The polls messed up
statmodeling.stat.columbia.edu
statmodeling.stat.columbia.edu
I would argue 538 made a similar error[1], but unlike Gelman, Nate Silver is a less likely to admit their model was off, and consequently examine/discuss the underlying factors that lead to this issue.
[1] I don't know if error is the right term here, but basically the fact that the actual election results seem to be landing near the lower end of the predicted 95% intervals for both the 538 and the Economist models.
EDIT: Let's look at a few of the tipping point results from 538 [1]
Florida: 50.9/48.4 -> 47.8/51.2 (2.8 diff) [99% reporting]
Arizona: 50.7/48.1 -> 50.7/47.9 (0.1 diff) [86% reporting]
Texas: 48.8/50.3 -> 46.4/52.2 (1.9 diff) [85% reporting]
Georgia: 50.1/49.2 -> 49.1/49.7 (0.7 diff) [98% reporting]
Minnesota: 53.7/44.6 -> 52.6/45.4 (0.9 diff) [99% reporting]
It's also worth noting that a lot of these are still at 85-95 percent reporting, so it's not over, but yes I don't see anything that's very off, even Florida that was so off was only 3 points off. I do consider 3 points to be well within the toss-up category.
[1] https://projects.fivethirtyeight.com/2020-election-forecast/
EDIT2: I think the main thing that surprised people is how in some places, mail-in votes came first, whereas in some other places, mail-in votes are coming much later. This really skews the early data but as things settle, everything is coming much closer to the predictions. But this is something 538 and others have been warning about since May or earlier.
That would make sense if some parts of the country had errors one way, and others another way. i.e. random error.
But if every part of the country has an error, all in the same direction, then there's something larger going on.
Did they say that their confidence interval was confidence against random error? I don't think that's a requirement of a confidence interval, I think it's certainly valid to encode e.g. the chance of "shy Trump" respondents in a confidence interval.
Figuring out the dependencies between demographics is how you come up with a useful model rather than just simulating everything independently based on raw polls numbers. Aka Michigan and Wisconsin have a lot in common where Florida is influenced by wildly different racial, demographic, and economic factors.
Back to my point, click Trump for Florida (which was considered a 31% chance) and the prediction looks a lot like our current map.
That's exactly why their probability forecasts tend to be more conservative than other forecasts. If you see a whole lot of possible paths for a candidate, and assume the errors are random, then that candidate will almost certainly win. If you know they might be correlated, then you know that a huge number of paths may shift at once.
As Silver pointed out in 2016 (and why his model then gave Trump a much better chance than the NYTimes model), historically, when election outcomes differ from polls in either direction, the polling error between states is correlated, not independent. That is, it is, to all evidence, a historical fact that poll-based predictions must account for that within election cycles, there is systematic polling error, even if the direction and magnitude of that error is not consistent between cycles.
Diff on final 538 -> current results
Wisc: ~8pt diff
Michigan: 4-6pt diff
PA: still counting but most likely 2-5pts off
Wisconsin: 53.7/45.4 -> 49.6/48.9 (3.8 diff)
Michigan: 53.5/45.5 -> 50.5/48 (2.7 diff)
PA: 52.0/47.3 -> 48.1/50.7 (3.7 diff), though will shift quite a lot as you mention
So even then it's still closer to 3-4 point, not 4-8 as you mention.
I.e. in Wisconsin, the forecasted margin was 8.3 points and looks to be closer to 0.7 points which is a difference of ~8 points as GP said.
Why do you say this? He did plenty of writing on what he, the site, and the whole industry got wrong in 2016 for Trump's win in both the primary[1] and general[2][3].
I really don't understand the hate Nate Silver gets from basically all sides.
[1] - https://fivethirtyeight.com/features/how-i-acted-like-a-pund...
[2] - https://fivethirtyeight.com/features/the-real-story-of-2016/
[3] - https://fivethirtyeight.com/tag/the-real-story-of-2016/
I don't mean to come off as a Nate Silver hater. My criticism is made with a lot of affection, and awareness of how much he discusses the role of uncertainty in his work. But the Economist team has made some legitimate critiques of some wacky tail behavior in the 538 model[1] which Nate Silver has basically dismissed without engaging, whereas the Economist team has written plenty critiquing their own model[2]. I mean this is Gelman's whole schtick about the scientific process: he thinks that errors are to be expected and should be transparently discussed and fixed and iteratively improved over time. That's partly why they've released all their code on Github, and Gelman has sort of poked at 538 for not doing the same with their code.
[1] https://statmodeling.stat.columbia.edu/2020/10/24/reverse-en... [2] https://statmodeling.stat.columbia.edu/2020/10/24/reverse-en...
It’s also worth noting that 538 doesn’t conduct any of the polls that feed into their forecast, and so aren’t pollsters in a relevant sense here. Their entire reason for existing is that they agree that there is systematic bias in polls so they attempt to model it.
The magnitude of their errors are shocking, stunning, and increasing over time. I can't understand how anyone can ever look at these again without a canister of salt. If Trump won in a landslide after they predicted Biden up by a landslide but assigned it some kind of probability you're going to give them a pass? Of course not.
The fact that there are so many polls and they are all wrong means that there is something foul at play at the heart of this entire enterprise. The most charitable thing I could say is that they are completely useless, a toy metric for amusements.
Given how negative you are about them “over time”, I wouldn’t have expected you to take a detailed look at their model this time around. Out of curiosity, what drove you to take a look this time around so you could claim their errors are “shocking, stunning, and increasing”?
[1]: https://projects.fivethirtyeight.com/trump-biden-election-ma...
But that range is so huge as to be useless.
Yes, it only predicts probabilities and can only be as accurate as the underlying data (which seems to be Bad.) But they don’t show/market it as being “anything could happen, this model is wide open and should only be used to rule out crazy outliers”.
In my opinion 538 nailed it in both 2016 and 2020. In 2016 they said Trump needed a full standard polling error to win. In 2020 they said he would need more than a full standard polling error.
The polls were wrong, and 538 told us what would happen in that scenario.
But really. If the polls are so unreliable then at some point people like FiveThirtyEight or the Economist need to understand that and stop incorporating them into models (which may mean not attempting to forecast at all). Just pointing the finger of blame around only gets you so far.
My opinion: I think deep down these people feel they are rationally wrong for voting in a president that promised to, and now is actually attacking USA's democracy and freedom which used to be touted as America's strong point. They know they're acting on instinct, and people tend to hide their instincts in public.
> My opinion: I think deep down these people feel they are rationally wrong for voting in a president that promised to, and now is actually attacking USA's democracy and freedom which used to be touted as America's strong point.
I think you underestimate peoples distrust of mainstream politicians.
Many find Trumps suggestions abhorrent, but still support investigations. Think if this happened to "your" candidate:
- There's no doubt that Twitter and Facebook has been uncritically spreading false rumors about Trump only to turn right around and supress the H.Biden leaks, even after it was clear at least some of the juicy mails was cryptographically verified. If Twitter had been posting a warning everytime someone pretended the "fine folks on both sides" hoax then they would have been in a better position now. Same if they had come up with the rule against hacked materials when Trumps tax papers supposedly leaked.
- Ballot collections
- Sweeping changes to election laws
etc etc.
I dislike Trump, but it is hard to not get some sympathy for someone who has managed to get a number of really fantastic peace deals, have finally started to improve lives of people in the midwest - without hurting the coasts - and yet is hounded by almost every mainstream media outlet.
Do you have a link that refers to this cryptographic verification? Last time I saw this talked about on HN, the summary was that there really wasn't even an ounce of evidence, and that the people who supposedly had it were not sharing it for no good reason, so I'd be interested if there has actually been a proper verification of this.
> have finally started to improve lives of people in the midwest - without hurting the coasts
What do you mean, and what's the source for this? Again I'm not that informed about this topic, so I'm genuinely curious about a source. That's also the problems when half his tweets make the headlines: that skews the information towards what he says rather than what he does, and given the votes of this election I tend to think that (at least the perception of) the latter is better than the former.
I don't know what GP is talking about here, but what I remember is at least one person mentioned in the emails confirmed to the FBI that they're real.
And yes, social and mainstream media has inflamed everything by misrepresenting Trump's side.
But let's not remove his guilt: he has said (in a speech that SHOULD be censored by both mainstream AND social media) that he won this election, and won't accept its results. This is ACTUALLY fascism (a word that's thrown around wrongly many times): for years the people's voice has been heard, and now we have a leader (and many supporters) that don't support a democratic system anymore. That's fine, democracy isn't the only available system, but it's a hallmark of the USA.
Totally agree, so let me clarify my position here.
I'm not defending the idiotic things Trump and certain Trump supporters say.
I didn't want Trump either and I think I started soon after last election (on another account) to say that I was afraid we would get more of this if the rest of the politicians didn't wake up for how massively unpopular they are in certain circles.
Yet here we are.
Yes, a lot of people love Trump. But very many hold their noses and vote for him despite thinking of him as a constant embarrassment.
Why?
My theories:
- BLM has been campaigning relentlessly for Trump - despite their intentions. Just think straight through this example: Your local shop get looted or someone you know loses their job or their house because of this. Will this increase the chance that you vote for the party who doesn't condemn these guys? Or the other party?
- Mainstream media, who doesn't care at all about you, even ridicule you as rednecks, will you vote for or against the candidate that they try to push.
- Will you want to support the party who has still not apologized for their "basket of deplorables" comment?
- They band up with none other than Lady Gaga, produces a video perfect for angering these people at release it a day or so before election day!
- Even a huge number of black people went to Trump this year. Guess why? No respect for them either!
So no, I don't like Trump. I don't defend his more idiotic ideas.
But I do think Democrats and many HNers have a lot to learn before they understand why Trump is so popular despite his truckload of flaws.
And no, it is not that everyone who voted for Trump is dumb racists.
That has been addressed before, and it had seemed unlikely—there are races where more "normal" Republicans on the ballot (for house or senate races) were polling worse than Trump. So it might be that they're ashamed to vote for Trump, but why would they be ashamed—let alone even more ashamed—to vote for a normal Republican?
How do you define the mythical "standard polling error" though? They have historical data, but polling techniques are always shifting.
Per the article:
> Suppose we’d included wider uncertainty intervals so the outcome was, say, within the 50% predictive interval. Fine. If we’d given Biden a 75% chance of winning and then he wins by a narrow margin, the forecast would look just fine and I’d be happier with our model. But the polls would still have messed up, it’s just that we would’ve better included the possibility of messing up in our model.
If presidential polls are consistently off by large margins, and 538 considers those margins to be unlikely every time, (which seems to be where we'll land this year as well as we did in 2016), then maybe it's not just the polls that are wrong, but the 538 assumptions about poll accuracy as well.
2016 on its own doesn't condemn the 538 methodology, but 2020 is starting to look like a trend. Yes, outlier events happen, but if they happen consistently... well, maybe your definition of "outlier" is incorrect. The sample size on presidential elections is small, and maybe there's just too much uncertainty for a model like 538's to end up being useful.
I get what you're saying, and I agree—but then, what is to be done? 538 doesn't do polls themselves, they just aggregate them. If the polls are incorrect, what do you replace them with? And how do you make that scientific?
They also assess their models in aggregate, but that obviously isn't applicable to the Presidential model in isolation, since it has so few runs; OTOH, it does validate their broad approach while leaving open the possibility that there is something especially wacky with the Presidential model. But without lots of runs of that model, you can't really clearly single that out from chance errors, and if there are isolated instances of unusual effects that show up only in the Presidential election and only, say, for one or two cycles (e.g., if there is an effect around Trump which is atypical but not new normal) that aren't accounted for but also aren't long-term features, you probably can't even reliably detect them once there are large numbers of runs of the Presidential model. Unfortunately, validating a statistical predictive model isn't itself a a probabilistic exercise subject to small-N problems.
I also don’t think they hide this uncertainty. The model results I’ve seen on FiveThirtyEight always feature a distribution that shows the electoral margin could end up being anything in a wide –X to +Y range.
Something still seems off about that. How do you do that many large polls of that many primaries and have such error still? Sure, there's going to be sampling error, hitting unrepresentative pockets and so on, but ... why can't they get any better after all this time (i.e. there was a previous Trump election) and research on the demographics?
- Nate Silver
https://twitter.com/NateSilver538/status/1324149741463228416
That's literally the opposite of what he's doing in the post you are replying to...
The irony is that Gelman has spent quite some time lately criticizing Silver et al. for forcing too much uncertainty into their model.
IIRC Silver got the same criticisms just before the 2016 election, too—that the only reason 538 didn't have Clinton's odds in the upper 90s, and had Clinton losing Florida, were to drum up interest and attention at the end of the race.
It's not surprising that technologists from the same schools and cities all failed to understand this again.
The filter bubble affects us all, including you, including me, including the people you think are completely brainwashed. I just happen to VPN with new browser sessions a lot and get to see it.
I also don't block people with different ideologies lol so there's that.
I've knocked on doors, and worked phones for various campaigns, and you learn a lot. The most important thing you learn, is that voters, and especially undecided voters are extremely ignorant (they are undecided for a reason) and are extremely susceptible to believe misinformation, or rumors. If you're not out there talking and debating with them, no one's going to knock them out of their bubble.
The Biden's campaign decision to not knock on doors, talk to voters, and get them registered just seems like a disastrous error to me. And yes, I know there's a pandemic going on, but there are safe ways to talk to people in person.
The parties never address or even acknowledge the criticisms about them. Choosing to instead imagine that everyone that doesn't believe what they do, already, actually believe the opposite. When people really just weigh different things and factor different things in, and aren't consciously thinking about whatever another person considers a trigger.
Critical error.
Blocking half of the people in the country and then being surprised that half the country didn't believe the same thing you did is not a productive use of energy.
When moderates and centrists can only be found way passed the downvote threshold for visibility, when the reality is that they are the only ones listening to the spectrum and can perceive what's going on, saying the least inflammatory things, but are perceived as a false negative by both parties because they don't sound like a party member.
The Democratic party did address and acknowledge criticisms in 2016, which is why the party leadership put their thumbs on the scales for Clinton over Sanders. Indeed, it's why they have super delegates to begin with--to prevent a Trump-like nominee. But doing that burned them.
The Republican leadership also tried to push back against Trump, at least very early on, but everybody fell into line fairly quickly once it became clear Trump was the front-runner.
At the end of the day a party can't simply wish away its fringe subgroups. At best it can occasionally give their leaders a cold shoulder, like Pelosi did with Ocasio-Cortez. For Democrats the cancel culture fringe and (much larger) equality-in-outcome movements cause tremendous difficulty in keeping moderates from straying, yet there's no easy way to assuage those moderates' fears.
The Republicans can keep winning with their white majoritarian populism even in a diverse America because the criteria for being "white" is rather flexible, especially for non-African Americans. For the plurality of Americans who have very short term memories, and who perceive everything as a zero sum game, the easiest default choice is choose what would be good for the "white" majority, or at least what they image it as. So long as minorities, particularly latino and asian, make some minimal progress--and with a growing share of the population that's inevitable--the GOP can continue to coast on their fear, uncertainty, and doubt strategy. "White" and majority will be synonymous in perpetuity. Republicans don't have to change, and with Trump they finally figured that out.
People keep repeating this mantra that the parties don't reflect the majority of Americans when what they really mean is that the parties don't reflect their own concept of America and American sentiments, and their own particular set of preferences. Even when we can all agree on general goals--liberty & justice for all--what matters are the particulars, and Americans have pretty radical and divergent ideas about the particulars. For example, most people agree we need stronger job security, especially for blue collar, semi-skilled labor. But the word "union" has become a dirty word even among many Democrats, and conservatives have preferences for "less regulation" and the "free market" at least as strong as they do for job security, preferences which are fundamentally contradictory.
This perception is mistaken: https://academic.oup.com/qje/article/135/2/711/5687353
> White and Hispanic children have fairly similar rates of intergenerational mobility.
As a result, incomes for Hispanic and Asian immigrants rapidly converge with those of whites. Cubans and Vietnamese who came here in the 1960s and 1970s as refugees are already fully converged economically with whites.
Economic convergence within a generation or two is not universal. Muslim immigrants to France suffer from multi-generational poverty. Asians and Hispanics have very real incentives to be wary of changing the current economic system. At least to the specific aspect of reaching economic parity with whites, it works extremely well. (There is a fair debate to be had about income inequality and social mobility writ large. I’m focused here on race.)
Also, unions are bad for immigrants, who tend to be excluded from the seniority system.
That paper covers the period 1989 to 2015. I based my idea that non-black minorities become "white" partly on the realization that in recent decades hispanic populations have integrated quite well, and are a growing Republican constituency. And I came to that opinion in the very early 2000s. (If not earlier--I distinctly remember mentally critiquing the assumption that both Francis Fukuyama and Samuel Huntington made in their late 1990s books regarding the political effects of a future hispanic-heritage majority.) I'm a [late] Gen X'er who grew up in the Midwest and Deep South and from my perspective hispanic culture and identity was already going mainstream, to the extent it was even distinguishable as alien. By contrast, almost all the adults in my community quite clearly held strongly and uniformly to old, persistent stereotypes of hispanics.
If you look at hispanic mobility over previous eras, I'm pretty sure it would look different. (Though I could be wrong--I don't recall seeing hard data, just plenty of anecdata, including centuries of received cultural narratives.) That historical disparity and relatively slow integration is presumably why older scholars just assumed a growing hispanic population would lead to a cultural and political shift. And that old scholarship still influences modern assumptions among the liberal intelligentsia. But, of course, the fact that their mobility is now much like that of whites (notwithstanding a relative decrease in white mobility as compared to earlier eras) is evidence that whiteness is sufficiently adaptive and inclusive to forestall the racial reckoning--i.e. between white and black identities--that Americans have been facing for centuries.
In many ways the liberal achievements of the Civil Rights era sowed the seeds of its own political destruction. Not simply because it pissed off a bunch of white people leading to Nixon's infamous Southern Strategy, but because it opened the doors of economic and, more importantly, cultural opportunity to Latinos and Asians, much more so than for African Americans. With those two large (and growing) groups increasingly knitted into the fabric of the American white mainstream, much like the Irish, Italian, and slavic groups in earlier decades, there's less pressure to maintain the social efforts regarding the cultural, political, and economic enfranchisement of African Americans.
While not quite the same, I see many parallels with sex and gender equality, too. Anyone who thought the enfranchisement of women would bring a substantially liberal turn in American politics 1) never paid attention to the Temperance movement in history class, and 2) never took a peek inside a church to see who was sitting in the pews.
As I grow older I've come to better appreciate the critiques of mid-century African American writers: prejudice and racism is fundamentally about power.[1] One not-so-obvious consequence is that empowering the disenfranchised is more likely to lead to the once disenfranchised adopting and adapting prevailing prejudices than for those prejudices to be fundamentally overturned, simply because the salient effect of those prejudices is to preserve power by distinguishing the powerful. You could even see this with the gay marriage debates--it was astonishing how quickly gay culture began espousing what were once "heteronormative" ideas about relationships and family matters as gay marriage became more mainstream, often leaving their bi- and trans-gender allies in the lurch.
[1] No need to get into the specifics of what power means or the mechanics of how it manifests. I certainly have no sympathies for Marxist philosophies. I think many writers on racial and sexual equality gravitated toward Marxism because Marxism provided a vocabulary for discussing power. But it was the wrong vocabulary, and the dissonance is rather apparent in hindsight. Which isn't to say there isn't a huge overlap between racial and class power struggles. See, e.g., https://www.commondreams.org/author/adolph-reed-jr
Infamously in 2016, the polls did not take into consideration or normalize for education level. This resulted in a oversampling for for Hillary supporters. So as we continue to see polarization due to personalized media, it may turn out that typical demographic markers are no longer as effective in normalizing these sample polls.
But I already know of the existence of a sizable number of conservative intellectuals. I don't see how that helps me extrapolate to whether it's going to be 60 or 70 million "regular" Americans that are going to vote against my candidate. (Not to mention that the difference between a razor-thin margin, and a huge Electoral College blow-out that would have resulted in articles like this not being written, is only a few hundred thousand carefully-placed votes.)
Quillette is part of the so-called “intellectual dark web” that prints articles too dangerous for places like The New Yorker and The Atlantic but in a similar vein. I would categorize it as a classically liberal magazine in the sense it is free thinking but the mainstream liberal press is too orthodox to print anything that challenges them.
Reason is the most mainstream libertarian news organization but is increasingly popular for its free thinking politics and resistance to woke trends. Highly recommend.
The term "intellectual dark web" merely means "disagreeing with liberals." I'm not sure how else you'd describe this beyond "socially conservative" -- it's the exact same stuff that Fox talking heads rail against day in and day out.
Idk it seems like liberals act like they're morally superior. That's insufferable and why minorities, etc. are voting against them.
No one likes being told they're irredeemable and need to repent.
And insofar as "the liberals" can't introspect, they're not considering that they're deluding themselves into thinking they represent the working class when they largely represent the affluent, educated people who know the intricate socially profitable phrases to say among their conformist city friends.
So basically - are liberals so infallible that "disagreeing with liberals" means you're socially conservative? Hogwash!
I mean, all the stances I posted above are ones firmly held by, say, Ben Shapiro, someone who is often considered "intellectual dark web," and absolutely considers himself "conservative."
No where did I suggest that liberals were infallible. I was simply pointing out that all those themes that I saw repeated day after day on Quillette were common grievances of socially-conservative writers. What's wrong with saying that?
You realize how presumptuous and lacking in self-awareness it is to dismiss criticism as essentially "clearly the enemy," instead of actually considering the content of their argument?
Would it really be hard to expand your mind by considering even the hypothetical possibility they're right?
Nowhere did I suggest that they are or aren't right about these issues.
And when I say Ben Shapiro's beliefs are socially-conservative, that's his language, it's nothing to do with "assuming" it's coming from conservatives, or "the enemy," or anything like that.
Again, I'm completely puzzled by your apparent horror that saying that the ideas expressed are "socially conservative." It sounds like you're the one who has a reaction to the "conservative" label, rather than seeing that all sorts of people may have a constellation of liberal beliefs and conservative beliefs.
I've been a long-time liberal and I sharply question aspects of all those things too as do many other long-time liberals I know. To suggest that people who aren't in agreement with the progressive left are social conservatives is to be utterly disconnected with what the general public really thinks, as current election results are showing.
Voter engagement is supposed to be high, higher than it is even this election.
Instead, one side will narrowly 'win' and then take control. At that point, it's time to pay back the loyal fringe of their base.
And when people view an election as a game to "win", why not lie to throw off the other side?
It's a basic art of war tactic.
Either conservatives have started lying to pollsters about their voting intentions or conservative officials started manipulating the vote totals in their favor.
Take your pick as to which possible explanation is most palatable and/or most probable.
If you don't have a representative sample of the population your poll will be inaccurate.
Do you have any reason to believe that pollsters and other statisticians are so ignorant of their own field not to understand this?
So they think they have a representative sample, but they really don't.
It can be explained simply by pollsters mainly living in large cities, and not really understanding rural Americans.
But that would start costing them elections, so instead they coddle and abet the worst of them.
What do you call 9 people staying quiet when a tenth is being blatantly racist? Ten racists. Take it from me, I saw it in action growing up white in Mississippi.
Of course pollsters already know this, so they weight the samples to reflect underlying demographics, but it gets tricky to do this accurately if your surveying response rate is 1% to 2%.
And of course... the forecast modelers know this. Which is what Gelman discusses in his post. Given that they now have 2016, and 2020 as examples of presidential polling errors, modelers need to shift their priors in a number of ways to account for what looks like a persistent undersampling of GOP voters.
That doesn't actually help unless the difference in response rate is explained by the underlying demographics. If it is correlated with preferences on the issues polled even controlling for non-preference demographics, it's not going to be correctable by that means, nor is it going to be measurable so that it can be corrected.
OTOH, if it is a long-term consistent effect tied to partisan presidential voting preferences across election (and not, e.g., tied to the particular transitory trends, or evolving significantly over time) you could, as a poll-based forecaster, correct for the effect without measuring the correlation in preference and response rates by adjusting how you draw forecasts from polling data by correctingnbased on empirical election results and their relation to polls.
In the article, he mentions the simplest and probably most correct way to do this: reduce the confidence these models have in the accuracy of the polling. The top line number he threw out as feeling "good" about was 75% for a Biden win.
As it is, it's becoming difficult to find any value in these models because the events they represent as outliers seem to be consistently occurring. On the other hand, if they adjust to more accurately reflect the uncertainty in the the polls, eventually everything starts looking close to a coin toss - in which case they have no practical value, either.
Propensity to participate in polls having even a slight correlation to preferences on the issues being polled will create a systematic polling error, however, and it's an effect that is difficult to measure because how do you measure the preferences of nonrespondents?
Without coordination a bunch of independent actors are generally predictable, because any random choice they makes averages out to the mean choice. With coordination the pollsters should be capable of finding out about it too.
So I think lying to polls only works while it's still rare enough (in terms of # of elections that it happens in) that the pollsters aren't accounting for it.
In a perfectly organized system (rarely found in reality) the counting would have finished within 2-3 hours after the end of voting.
American-style slow results have more in common with the kinds of places that need time to "massage" the results than with the electoral processes in most democracies.
The only real impediment/threat to voting is the Republican party doing any and everything they can to limit peoples’ ability to vote.
The overwhelming problem with American elections is electoral fraud, a term which encompasses such things as preventing eligible voters from voting, not counting legitimate ballots, voter intimidation, and jerrymandering. These problems cannot be fixed through measures intended to minimize voter fraud as they are the result of the state, and/or parties, abusing the democratic process rather than the citizen abusing it.
[0] https://www.cnbc.com/2020/09/24/fbi-has-not-seen-evidence-of...
You need to commit a lot of voter fraud to swing an election, it has to be a very close election. Like one state and 20,000 to 100,000 vote. A once in a 30-50 year election. You need to figure out before hand which state will be swing state that decides the election.
Now you have to commit voter fraud 50,000 times in that state. You can't do it in person because the numbers are too larger so you have to do mail-in. How do you find 20,000-50,000 people across the state who you can get a mail in ballot for, steal it, and turn it in without anyone noticing.
This is basically an impossible task that even if you pull off has a low chance of making a difference, and a large chance of sending you to jail.
I would trust mail ballots because they are delivered to your residence. Pretty hard for a foreign actor to fake/game that kind of system -- or at least, harder than posting a bunch of inorganic FUD comments on social media such as HN.
Here’s a recording from 2019 elections in Canada. The last seat was determined 4 hours and 44 minutes later, and that is unusually delayed, but still a record time compared to U.S. counts: https://youtu.be/GYrY_945co0
Of course it might vary by population, but it shouldn’t need to be this inefficient to run or count advance votes, etc. Elections Canada might also make a difference rather than going state by state. And we don’t have a step after where votes are cast, etc., instead it’s a simple first past the post system.
Some background:
https://www.cbc.ca/news/politics/canada-votes-2019-election-...
https://www.elections.ca/content.aspx?section=vot&dir=faq&do...
As a Canadian I’d often watch as polls close in different time zones and then roughly a half hour later, results trickle in. Usually within the first two or three hours, up to half the country’s votes, by population (at least), have been counted and results shared — even as polls are still closing in other parts. It’s expected that by 2am at the latest you could usually know the outcome, even in a contested race. The exception would be minority governments, which can form their own majority from coalitions over the next few days after a Canadian election. And they do still count all the votes by hand in Canada.
This kind of vote count inefficiency reminds me of tax code inefficiencies. Just because taxes are complicated to file in the U.S. does not mean it has to be that way in the rest of the world! :)
That's 51 different voter registration laws, 51 different sets of candidates, 51 different ballots, 51 different voting procedures, etc. Each state's laws have been subtly (and sometimes not so subtly) changed to provide a benefit to whoever was in power at that time.
There's a hell of a lot more than that, the ballots include local votes as well.
> The forecasts were off. We were forecasting Biden to get 54.4% of the two-party vote and it seems that he only got 52% or so. We forecasted Biden at 356 electoral votes and it seems that he’ll only end up with 280 or so.
> – The polls messed up more in some states than others. Florida is the clearest example of where the polls got it wrong.
No major poll I saw had Trump anywhere near these numbers. Clearly something went wrong.
Example: Poll saying Biden winning Wisconsin by +17: https://www.wbay.com/2020/10/28/biden-leads-trump-by-17-poin...
Actual result: Biden currently winning by .6%. If the poll is off by 16.4% its a garbage poll and needs serious re-evaluation.
What I'd be more interested in is an article with more meat than this one, something that analyzes a single model and talks about what, if anything, went wrong. 538 has already mentioned that they plan on doing a post-mortem.
The problem with presidential elections is that they happen infrequently with a large time gap between, and we only have a few decades of useful data to model from. For me, as a non-statistician, it's difficult for me to look at the numbers and know whether this one particular reality was adequately accounted for by the model. 538 gave chances of each candidate winning - they didn't state that they'll have exactly x margins of victory. Nate's stance before the election was that Biden could survive a significant polling error, and that's about where we seem to be now.
I'm looking forward to the 538 post-mortem.
Popular vote margins will also change once the big blue states finish counting which might take a while.
Sure Florida was way off, but correct me if I’m wrong: If you rely on 95% confidence you can expect one prediction in 20 to go outside of it. There are 51 elections to the electoral collage, so you can expect two ore three states/districts to be outside of it.
This time, we might end up in the counter-intuitive territory where the poll were "wrong" even if they pointed in the direction of the winner, because the results are outside the margin of error.
The problem is that Presidential elections are always pretty close. It has been almost a half century since any candidate has gotten more than 60% of the popular vote.
If the result is always within the margin of error, how useful does that make the polls?
From the perspective of ad clicks and news readership, very.
That's what Gelman's post is about. There's something wrong with polling methodology and statisticians' models, and it's time to see what.
The problem is that it's very hard without more data than can be gathered, and it may be a moving target.
This same fact (or, rather, highly likely conclusion of substantial, correlated polling error indicating a systemic error) was, remember, widely observed in the wake of 2016 and the common hypothesis (for which fixes were implemented in 2020) was that the problem was pollsters not weighting samples for education.
I'd like to see more polls after the election that ask people "who did you vote for". If those polls are skewed against actual results in the same way as pre-election polls, that might point in the direction of the problem.
I believe in a constitutional amendment that guarantees all citizens the right to vote, including convicted felons. Not only that, but every citizen should be required to vote (or face a fine, like in Australia). Forget voter registration pushes and poll vans. There should be a whole department of government whose job it is to go out and collect your vote, the way they collect your taxes.
Until machines are taken out of the equation, I find it difficult to consider any election legitimate.
The general public could swallow this because they're completely clueless about computer security, but I'm surprised HN readers would be so trusting of these machines when there's been expose after expose (many of them landing on HN's front page) about how vulnerable they are.
Only 8 states use electronic voting anyways.
As it stands now, that sounds pretty good, however we have to wait for the end of the count of course.
And in general, I am a fan of not overly editorializing mathematical output. Nate Silver made on his podcast today the point that the popular vote polls look quite good right now, because mail ballots in very blue states like NY are still be counted. However, I kinda enjoy how the forecasters are beating themselves up right now, even if I think that is probably somewhat premature right now.
Yeah, I heard a pundit on NPR describe a polling bias that arises from individuals' distaste for the news agency conducting the poll. I would think you could model this phenomenon (and/or create a polling spinoff marketed with a more independent brand).
However you slice it, it seems like the polls just aren't painting a picture as accurate as they have in decades past.
They may be the ones who got it the most right ... Again. They also discuss some of their methodologies:
https://www.nationalreview.com/podcasts/the-editors/special-...
They could have been really dramatically off in many states, or in the end they could be really, really close. In most states, we don't actually know yet!
He adds a P.S. suggesting Florida as a counter-example, but polls could be off in Florida without being off elsewhere. It just seems too soon to be doing analysis of a moving target.
Anyone knows which website I'm talking about? Or has a link?
What went wrong in Florida shows up clearly in the results for Miami-Dade county.
The Biden vote in that county was way down on the results acheived by Hillary Clinton and that then meant Florida was lost.
And in my opinion this shows exactly why polls can go wrong.
In the last days of campaigning Trump ran hard on the idea Biden was a communists, hell bent on turning America into the next Cuba.
That theme resonated strongly with the Cuban-American population of Florida.
I suspect in the early days of polling that Trump fear campaigning was just simmering away, having little effect.
But it really came to the boil on election day.
At best, polls are a simulated election, wrapped in a facade of a forecast. News orgs offer them as a surrogate for factual election results.
It is unclear how presenting this inherently unreliable mess as news helps any voter choose a responsible candidate.
It also may have an effect on turnout and peoples votes, but that’s a lot more wooly.
“Internal” polls are polls that the campaign has contracted a polling organization, quite often an organization that does its own (or is contracted by media outlets to do) public polling, as well.
Political campaigns are comparatively ephemeral, they don't have time to build polling organizations from scratch.
The demographic outcomes of this country are getting more and more unpredictable.
It looks like the FiveThirtyEight popular vote forecast will end up and few points off of the final count, but it will be within the confidence interval.
There are other complex factors, too. I keep seeing polling/forecast methodology explainers mention calibration to historical data so that if the same method was applied to historical elections then it would have been accurate. But Trump is a way different candidate than any other Republican (or Democrat); he is appealing for reasons that other politicians aren’t. It may be that no historically calibrated forecasting methodology will be accurate when you have a Trump.
I guess all this depends on the disparate sorts of polling data that was available and how it was all weighted. Gelman et al published their model so it's worth digging into.
Also, Trump gained ground among demographics (compared to 2016) that would have been most affected by voter suppression, and lost ground among white men.
Oh, really?
CNBC: Postal Service data shows poor mail-in ballot delivery rate in key swing states, judge suggests Postmaster General DeJoy might have to testify.
From: https://www.cnbc.com/2020/11/04/2020-presidential-election-p...
--- KEY POINTS
The Postal Service on Election Day failed to deliver a significant percentage of mail-in ballots to several states that could determine who wins the presidential contest between President Donald Trump and Joe Biden, data filed with a federal court suggested Wednesday.
The states seeing relatively poor levels of mail deliveries of ballots include Georgia, Pennsylvania, Michigan, North Carolina and Wisconsin, all of which have yet to declare victors in their tallies for Trump and Biden.
The states affected include Georgia, Pennsylvania, Michigan, North Carolina and Wisconsin.
Federal Judge Emmet Sullivan was furious at Postal officials for failing to inform him Tuesday of their failure to do sweeps at mail facilities for ballots by a mid-afternoon deadline.
Lying, even when you're completely anonymous, is a serious problem in social sciences, and one for which sociologists have had to correct for.
While most people will proudly claim they want to vote one way or the other, a good percentage of the population, doesn't want to say 'I Vote Trump', because they are afraid that their voting choice will not be socially acceptable in their circles.
The tl;dr is that there's more evidence against it than for it.
Polling errors are typically caused by unrepresentative polling - southern latino voters (one of the key shifts for Trump in Florida) in particular are hard to reach for polls, because there can be a language barrier.
There is a huge distrust of media in the US, and pollsters are seen as part of that. I do think there is some component to bias influencing the election and how polls can be either manipulated or withheld to manipulate that.
Personally, I think we'd be better off politically without pollsters and less polarized without the media complex.
> when the vast majority of polls in Florida showed Biden in the lead, and then Biden lost there by a few percentage points, that’s a polling error.
The job of the polls is to gather information, not forecast. It sounds like forecasters are setting up to pass blame off onto the polls instead of taking responsibility. And I guess they got away with this in 2016.
The polls asked people questions, the people answered, the polls reported the answers. So one would have to dig a lot deeper into the poll's claims to say they messed up.
There’s also the factor of questioning, both how things are worded and the means they use to ask them. Landline polling with 50 questions (yes, some pollers do this and you can see the questions they ask) aren’t likely to reach people with jobs or people who are young, so that results in the few people they do reach giving undue weight to “make up” for the voters they couldn’t poll. Wording of a question can also change how a person reports their result, and there’s always the lizard man constant and people who intentionally screw with pollers, some being easy to detect and discard.
In a way, they do end up as forecasters.
Polls tend to focus on "likely voters", which is defined mostly by who's voted before. So something like the massive surge we saw in Latino Trump voters in Florida and Texas would be difficult to see coming unless you have some reliable way of forecasting which seemingly-unlikely-voters will actually turn out this time around, which is extremely difficult -- people tend to be much more likely to say they'll vote than to actually vote, so simply asking them isn't reliable.
"Shy Trump voters" probably isn't the core problem in an unusually high-turnout election. The problem is that there's no real way of predicting who pollsters should be interviewing.
That isn't a problem. That's why random sampling and confidence intervals exist.
"Polls tend to focus on 'likely voters'" - no they don't. They randomly sample, and then ask people whether they plan to vote or not; they still get, and report, the numbers amongst registered voters, vs likely (as defined however) voters.
His entire job is poll analysis and he fails to see how they're fundamentally flawed when some of them were off by more than double the margin of error.
Why is he still relevant?
In any democracy the elections would be over by now, Biden won the popular vote, that’s it. But in the US...
And “any democracy”? Is Canada not a democracy either? We don’t elect our leader based on popular vote either.
If you don't like the electoral college system, go look at what it was negotiated for and also stop wanting whatever that was too. Otherwise you're just like someone saying "That other country should be part of my country! Not fair!"
In New Zealand's last election, the National Party won the popular vote but their opponent the Labour Party actually won the election and is now in power. So New Zealand isn't a democracy either?
Incorrect. In the last election, a couple of weeks ago, Labour easily won the popular vote and is also in power.
There is no "popular vote" for the President, legally speaking. The popular vote is essentially a colloquial totaling of 51 separate vote-totaling processes.
That's why the system was designed with an electoral college. So that rural farmers could be represented as equally and strongly as city business people. Should they be? Probably, if we want to eat.
Even in the current system, the election is largely chosen by people living in large cities that are completely disconnected from the rest of the 95% of the country. There is almost a city limits split between red and blue. Maintaining balance in the complex system is more important than making each voice heard.
This isn't all that accurate. One of the main reasons why the framers came up with the electoral college was due to the fact that, 200+ years ago, information exchange was extremely low. The US was already fairly large at that point, and the concern was that 18th-century voters in rural areas would lack the resources to have an informed vote about national candidates.
There was concern about equal representation, but reducing that down to just "farmers" is a little disingenuous. The reason why Southern states agreed to a voting system that granted electors based off of population was because they were able to work in the Three-Fifths compromise. States like Virginia, where enslaved people counted 60% of the population, greatly benefited from this, and at one point enabled Virginia to count for 25% of the national electorate.
These decisions might have been reasonable 200+ years ago, but neither of these are the case anymore, and are not good reasons to cling on to a system that was largely born out of necessity.
Also, one point that's usually glossed over is that when people gripe about the electoral college, it's usually referring to the fact that the candidate with the majority of votes in a state takes all of their electors (e.g. winner takes all). However, the framers never defined how states should conduct their own elections. This system of allocating votes only began catching on around 40 years after they ratified the Constitution, where individual states gradually decided to implement this system. There's nothing in the Constitution about it, and even James Madison famously wrote a letter in 1827 complaining about how adopting this system could undermine our entire Democratic process.
I agree there are possibly better ways, possibly force choice ranking, etc. But all a simple majority vote would do would be to ensure we would become a one-party system, the opposite direction we should be headed (a multiple-party system).
Having it be controlled by the states is a great testing bed for that, for example, in Maine. There's no reason to take that power away from states and make it federal.
IANAL, but my interpretation is that the part that is in the constitution is that it is up to the individual states to decide how to run their elections and assign electoral votes.
Let me give a practical example. Aboloshing guns. Note I'm not even talking gun control, but actually getting rid of classes of guns.
I've lived both in Texas ranch land, and in New York (Jersey, commuting to Manhattan). Those are very different contexts. In New York, guns are seen as nothing but a weapon. In Texas, guns are seen as a necessity in places. In New York, guns are to defend yourself or to hurt someone else, and are almost exclusively used against people. In Texas, guns are used for those purposes, but also to hunt, to kill boars that would hurt people otherwise (and are invasive, dangerous, and could kill animals and children), to put down animals, to kill coyotes who steal chickens, to kill a rattlesnake that is in the horse pen...they're tools. You know what AR-15s are useful for? Boars and coyotes, as well as cougars (not in Texas that I'm aware of though). They're great against small fast predators where you would need more than one shot in an easy to shoot gun. Wouldn't carry one for a minute in Alaska against Grizzly bears. That's an entirely different context than Texas, even though it's "rural". I don't think there's much need for one at ALL in Manhattan where if you fired one it'd shoot through three apartments' walls defending yourself. Second amendment and the right to defend yourself aside.
Even talking about defending yourself, how long do you think it'll take to get a county sheriff out to a 1,000 acre ranch in the middle of nowhere if someone is shooting at you?
The context matters. There's a lot fewer rancher Texans than there are people in Manhattan. Why should Manhattan get the full say in saying that the rancher's way of life is invalid? Maybe there is compromise, but I honestly don't think there is such a thing in our current climate.
This crosses many areas...which business get incentives? Should we give subsidies for corn, for food, etc. Should we encourage tech growth? What about migrant workers? Visa workers are very different than someone coming in to work a ranch or construction.
On some issues I agree with you. Particularly moral issues. We need to be "American" at some point, and not bipartisan. We need to have discussions on issues and not fight across party lines.
But some issues, the party lines are there because the parties represent different groups of people. We should discuss those issues too, but in almost all cases I've seen both sides don't even begin to understand the other side. They can't comprehend it because they've not walked a mile in the other people's shoes. "Who cares about coal miners?" Or the thoughts of Ivory Tower academics.
There's probably a more fair system than what we have. We need to destroy bipartisanship and limit the federal government so that local governments can give their people what they need, in my opinion, while keeping national issues like the environment as a topical national discussion. We're a LONG way from any of that though, and will continue to as long as there's this left/right divide, unwillingness to cross the aisle, and the view that the other side is the enemy, stupid, or evil/racist. BOTH sides see each other that way, my Facebook is full of people from both Texas and New York who see each other that way, and just don't understand...both sides mean well.
Economically, structurally, contextually, we need a government that gives some level of power to people who live in more rural areas. Not absolute, but a balance. Pretty much everyone in Manhattan I knew were either party line democrats or libertarians. Nothing wrong with that, but that is a big bucket of people that pretty much all believe the same things because of their context, and don't understand other areas as well. New York itself is split into different contexts, upstate New York seems an entirely different state/country than New York City.
Maybe some people do understand the issues and each other...again, we're people, it's complex, we have different backgrounds. But as it stands, our system is polarized into two groups that don't understand each other, but help each other in a weird symbiotic way. If you switch it to be purely population based...it's going to be one group that now doesn't understand 95% of the country's land area, and very little of the supporting infrastructure that enables things like tech or finance.
Note what I'm NOT saying...that the electoral college fixes all of this. I'm just saying that a purely popular democratic vote would not work like people think they would, because all they see are their issues, and the most populous groups mostly live in the same types of contexts (college educated cities with tech, finance, medical, education, and business sectors).