guy. But then the past few cycles have taught me that pre-election polling of a nearly evenly split electorate where fealty counts more than platform and balloons, is of dubious value, except as an adrenalin rush.
Was polling that bad? Or was turnout so over whelming that it rendered those 1,200 voters were surveyed useless? How good is a poll in say NC of 1200 voters when over 5 million vast votes? The fine print on most of the polls are basing the accuracy of previous turn outs. That sure as hell didn't work out too well for them did it?
So what happens in 2022? If we return to normal mid term turnout levels who does that help/hurt? I think the GOP in Georgia made a gross assumption that DEMs dominated mail in and absentee voting. I'm guessing that republicans thought that THEY would dominate same day voting enough to overcome that/
But turnout on voting day was heavy for DEMs as well.
So now if mail in voting levels decline AND in person levels decline what do we expect on voting day 2022?
The answer? More BS leading up to the election with the various factions falling all over themselves when a poll paints a positive picture or a positive trend for their particular party. "Oh Obama carried that district by +5 in 2012 so the polling says Biden is at +3 that should be enough to carry that district."
When turnout hits 74% in NC and we haven't seen that since the 1950s I have to wonder where these hot shot pollsters are getting their info from?
What was so different in 2020?
The COVID Plague
A lying president
An aged opponent
Both parties split
More money raised and spent
More mail in ballots allowed
Enormous surge in both mail and in person voting
Sure polling based on previous election turnouts were gonna catch all that.
If there's the same bias everywhere, wouldn't that make it easier to figure out why the pollsters keep getting it wrong? They seem to have been off most places in 2016, but did it again in 2020.
To really find out what is going on, it would probably be necessary to contact every single voter in some hopefully representative geographic sampling, to find out how the sampling is going wrong (if that is actually the problem). That may be beyond the capabilities of the pollsters.
2012 was off like 2016
+1 drunken Peggy Noonan seeing double the number of Rmoney/Ayn Ryan yardsigns from her chauffeured car
Or we could cease spending so much energy on polls.
I mean, yeah, I think we all figured that out on election night.
I've read a number of plausible explanations regarding polling inaccuracy suggesting the universe of people who are actually willing to give their opinion when polled is itself very heavily biased in favor of voters likely to support the Democratic Party. This obviously must be corrected for. In 2016 (so the theory goes) the analysts and pollsters were barely cognizant of the phenomenon. They were aware of the phenomena in 2020, but didn't correct their models enough. (I'd say it's possible the phenomenon of "MAGA unwillingness to be polled" thus intensified between 2016-2020).
So, who knows? Maybe polling in presidential contests has indeed become relatively worthless for the foreseeable future. But I also think it's possible by 2024 the models (both on the polling and aggregation side) will have been recalibrated to the point of improved accuracy and predictive value.
I'll also add: I've seen seemingly plausible claims that, in addition to the dynamic above (polling participation reluctance), those MAGA/GOP leaners who are willing to be polled sometimes exhibit a tendency to lie about their voting intentions (in an attempt to muddy the polling waters). It's actually a pretty savvy strategy on their part, if true and widespread.
Just caught a clip of David Shor via Twitter: he claims a big source of additional polling error this past cycle were lockdowns: sounds as though knowledge workers -- who lean Democratic -- were home, and easily reachable, and perhaps starved for social interaction. So they eagerly responded to polls (way out of proportion, presumably, to Trump-supporting working class voters).
I think that is unlikely. Ability to work from home correlates pretty strongly with income, at least for low to moderately-high levels. Pollsters work with stratified samples — if they have too many high-earners in their raw samples, they would throw some out at random until proportions in the sample match proportions of the same groupings (gender, age cohorts, incomes, race/ethnicity) in the population of recent voters.