The number of people in the comments, who are provided with this information, either from this post or others on this, who think that 55/45 is substantially different 45/55 staggers me.
If you're paying for a subscription and aren't bothered to read and understand what Nate and Eli write, what's the point?
Another point. One can't be "wrong" if one presents probabilities. The heat Nate took over Trump's 2016 win astounds me. An underdog came in. It happens all the time.
Mmmm, yes and no. I think Sam Wang was definitely wrong about the 2016 election when he gave Clinton a 99% chance of winning. And honestly, I kind of blame him (and some of the other modelers with overconfident predictions) for the backlash Nate faced. The public were told "The Models" said Clinton was cruising to victory and 538 took the heat because they were the biggest dog in town
Though props to him, Sam said he'd literally eat a bug if Trump won more than 240 electoral college votes and he followed through.
Probabilities can be unreasonable, like the ones you mention. But they can’t be wrong, since there is no actual thing in the world the probability is trying to match.
With a binary prediction, you can be wrong or unreasonable, but they are different. Someone who predicted Trump to win in November 2016 was probably a bit unreasonable (unless they had information about polling errors that wasn’t part of the various models), but they ended up being right.
Probabilities can be wrong, because they can be calculated incorrectly or based on bad assumptions.
Some theoretical probability calculated with perfect information, I guess, couldn't be wrong, per se, but any set of probabilities crafted by humans relying on imperfect information and subjective assumptions can indeed be wrong. Or at least bad, misleading, or disingenuous.
Probabilities can be bad, misleading, or dangerous. But they can't be wrong.
Given perfect information, a correct calculation will always be 0 or 1 (at least, in a deterministic universe - which many theories say that a quantum universe is, despite its popular interpretations). So any probability other than 0 or 1 is in that idealized sense, "wrong", but I think that's not a useful sense.
I like this point. It seems to me that probability only works in a limited field: a deck of 52 cards, or a chess game. But an election is the whole world, and that isn't limited in any way that makes sense of probability, IMO. An asteroid could hit the planet. There could be 10-foot-tall space invaders, or a zombie war. Marburg virus could wipe out much of the human race. Seems to me the probabilities being considered leave out a lot of possibilities; in fact, most of them.
Long tail outcomes like that won't shift the final estimate much and would be unreasonably computationally expensive to model. Of course the model is a simplification, it has to be - you can't fully simulate the mental state of every potential voter from now until election day.
The whole point is building a tool that will be close enough to be useful.
“ It seems to me that probability only works in a limited field:”
That’s because you’re a GD ignorant fool, stuck in a spotty elementary school level of understanding of probabilities.
The last century plus of scientific advancement is based upon [successful] use of these techniques applied to the messiest “real world” problems. Quantum mechanics, an EXTREMELY successful model without which we’d have basically none of the last 50 years of advancement, is entirely composed of calculation of probability of things we cannot possibly directly observe.
Somebody told Trump to go campaign in WI/MI/PA. At the time nobody seriously thought the blue wall would crack. Whoever that person was they had extraordinary insight.
It's just like an on side kick at the end of the game. It's not like it's a great play, but it's the only potential option to win so obviously you do it. Clinton team forgot the converse. You put in the hands team.
Absolutely probabilities can be wrong, though -- you are trying to match how often some event happens given the available evidence. In politics, you can't run the same election twice, so yes the set of "all presidential elections where the major party candidates are Harris and Trump and it's 2024" will only ever have one item in it. But that doesn't mean the probability can't be wrong -- and in repeatable cases (e.g., coin flips) a model's probabilities can absolutely not match the world's actual outcomes: for a given approximately-fair coin, saying the probability of heads is 10% is quite likely wrong.
> you are trying to match how often some event happens given the available evidence.
If you are trying to match frequencies, then yes, frequencies can be right or wrong. But the kinds of probabilities we are talking about here are not frequencies. It is reasonable for them to be informed by frequencies. But they are importantly different.
They can't be wrong but they can be unreliable. If Trump ends up winning in an EC landslide and ties the national popular vote, then Nate's model can then be deemed as unreliable and uncredible.
"Unreliable" and "uncredible" is always a comparative concept. And it is also inherently one that applies to a strategy of making predictions, not just a single prediction.
In order to show that Nate's model is unreliable, you would need a specific alternative, and you would need a sequence of events, where following the alternative has regularly done better than following his model.
But sometimes, two equally reliable models give different predictions in the same case. And sometimes, a reliable and credible model is far from the truth in a particular case.
You most certainly can be wrong, you can be an absolute bunko artist, but your audience will never know because your claims and findings are unfalsifiable to them, and therefore literally non-scientific.
Not true in Nate's case. 538 published data showing their models were well-calibrated (e.g. 70% probability events happen approximately 70% of the time, etc.)
Given a large enough sample you can demonstrate that the distribution of probabilities matches what the model predicted. The problem with elections is that the sample size is nowhere near that threshold.
It's probably exactly that simple. The sample sizes are just way, way too small to do what we want ----------- predict for sure who will win.
Not "take a snapshot," not "of course this poll will vary from the Election Day poll" ----- all the weasel words that are coming at us saying that they don't know who will win and can't fix that so far. We just want one thing, to know the future, and the pollsters cannot give that to us.
They say math is an art rather than a science, so maybe that’s the point of confusion? Because this is math. Probabilities are exactly what the word says, they are not predictions or certainties because we cannot predict the future (aside from stuff like the sunrise or the tides).
Which brings up another difference between elections and sports—elections, so long as they are fairly administered, are determined by the people. We only know ahead of time who will win when it is very clear that there are far more people who are going to vote for this candidate rather than that one (and also, in presidential races due to our crazy system, that those percentages hold for enough states to total 270 ev’s or more). When things are close like this, it all depends on who turns out to vote.
You certainly can be wrong, it’s just usually harder to demonstrate. Ironically the key difficulty that makes Presidental election models hard to build, low n, also makes it harder to verify their accuracy.
Just tally up Nate's forecasted EVs for whichever party you support, then measure against the final result.
The closer Nate is, the better.
In 08, 12, 20 he was very close. In 16 he was fairly off but less off than most. That record is good enough that you and I both subscribe to his substack.
I subscribed because I thought Nate had become non-partisan. But I was wrong. Canceling Nov 6 after Trump proves that Nate's model is junk. Garbage in, garbage out is the best way to describe his model.
He's mentioned multiple times he hopes Kamala wins. It's not exactly a state secret that Nate is left. And that's ok. Everyone is a partisan. But I prefer analysts who are non-partisan in their analyses.
Naaaaaaaah ---- tell me weather forecasts are accurate when they make a prediction on Wednesday for the next ten days ----- and then actually stand by it!! Rise or fall professionally by their prediction! Wow, wouldn't that be something unusual. Instead, they change the forecast every second five minutes and in the end they even fail the "look out the window" weather test.
Polling has looked at this model and realized there is a lot of profit in pretending they are doing something else than what we want them to do (predict the winner accurately), and so they talk about snapshots, about how it's really okay that the polls keep changing, see, because they can't be expected to reflect anything about the final Poll Nov. 5. (!!??!!) Chaos theory rules in both weather forecasting and polling and since they can't do anything about that butterfly in Hong Kong, they cash in on our interest without fulfilling our need.
I think I am switching over to fundamentals as a model for election prediction.
It happens some of the time, not “all of the time,” although that is an expression usually used to signify that “less likely” things happen often enough.
There have been ~80/4=20 presidential elections since the start of modern polling. Consequently it's not too unreasonable to expect that choices in polling model formulation should give probabilities varying by about 100/20=5%. Now, maybe it's just numerology, but Twitter was recently going nuts about how Polymarket is about 2∙5%=10% different from Nate's model.
With that said, until I sat down and wrote this comment, I expected my final numbers to be 2∙3%=6%. I'm not sure where that came from, but if it were true, then a 10% variation would reflect a surprising disagreement about the nature of US presidential elections.
A big enough difference that you'd change your thought/behavior on it. Both are in the territory of a tight race and would lead people to think about it as a tight race. Neither are so far off 50/50 that you would think "This person is the favorite to win".
In terms of this specific race, there's been commentary about how campaigns take different actions depending on if they think they're ahead/tied/behind. 55% to win is still close enough that you would campaign as if you were tied, as opposed to starting to take actions as if you were ahead/behind. Not having a third debate is a good example of that. If Trump thought he was behind, a third debate might be a smart Hail Mary play, but he wouldn't want one if he thinks it's tied.
Changing your thought isn't a binary state, though. Any small change in the inputs should lead a Bayesian to make concomitantly small adjustments to his priors.
Trump is kind of unique though—that would hold true for an ordinary candidate. But given that he kind of had his ass handed to him not because he made normal debate mistakes but because he sounded insane, he probably knows deep down that a debate against Kamala Harris is not in his interest. Of course, he cannot admit in public that he lost, which is another reason not to do it again. So his refusal to debate is really neither here nor there—it’s just not his venue.
Another W article from Nate. You are going to get hated on no matter what way the election goes, but real ones understand how valuable your predictions are.
There is simply no predicting, the word just doesn't belong. If you walk up to a roulette table, each spot gives you different odds... Nate is just trying to tell you the odds in a game where the table and odds are continually changing. Placing a bet on a roulette table is not a prediction; the lottery picks winners regularly but it doesn't change the odds for you being a winner.
Serious question -- how valuable do you believe a coin flip prediction actually is (outside of deciding whether or not to wade into the prediction market.)
I think its extremely hard to look at a 55/45 split and not have your own personal preference, beliefs, or desired outcome cloud your judgment into thinking 55 is more of a lead than it is. The higher the stakes, the worse that problem is I think. I’m not anything close to a statistician, but I could be reasonably called a subject matter expert on politics… and I’m constantly having to explain this to problem to colleagues, friends and family. I appreciate how hard Nate works to provide us with examples and analogies to help those conversations go a little easier.
Yeah, psychologically speaking I think most people can understand basic probability, but they also don't know how to think objectively enough for it to matter.
One's tendency to think of this as a life changing event, and then propensity for optimism or doom spiralling, for example can have a major effect on their perception of the likelihood.
I always think about it this way... You have a bag with 100 balls in it. 55 are blue, 45 are red. How sure do you feel that you can reach in and draw a blue (or a red) ball? How much of your next paycheck would you bet on either one? As a non-gambling, regular person, my answers are "not very" and "none".
"how often do I get to bet on this?" is also important.
If you will let me keep pulling bags out of the ball, and bet on the outcome each time, there may be no end to how much money I will sink into it (though there would be a cap on how much I would bet on one individual pull).
Indeed! If you can draw as many times as you like, and drawing the blue ball pays double your bet and the red ball zero, you should bet 10% of your bankroll every time- thank you, Kelly criterion.
Yeah, you're correct and I think in dice gaming turns as well. I've played to win tournaments with prize money on the line and it really brings home how when 1/20 odds happen quite often!
These things are pretty intense when it feels it will have a significant impact on the world my children grow up in.
The idea is correct, but as of the last time I checked, your numbers are wrong. the correct version is: Nate's model says that Kamala needs to get a 9 or higher. Polymarket says she needs to get an 11 or higher.
On a d20, a perfect 50/50 would be 10 winning faces, and 10 losing faces. In that scenario, the fails would be 1, 2, 3, 4, 5, 6, 7, 8, 9, 10. So "11 or higher" is 50%. From there, we add or remove a number to get the 45/55 and 55/45 odds.
For the D&D nerds: yes, that does actually mean death saving throws (pass at 10 or higher) are slightly tilted in the player's favor.
I don't disagree with any of this (it'sactually quite good), but I view things slightly differently, and it ties to something I've been thinking about lately.
Even though polling is closer than it's ever been and most models have the election at a toss-up, it is actually quite unlikely that the 2024 election will be closer than the 2020 election, and there's a decent chance the election won't be close at all. In 2020, Biden won WI (which gave him his 270th electoral college vote) by about 0.6 pt. Meanwhile, Nate's model has P(decisive state being decided by 0.5 pts or less) at only about 10%.
So if the result potentially ends up being not as close as everybody thinks it will be, people will inevitably bitch and complain, demonstrating how the pollsters and modelers were wrong, when in reality, people (still) don't understand how variance and volatility work and how it relates to probability.
I don't know if you're a subscriber, but Nate had an earlier post (https://www.natesilver.net/p/the-128-paths-to-the-white-house) that points this out explicitly: of all the possible ways the 7 swing states can fall, by *far* the two most likely according to the model are either all 7 going to Harris or all 7 going to Trump. As the model sees it, any deviation from the polling average is likely to be highly correlated at the state level. It's not fifty-fifty because we're very sure there's a bunch of races that will individually be very close, it's fifty-fifty because the range uncertainty is wide enough to cover moderately strong wins for either candidate, in a roughly equal manner.
In baseball, mean reversion is more of a thing than in other sports. Because there are nine batters and pitchers can’t play every game, no one player can dominate the way a goalie can dominate in hockey or an nba star can carry a team. Even an nfl quarterback has much more leverage than a strong center fielder.
Yes, it's true that baseball's structure leads to a more pronounced mean reversion due to less dominance by single players. However, the underlying concept of reversion to the mean applies broadly (for example, Nick Foles’ 2013 season vs his career). My main point was that in any field involving statistical analysis, from sports to political polling, it's important to consider (and remember) how results generally trend back toward the average over time.
To patch up the sports analogy, the final score is the EC count accepted by Congress. We’re watching a game with imperfect know of the actual points/goals/runs that have already happened. That’s only revealed at the end of the game.
Edit: Basically this is panel judged boxing match, where scoring is announced at the end of the match. A lot of the time it’s really obvious who is ahead, but occasionally it’s vague enough that there’s reasonable doubt until the end
Pollsters have been making an active effort to avoid underestimating Trump after doing so for two cycles in a row. That does make me think the chance that they underestimate him is lower than normal, but I don't have any evidence for that.
Exactly. I think the same will happen here. The 3% that are undecided and the 5 or so percent that could change their mind will swing to Harris. They would of swung to Trump alalready if they were going to choose him
Perhaps you’re right, but I’d be just as hesitant to be confident of that as to be confident that Trump is underestimated. Harris is still considered more acceptable in polite society, and Trump’s voters (based on from what I would characterize as a lack of civic engagement beyond voting, not “shyness”) still may be undersampled.
The best case for Harris being currently underestimated is, in my opinion, that her marginal voters are probably higher-propensity than Trump’s marginal voters.
The case for Harris being underestimated is the idea that pollsters got a ton of egg on their face because they missed in the same direction twice. They are strongly incentivized not to underestimate Trump again, while underestimating Harris wouldn't look as bad for them.
Pollsters are well aware of "shy Trump" voters, and after two cycles should be able to correct for it.
It would be interesting if some pollster would post their poll, and then also post what their poll would have shown if they still used the same methodology they used in 2016. I would expect that the 2016 methodology would show Harris with a much larger lead than the 2024 methodology.
The "shortfall" was compared to the expected benefit from the House gerrymander. in 2020, the D's were +3.1% in the popular vote, but only +2% in seats.
If the relative difference had held, the R's in 2022 should have have a 16 seat majority, but they only got a 9 seat edge.
Of course extrapolating 2022 to 2024 is risky because of off-presidential election effects. The total congressional vote in 2022 was 30% less than in 2020.
"Late polling was R+1.2 at 538, and R+2.5 at RCP."
I mean it seems peculiar to me that a lot of Democratic partisans seem to be hanging their hopes on an election where the R vote was still underestimated (although by not as much as the Presidential elections) and the R's won at the national level.
I don't know what late polling means, but the final 538 prediction was Rs winning 230 seats and Rs winning the national house vote by 4%, so Ds outperformed those two benchmarks.
The question is _how_ they are tweaking their numbers. If, for example, they give extra weight to blue collar whites without a degree do they risk missing blue collar minorities without a degree if that latter group makes up an increasingly large share of "shy" Trump voters?
Yes, of course the specifics of the adjustments are important. The article I posted states that many pollsters are weighting based on recalled 2020 vote, which is an odd metric to use. Many voters in 2020 will not vote in this election, and many non-voters in 2020 will vote in this election. Plus some people will have had their minds changed in one way or the other.
The polls may be underestimating Harris vote share to avoid a repeat of the systemic Trump error in the past, however, DJT seems to be hovering around his ceiling in the polls and the ballot box, which has historically been between 46% and 47%. DJT is arguably the best known political product in the modern era, and yet has never achieved a majority of the vote in either elections, and seems to have hit this ceiling in the polling averages. If indeed Harris polling is being underestimated, as I agree, it is likely she will, as did the last two Democratic Presidential contenders, achieve a popular vote victory. The real question, is can she get over the 51% threshold needed for the EC edge needed? And is it 51% or closer to 52?
I don’t think it’s actually a popular vote threshold she needs to get over, she just needs those votes in the right places. A few tens of thousands of votes in battleground states are more important than millions of votes nationwide.
Agreed. My question/argument is that there is a National popular vote threshold that will carry her over the finish line in those battleground states. For instance, Biden received just over 51% and carried MI, GA and PA, Clinton on the other hand barely topped 48% and lost MI, GA and PA (and others).
2/3 of polling organizations are using weights to compensate for past polling errors in favor of Dems.
Nate has said that he believes there is not a long term systemic bias in the polls, ie random walk; the systemic Trump error of the past doesn't mean it will be there this election too, like you said, the polls could be overcompensating for it. Time will tell.
Yes, I love this number. I use 2.1 national popular vote lead which lost the election for Hillary, and 4.1 national popular vote lead that won the election for Biden. So it's a useful shortcut for me to look at the national popular vote polls. Hey, at least the n is bigger.
You haven't heard much about third-party votes since RFK Jr dropped out, but I'm guessing they may wind up being critical. Total third-party vote share was 5.8% in 2016 and 1.8% in 2020. My feeling, admittedly based on "vibes" more than anything else, is that it will bounce up a bit this year, maybe to the ballpark of 3%. Harris winning the national popular vote 50-47 would then be plausible. The question is how many third-party votes would come at Trump's expense in the key states.
Looking at MI, where Kennedy is still on the ballot as the Natural Law party candidate, it's interesting to look at recent third party impact in the state. Gary Johnson in 2016 running as a libertarian received 3.6% of the vote. Hillary barely lost MI in '16 to Trump 47.5% to 47.3%. In my opinion, Gary Johnson provided a "safe harbor" for "Never Trump" and "Not another Clinton" conservative voters. Arguably, there is a fine line between an old school conservative Republican and a Libertarian. Some Libertarians, e.g. Justin Amish, like to hang out in Republican ranks understanding the two party system is what it is. (I limit my rambling on this). In 2020, in MI, there was not a viable Libertarian alternative for Never Trump voters, and we saw 3rd party vote share drop by 2.5% (112,000) votes, while Trump only gained .3 % more than '16, allowing Biden to turn the state blue garnering 50.62% of all votes cast. Now MI experienced record turn out, and it is probably intellectually dishonest to suggest that all of the Libertarian vote decline was Never Trump R vote from '16 who found Biden to be less enough of an evil than Hillary and Trump enough of a menace that they voted D, but Biden won MI by about 150K votes, and the 3rd party decline was about 120K ish. That's all a long-winded explanation to suggest that if Kennedy presents the same "safe harbor" for the "Never Trump" voter as Johnson in '16, the third party voted could impact Harris more than is currently anticipated. I don't adhere that school of thought, as Kennedy seems to have lost all credibility since endorsing Trump, and Trump has never received a greater share of the vote in MI that that 47 ish %, right where he seems to hang out in all polling. Like I said, fun thought experiment.
Yes, 3.6% is a LOT in an election as close as this. Michigan is delicate because of the Muslim thing at least, and maybe the union thing too. As you say, analysts blamed third party voters for Hillary's loss in 2016 and a narrow loss for that reason could happen again. Certainly Kennedy is trying to help Trump in this way. I like your point.
But I can't see why the RFK Jr. vote would matter. Because I was looking at a NYT graph today that showed the bump-up in votes after RFK left the race and BOTH candidates profited exactly equally: the graph bumped up for both at the same place, same amount. Why Trump offered RFK a job, I don't know --- he must know something I don't understand. Okay, it's a trade for RFK going around taking his name off the ballots in battleground states, so they must think if people can't vote for him, they'll vote Trump instead. It's not how the graph looked ---
Agree. I think Trump is still leaking some blood from the debate fiasco. And the recent good economic news (lower interest rates, lower inflation, good jobs report) will give Harris a small boost.
Agh, yes, this is so annoying. A few weeks ago, when you had Trump as 62% likely to win briefly, the Daily Mail (which is admittedly only slightly more reliable a news source than Fox News) had a headline saying you were predicting a "blow-out win" for Trump. I suspect they do know this is not what probabilities mean, but pretending they don't gives a far better headline
Also, re the election, when Harris smoked and depantsed Trump in the debate and THAT did not move the needle… unfortunately feels like people just want Trump back
The debate did not move the needle because while Harris voters were ecstatic at her performance and Trump voters disappointed that he was totally by her bait , I did not hear anyone say that she provided any substance or clarification of her views. In fact, most undecided voters felt that she disappointed them. Style points alone never get you a real win in any sport of which I am aware, even if they make you look good. Her interviews, even with sympathetic interviewers, continue to be content free and/or word salads.
I don't think you know what word salad means. You can say you don't like what she's put out there so far. You can say her plans aren't substantive enough for you. But only one candidate is spewing incoherent nonsense. Just because you're accustomed to unhinged language from him doesn't mean it isnt much closer to word salad.
It's not word salad, technically --- I've seen that, idiopathic schizophrenic talk. But she does have pretty bad perseveration, a brain damage symptom: all that repetition of the same word or phrase that she catches onto and then repeats throughout the sentence and the paragraph, making no sense. She is 60 --- it could be mini-strokes, I suppose.
How many people do you know who do that perseveration on one word? Right, none.
Lol most women with jobs ranking higher than manager whom I know talk this way. Me included. It's how we're trained not to appear too threatening to men.
That isn't a word salad. If it is "unintelligible" to you, it indicates your comprehension problems.
As Crocodile Dundee would say, "You think that is a word salad? Now this is a word salad".
Q : "If elected, what specific legislation would you seek to pass to lower childcare costs in America? "
Trump : "Well, I would do that. And we’re sitting down – you know, I was – somebody we had, Sen. Marco Rubio, and my daughter Ivanka was so impactful on that issue. It’s a very important issue. But I think when you talk about the kind of numbers that I’m talking about, that – because, look, childcare is childcare, it couldn’t – you know, there’s something – you have to have it. In this country you have to have it."
Nate is absolutely right that there's no meaningful difference between a 40-60 probability vs a 60-40 probability for a one time event (unless you're a gambler. Unfortunately none will need Nate's advice.
Nate's point on the precision of polls is also important. If you wanted to know what margin would be rounded to the nearest 10th (Trump +10, tied, Kamala +10, etc) polls would rarely be wrong! Polls are even useful if you want to round to the nearest 5th (Trump +5, tied, Kamala +5, etc). But happens when the race is Kamala +2.5? If you trust RCP's average of Kamala +2, you likely think Trump is the favorite. If you prefer Nate's Kamala +3, you likely think she is winning. But those are arbitrary and self motivated reasons for preferring an average that are subject to high degrees of error.
The race is 50-50, it's always been 50-50, it always will be 50-50.
Absolutely love this, and why I’m a paid subscriber (Trump voter, btw…holding my nose).
@Nate - Question: If you were to frame the race in sports betting odds (-110, +215, etc), what would it look like? For example, something like, Harris -115, Trump +105…?
One thing that Nate is leaving out is MOMENTUM. In sports, this plays a big part in who winds up winning a particular game. I have no doubt that TRUMP has that momentum at the present time. When a candidate can draw a crowd of 80,000 people to his rallies compared to 5,000 for Harris (many of which are bused in by the candidate) I have to believe that Trump's base is fired up. The Sportsbooks now have Trump as the slight favorite in the betting odds. They are the best indicators of who is going to win or lose, IMO. When I see Harris answering questions with a bunch of word salads & that ridiculous habit of shaking her head yes, at the end of every sentence, I see a person who is not being authentic, & quite frankly not fit to be the leader of the free world. But, of course, we have to factor in the elephant in the room. The legacy media & multinational Corporations all want Harris to win. And make no mistake, they are the ones that are both manipulating & counting the votes. That may be an obstacle that Trump will not be able to overcome. In a free & fair election my money would be on Trump, however in today's corrupt political process, I'm probably going to bet on Harris. Sad but True.
Have you heard a 5 minute slice of anything from Trump? If it weren't for the sanewashing in the media he would have been dropped from the election before Biden.
Harris on the other hand is only unintelligible if you have problems with complex concepts and rhetorical devices.
The Harris people are trying to say her crowds are also big ---- doesn't look like it, though, and this enthusiasm gap is VERY reminiscent of the problem Hillary had. Both Kamala and Hillary have small inside groups of bused-in audiences ------ huh. Probably not a good sign.
As a former Democrat, let me also state that everything that the Democrat Party accuses Trump of either doing or going to do, is something that they are either doing now (at this very moment), or plan to do in the future, if they remain in power. The only End To Democracy that will happen is when the federal government controls what we are allowed to say, either in a private or in a public venue. George Orwell predicted what was going to happen. The only thing he got wrong was the year it would take place. Be warned!
Fox is not a legacy media company. CBS, NBC, ABC CNN, MSNBC, all push the legacy corporation party line. Legacy media refers to traditional forms of mass communication that have been established for a significant period, such as newspapers, television, and radio. These mediums were dominant before the rise of the internet and continue to shape public discourse and information dissemination today. Get a clue before you respond.
I just saw a video of a nearly empty Trump rally. Don't blame the media, it was a video recording, and Trump was talking to a mostly empty room. They panned across the room, and most attendees were behind Trump. Nobody really has momentum at the moment.
PS - There is no factor in the model for momentum. Nate addressed this a while back when Harris had the momentum. If somebody has momentum, it needs to show in polling.
If someone has momentum, it WILL show in polling, unless the polls are corrupt. Which they may be: keeping it close to keep us all clicking. I remember George H.W. Bush used to call it "the Big Mo."
There can be a Big Fade, too, as when Comey interfered with the Hillary election by tying her to that Anthony Weiner character, because he had one of her old laptops and was sexting on it.
Or when little Dukakis wore the oversized helmet in the oversized tank and looked like an idiot.
Did I say that? I don't think so. Butler v2 was obviously a special case.
Rally size really doesn't mean much. We all have no doubt that Trump has a deeply devoted base of followers who will show up, even if it kills them (see 2020 superspreader rallies, for example). That doesn't win elections.
Big crowds, however, HAVE to be a better sign than bussing in a hundred people because nobody will come otherwise. I think Trump has always been correct to focus on that and publicize it. It's impressive. Dems often sneer at Trump's "deeply devoted base of followers" -------------- but I bet Harris' campaign wishes they had that.
Harris campaign definitely has that. I live in a purple part of a blue state and am experiencing it directly. It's not the same as Trump's base, but it is there. By the way, it includes a number of Republicans who voted for Trump in past elections.
PS - Note that Ralph Nader had multiple "super rallies" with well over 10,000 attendees - and he got 3%. Rallies have value for getting media attention and volunteers, but mean little otherwise.
The number of people in the comments, who are provided with this information, either from this post or others on this, who think that 55/45 is substantially different 45/55 staggers me.
If you're paying for a subscription and aren't bothered to read and understand what Nate and Eli write, what's the point?
Another point. One can't be "wrong" if one presents probabilities. The heat Nate took over Trump's 2016 win astounds me. An underdog came in. It happens all the time.
Mmmm, yes and no. I think Sam Wang was definitely wrong about the 2016 election when he gave Clinton a 99% chance of winning. And honestly, I kind of blame him (and some of the other modelers with overconfident predictions) for the backlash Nate faced. The public were told "The Models" said Clinton was cruising to victory and 538 took the heat because they were the biggest dog in town
Though props to him, Sam said he'd literally eat a bug if Trump won more than 240 electoral college votes and he followed through.
Probabilities can be unreasonable, like the ones you mention. But they can’t be wrong, since there is no actual thing in the world the probability is trying to match.
With a binary prediction, you can be wrong or unreasonable, but they are different. Someone who predicted Trump to win in November 2016 was probably a bit unreasonable (unless they had information about polling errors that wasn’t part of the various models), but they ended up being right.
Probabilities can be wrong, because they can be calculated incorrectly or based on bad assumptions.
Some theoretical probability calculated with perfect information, I guess, couldn't be wrong, per se, but any set of probabilities crafted by humans relying on imperfect information and subjective assumptions can indeed be wrong. Or at least bad, misleading, or disingenuous.
Probabilities can be bad, misleading, or dangerous. But they can't be wrong.
Given perfect information, a correct calculation will always be 0 or 1 (at least, in a deterministic universe - which many theories say that a quantum universe is, despite its popular interpretations). So any probability other than 0 or 1 is in that idealized sense, "wrong", but I think that's not a useful sense.
I like this point. It seems to me that probability only works in a limited field: a deck of 52 cards, or a chess game. But an election is the whole world, and that isn't limited in any way that makes sense of probability, IMO. An asteroid could hit the planet. There could be 10-foot-tall space invaders, or a zombie war. Marburg virus could wipe out much of the human race. Seems to me the probabilities being considered leave out a lot of possibilities; in fact, most of them.
Long tail outcomes like that won't shift the final estimate much and would be unreasonably computationally expensive to model. Of course the model is a simplification, it has to be - you can't fully simulate the mental state of every potential voter from now until election day.
The whole point is building a tool that will be close enough to be useful.
“ It seems to me that probability only works in a limited field:”
That’s because you’re a GD ignorant fool, stuck in a spotty elementary school level of understanding of probabilities.
The last century plus of scientific advancement is based upon [successful] use of these techniques applied to the messiest “real world” problems. Quantum mechanics, an EXTREMELY successful model without which we’d have basically none of the last 50 years of advancement, is entirely composed of calculation of probability of things we cannot possibly directly observe.
Somebody told Trump to go campaign in WI/MI/PA. At the time nobody seriously thought the blue wall would crack. Whoever that person was they had extraordinary insight.
It's just like an on side kick at the end of the game. It's not like it's a great play, but it's the only potential option to win so obviously you do it. Clinton team forgot the converse. You put in the hands team.
Maybe. Ot maybe somebody on Trump's team sae an opportunity and ran with it.
Cnn just ran a story about the Trump team focusing on low propensity voters instead of turnout. Same flavor.
Absolutely probabilities can be wrong, though -- you are trying to match how often some event happens given the available evidence. In politics, you can't run the same election twice, so yes the set of "all presidential elections where the major party candidates are Harris and Trump and it's 2024" will only ever have one item in it. But that doesn't mean the probability can't be wrong -- and in repeatable cases (e.g., coin flips) a model's probabilities can absolutely not match the world's actual outcomes: for a given approximately-fair coin, saying the probability of heads is 10% is quite likely wrong.
> you are trying to match how often some event happens given the available evidence.
If you are trying to match frequencies, then yes, frequencies can be right or wrong. But the kinds of probabilities we are talking about here are not frequencies. It is reasonable for them to be informed by frequencies. But they are importantly different.
They can't be wrong but they can be unreliable. If Trump ends up winning in an EC landslide and ties the national popular vote, then Nate's model can then be deemed as unreliable and uncredible.
If you have a dice model that projects an 83% chance of rolling anything but a 6, then you roll a 6, would that model be unreliable and uncredible?
The question at that point becomes how useful is the model?
To Ed Y, yes.
"Unreliable" and "uncredible" is always a comparative concept. And it is also inherently one that applies to a strategy of making predictions, not just a single prediction.
In order to show that Nate's model is unreliable, you would need a specific alternative, and you would need a sequence of events, where following the alternative has regularly done better than following his model.
But sometimes, two equally reliable models give different predictions in the same case. And sometimes, a reliable and credible model is far from the truth in a particular case.
“ then Nate's model can then be deemed as unreliable and uncredible.”
What basis do you suppose for that? Your obvious statistics illiteracy? 😉
No, for real world usefulness. Otherwise what is the utility of such a model that is off by so much?
Does anyone buy this "weighted polling" stuff? I dont think its a good methodology
Because?
You most certainly can be wrong, you can be an absolute bunko artist, but your audience will never know because your claims and findings are unfalsifiable to them, and therefore literally non-scientific.
Not true in Nate's case. 538 published data showing their models were well-calibrated (e.g. 70% probability events happen approximately 70% of the time, etc.)
No, you can’t be wrong, because probability is not a fact that one can report correctly or incorrectly.
Given a large enough sample you can demonstrate that the distribution of probabilities matches what the model predicted. The problem with elections is that the sample size is nowhere near that threshold.
It's probably exactly that simple. The sample sizes are just way, way too small to do what we want ----------- predict for sure who will win.
Not "take a snapshot," not "of course this poll will vary from the Election Day poll" ----- all the weasel words that are coming at us saying that they don't know who will win and can't fix that so far. We just want one thing, to know the future, and the pollsters cannot give that to us.
One can dig into the data and evaluate the accuracy of the model.
Was the prediction right because major assumptions were wrong in ways that balanced out?
Was the prediction wrong because a collection of close factors all flipped in the same direction?
Doing a deep dive into the crosstabs is something that Silver specifically disavows.
?
They say math is an art rather than a science, so maybe that’s the point of confusion? Because this is math. Probabilities are exactly what the word says, they are not predictions or certainties because we cannot predict the future (aside from stuff like the sunrise or the tides).
Which brings up another difference between elections and sports—elections, so long as they are fairly administered, are determined by the people. We only know ahead of time who will win when it is very clear that there are far more people who are going to vote for this candidate rather than that one (and also, in presidential races due to our crazy system, that those percentages hold for enough states to total 270 ev’s or more). When things are close like this, it all depends on who turns out to vote.
"They say math is an art rather than a science,"
THEY are wrong.
It is both.
You certainly can be wrong, it’s just usually harder to demonstrate. Ironically the key difficulty that makes Presidental election models hard to build, low n, also makes it harder to verify their accuracy.
I'd love a job where I could never be wrong.
Just tally up Nate's forecasted EVs for whichever party you support, then measure against the final result.
The closer Nate is, the better.
In 08, 12, 20 he was very close. In 16 he was fairly off but less off than most. That record is good enough that you and I both subscribe to his substack.
I subscribed because I thought Nate had become non-partisan. But I was wrong. Canceling Nov 6 after Trump proves that Nate's model is junk. Garbage in, garbage out is the best way to describe his model.
What about his posts in any way indicate partisan learnings? This is about the data.
He's mentioned multiple times he hopes Kamala wins. It's not exactly a state secret that Nate is left. And that's ok. Everyone is a partisan. But I prefer analysts who are non-partisan in their analyses.
Be a weatherman. They can change their "prediction" hourly: there is no penalty for being wrong because they can change their forecast constantly.
Spoken like a person that doesn’t actually rely on forecasts, or actually have any clue about how accurate contemporary weather forecasting is. 🙄
Naaaaaaaah ---- tell me weather forecasts are accurate when they make a prediction on Wednesday for the next ten days ----- and then actually stand by it!! Rise or fall professionally by their prediction! Wow, wouldn't that be something unusual. Instead, they change the forecast every second five minutes and in the end they even fail the "look out the window" weather test.
Polling has looked at this model and realized there is a lot of profit in pretending they are doing something else than what we want them to do (predict the winner accurately), and so they talk about snapshots, about how it's really okay that the polls keep changing, see, because they can't be expected to reflect anything about the final Poll Nov. 5. (!!??!!) Chaos theory rules in both weather forecasting and polling and since they can't do anything about that butterfly in Hong Kong, they cash in on our interest without fulfilling our need.
I think I am switching over to fundamentals as a model for election prediction.
🤣🤣🤣🤣🤣
You ignorant bucket of slop. Clutching your pearls about 10 days out? Who TF even pays attention to 10 day forecasts? Why?
Contemporary 1 week forecasts are around 75% to 80% accurate, 5 day forecasts reach into 90% range. 24hr.
And WTF do you mean by “stand by”?
PS “ I think I am switching over to fundamentals as a model for election prediction.”
PRICELESS! 🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤦♂️
It happens some of the time, not “all of the time,” although that is an expression usually used to signify that “less likely” things happen often enough.
Having $10, or even $100 in your pocket ain’t no guarantee of basic understanding of a topic you profess interest in
There have been ~80/4=20 presidential elections since the start of modern polling. Consequently it's not too unreasonable to expect that choices in polling model formulation should give probabilities varying by about 100/20=5%. Now, maybe it's just numerology, but Twitter was recently going nuts about how Polymarket is about 2∙5%=10% different from Nate's model.
With that said, until I sat down and wrote this comment, I expected my final numbers to be 2∙3%=6%. I'm not sure where that came from, but if it were true, then a 10% variation would reflect a surprising disagreement about the nature of US presidential elections.
Being fair, what does “substantial” mean?
A big enough difference that you'd change your thought/behavior on it. Both are in the territory of a tight race and would lead people to think about it as a tight race. Neither are so far off 50/50 that you would think "This person is the favorite to win".
In terms of this specific race, there's been commentary about how campaigns take different actions depending on if they think they're ahead/tied/behind. 55% to win is still close enough that you would campaign as if you were tied, as opposed to starting to take actions as if you were ahead/behind. Not having a third debate is a good example of that. If Trump thought he was behind, a third debate might be a smart Hail Mary play, but he wouldn't want one if he thinks it's tied.
Changing your thought isn't a binary state, though. Any small change in the inputs should lead a Bayesian to make concomitantly small adjustments to his priors.
Trump is kind of unique though—that would hold true for an ordinary candidate. But given that he kind of had his ass handed to him not because he made normal debate mistakes but because he sounded insane, he probably knows deep down that a debate against Kamala Harris is not in his interest. Of course, he cannot admit in public that he lost, which is another reason not to do it again. So his refusal to debate is really neither here nor there—it’s just not his venue.
“Muhhh Red Wave”
Another W article from Nate. You are going to get hated on no matter what way the election goes, but real ones understand how valuable your predictions are.
I thought the whole point was that it’s not a prediction, it’s calculation of probability?
It's a probabilistic prediction. 🤷♀️
There is simply no predicting, the word just doesn't belong. If you walk up to a roulette table, each spot gives you different odds... Nate is just trying to tell you the odds in a game where the table and odds are continually changing. Placing a bet on a roulette table is not a prediction; the lottery picks winners regularly but it doesn't change the odds for you being a winner.
Serious question -- how valuable do you believe a coin flip prediction actually is (outside of deciding whether or not to wade into the prediction market.)
I think its extremely hard to look at a 55/45 split and not have your own personal preference, beliefs, or desired outcome cloud your judgment into thinking 55 is more of a lead than it is. The higher the stakes, the worse that problem is I think. I’m not anything close to a statistician, but I could be reasonably called a subject matter expert on politics… and I’m constantly having to explain this to problem to colleagues, friends and family. I appreciate how hard Nate works to provide us with examples and analogies to help those conversations go a little easier.
I allow my heart to round 55 up to 99.
I try to keep my head at 55, tho.
Yeah, psychologically speaking I think most people can understand basic probability, but they also don't know how to think objectively enough for it to matter.
One's tendency to think of this as a life changing event, and then propensity for optimism or doom spiralling, for example can have a major effect on their perception of the likelihood.
I always think about it this way... You have a bag with 100 balls in it. 55 are blue, 45 are red. How sure do you feel that you can reach in and draw a blue (or a red) ball? How much of your next paycheck would you bet on either one? As a non-gambling, regular person, my answers are "not very" and "none".
"how often do I get to bet on this?" is also important.
If you will let me keep pulling bags out of the ball, and bet on the outcome each time, there may be no end to how much money I will sink into it (though there would be a cap on how much I would bet on one individual pull).
Yup - it is like the odds in favor of a casino. At +2% and a big enough bank account, you just keep going and let the money roll in.
Unless you are Donald Trump of course.
Indeed! If you can draw as many times as you like, and drawing the blue ball pays double your bet and the red ball zero, you should bet 10% of your bankroll every time- thank you, Kelly criterion.
Rationally, the same applies even if you only get one draw.
His point is that you don't get to rerun the election over and over again. You get that one individual pull.
Yes. I think the "next paycheque" thing is important though.
How much of a fixed and very limited amount of money would anyone rational bet? Not much, I guess.
As a D&D nerd, I find myself thinking in dice. So correct me if I'm wrong:
Kamala Harris has a d20, metaphorically speaking. Polymarket says she wins if she rolls 12 or higher, Nate Silver says she has to roll a 10 or higher.
Either one's a pretty nerve-wracking roll, given the consequences.
Yeah, you're correct and I think in dice gaming turns as well. I've played to win tournaments with prize money on the line and it really brings home how when 1/20 odds happen quite often!
These things are pretty intense when it feels it will have a significant impact on the world my children grow up in.
The idea is correct, but as of the last time I checked, your numbers are wrong. the correct version is: Nate's model says that Kamala needs to get a 9 or higher. Polymarket says she needs to get an 11 or higher.
On a d20, a perfect 50/50 would be 10 winning faces, and 10 losing faces. In that scenario, the fails would be 1, 2, 3, 4, 5, 6, 7, 8, 9, 10. So "11 or higher" is 50%. From there, we add or remove a number to get the 45/55 and 55/45 odds.
For the D&D nerds: yes, that does actually mean death saving throws (pass at 10 or higher) are slightly tilted in the player's favor.
Oops! Lmao. Well you know what they say, there are only three hard problems: naming things, and off-by-one errors
I don't disagree with any of this (it'sactually quite good), but I view things slightly differently, and it ties to something I've been thinking about lately.
Even though polling is closer than it's ever been and most models have the election at a toss-up, it is actually quite unlikely that the 2024 election will be closer than the 2020 election, and there's a decent chance the election won't be close at all. In 2020, Biden won WI (which gave him his 270th electoral college vote) by about 0.6 pt. Meanwhile, Nate's model has P(decisive state being decided by 0.5 pts or less) at only about 10%.
So if the result potentially ends up being not as close as everybody thinks it will be, people will inevitably bitch and complain, demonstrating how the pollsters and modelers were wrong, when in reality, people (still) don't understand how variance and volatility work and how it relates to probability.
I don't know if you're a subscriber, but Nate had an earlier post (https://www.natesilver.net/p/the-128-paths-to-the-white-house) that points this out explicitly: of all the possible ways the 7 swing states can fall, by *far* the two most likely according to the model are either all 7 going to Harris or all 7 going to Trump. As the model sees it, any deviation from the polling average is likely to be highly correlated at the state level. It's not fifty-fifty because we're very sure there's a bunch of races that will individually be very close, it's fifty-fifty because the range uncertainty is wide enough to cover moderately strong wins for either candidate, in a roughly equal manner.
Baseball has also taught me to always take statistics with a grain of reversion to the mean.
In baseball, mean reversion is more of a thing than in other sports. Because there are nine batters and pitchers can’t play every game, no one player can dominate the way a goalie can dominate in hockey or an nba star can carry a team. Even an nfl quarterback has much more leverage than a strong center fielder.
Yes, it's true that baseball's structure leads to a more pronounced mean reversion due to less dominance by single players. However, the underlying concept of reversion to the mean applies broadly (for example, Nick Foles’ 2013 season vs his career). My main point was that in any field involving statistical analysis, from sports to political polling, it's important to consider (and remember) how results generally trend back toward the average over time.
To patch up the sports analogy, the final score is the EC count accepted by Congress. We’re watching a game with imperfect know of the actual points/goals/runs that have already happened. That’s only revealed at the end of the game.
Edit: Basically this is panel judged boxing match, where scoring is announced at the end of the match. A lot of the time it’s really obvious who is ahead, but occasionally it’s vague enough that there’s reasonable doubt until the end
And occasionally the judges were apparently watching a completely different match than everyone else and the clear loser is declared the winner.
If it was a Don King run fight they might have been watching their Swiss bank account? 🤣
Yup
I think the polls will be wrong again this election. Except they will undercount Harris votes instead of Trump votes. We will find out in 4 weeks
Pollsters have been making an active effort to avoid underestimating Trump after doing so for two cycles in a row. That does make me think the chance that they underestimate him is lower than normal, but I don't have any evidence for that.
https://www.nytimes.com/2024/10/06/upshot/polling-methods-election.html?unlocked_article_code=1.QE4.vM5I.1z8G-fzW8nUa&smid=url-share
I think they may go the other way and underestimate Harris voters. Look at the 2022 mid terms. The democrats out preformed their polls
Iiirc there wasn't a systemic error in 2022 but rather they underestimated Democrats by a bit in a few key races.
Exactly. I think the same will happen here. The 3% that are undecided and the 5 or so percent that could change their mind will swing to Harris. They would of swung to Trump alalready if they were going to choose him
Perhaps you’re right, but I’d be just as hesitant to be confident of that as to be confident that Trump is underestimated. Harris is still considered more acceptable in polite society, and Trump’s voters (based on from what I would characterize as a lack of civic engagement beyond voting, not “shyness”) still may be undersampled.
The best case for Harris being currently underestimated is, in my opinion, that her marginal voters are probably higher-propensity than Trump’s marginal voters.
We’ll see.
The case for Harris being underestimated is the idea that pollsters got a ton of egg on their face because they missed in the same direction twice. They are strongly incentivized not to underestimate Trump again, while underestimating Harris wouldn't look as bad for them.
Pollsters are well aware of "shy Trump" voters, and after two cycles should be able to correct for it.
It would be interesting if some pollster would post their poll, and then also post what their poll would have shown if they still used the same methodology they used in 2016. I would expect that the 2016 methodology would show Harris with a much larger lead than the 2024 methodology.
Last I checked in 2022 the Republicans still won the national House vote by about 1%.
Hmm. Wiki says the 2022 House vote was R+2.7
Late polling was R+1.2 at 538, and R+2.5 at RCP.
The "shortfall" was compared to the expected benefit from the House gerrymander. in 2020, the D's were +3.1% in the popular vote, but only +2% in seats.
If the relative difference had held, the R's in 2022 should have have a 16 seat majority, but they only got a 9 seat edge.
Of course extrapolating 2022 to 2024 is risky because of off-presidential election effects. The total congressional vote in 2022 was 30% less than in 2020.
"Late polling was R+1.2 at 538, and R+2.5 at RCP."
I mean it seems peculiar to me that a lot of Democratic partisans seem to be hanging their hopes on an election where the R vote was still underestimated (although by not as much as the Presidential elections) and the R's won at the national level.
I don't know what late polling means, but the final 538 prediction was Rs winning 230 seats and Rs winning the national house vote by 4%, so Ds outperformed those two benchmarks.
The question is _how_ they are tweaking their numbers. If, for example, they give extra weight to blue collar whites without a degree do they risk missing blue collar minorities without a degree if that latter group makes up an increasingly large share of "shy" Trump voters?
Yes, of course the specifics of the adjustments are important. The article I posted states that many pollsters are weighting based on recalled 2020 vote, which is an odd metric to use. Many voters in 2020 will not vote in this election, and many non-voters in 2020 will vote in this election. Plus some people will have had their minds changed in one way or the other.
My point is it's a moving target and a complicated task.
The polls may be underestimating Harris vote share to avoid a repeat of the systemic Trump error in the past, however, DJT seems to be hovering around his ceiling in the polls and the ballot box, which has historically been between 46% and 47%. DJT is arguably the best known political product in the modern era, and yet has never achieved a majority of the vote in either elections, and seems to have hit this ceiling in the polling averages. If indeed Harris polling is being underestimated, as I agree, it is likely she will, as did the last two Democratic Presidential contenders, achieve a popular vote victory. The real question, is can she get over the 51% threshold needed for the EC edge needed? And is it 51% or closer to 52?
I don’t think it’s actually a popular vote threshold she needs to get over, she just needs those votes in the right places. A few tens of thousands of votes in battleground states are more important than millions of votes nationwide.
Agreed. My question/argument is that there is a National popular vote threshold that will carry her over the finish line in those battleground states. For instance, Biden received just over 51% and carried MI, GA and PA, Clinton on the other hand barely topped 48% and lost MI, GA and PA (and others).
2/3 of polling organizations are using weights to compensate for past polling errors in favor of Dems.
Nate has said that he believes there is not a long term systemic bias in the polls, ie random walk; the systemic Trump error of the past doesn't mean it will be there this election too, like you said, the polls could be overcompensating for it. Time will tell.
Yes, I love this number. I use 2.1 national popular vote lead which lost the election for Hillary, and 4.1 national popular vote lead that won the election for Biden. So it's a useful shortcut for me to look at the national popular vote polls. Hey, at least the n is bigger.
You haven't heard much about third-party votes since RFK Jr dropped out, but I'm guessing they may wind up being critical. Total third-party vote share was 5.8% in 2016 and 1.8% in 2020. My feeling, admittedly based on "vibes" more than anything else, is that it will bounce up a bit this year, maybe to the ballpark of 3%. Harris winning the national popular vote 50-47 would then be plausible. The question is how many third-party votes would come at Trump's expense in the key states.
Looking at MI, where Kennedy is still on the ballot as the Natural Law party candidate, it's interesting to look at recent third party impact in the state. Gary Johnson in 2016 running as a libertarian received 3.6% of the vote. Hillary barely lost MI in '16 to Trump 47.5% to 47.3%. In my opinion, Gary Johnson provided a "safe harbor" for "Never Trump" and "Not another Clinton" conservative voters. Arguably, there is a fine line between an old school conservative Republican and a Libertarian. Some Libertarians, e.g. Justin Amish, like to hang out in Republican ranks understanding the two party system is what it is. (I limit my rambling on this). In 2020, in MI, there was not a viable Libertarian alternative for Never Trump voters, and we saw 3rd party vote share drop by 2.5% (112,000) votes, while Trump only gained .3 % more than '16, allowing Biden to turn the state blue garnering 50.62% of all votes cast. Now MI experienced record turn out, and it is probably intellectually dishonest to suggest that all of the Libertarian vote decline was Never Trump R vote from '16 who found Biden to be less enough of an evil than Hillary and Trump enough of a menace that they voted D, but Biden won MI by about 150K votes, and the 3rd party decline was about 120K ish. That's all a long-winded explanation to suggest that if Kennedy presents the same "safe harbor" for the "Never Trump" voter as Johnson in '16, the third party voted could impact Harris more than is currently anticipated. I don't adhere that school of thought, as Kennedy seems to have lost all credibility since endorsing Trump, and Trump has never received a greater share of the vote in MI that that 47 ish %, right where he seems to hang out in all polling. Like I said, fun thought experiment.
Yes, 3.6% is a LOT in an election as close as this. Michigan is delicate because of the Muslim thing at least, and maybe the union thing too. As you say, analysts blamed third party voters for Hillary's loss in 2016 and a narrow loss for that reason could happen again. Certainly Kennedy is trying to help Trump in this way. I like your point.
"Kennedy seems to have lost all credibility since endorsing Trump"
No, I think he's lost credibility since he said his brain was eaten by a parasite -----
But I can't see why the RFK Jr. vote would matter. Because I was looking at a NYT graph today that showed the bump-up in votes after RFK left the race and BOTH candidates profited exactly equally: the graph bumped up for both at the same place, same amount. Why Trump offered RFK a job, I don't know --- he must know something I don't understand. Okay, it's a trade for RFK going around taking his name off the ballots in battleground states, so they must think if people can't vote for him, they'll vote Trump instead. It's not how the graph looked ---
Agree. I think Trump is still leaking some blood from the debate fiasco. And the recent good economic news (lower interest rates, lower inflation, good jobs report) will give Harris a small boost.
This is a topic Nate could revisit again and again. If there is a bias in the model or the polls this might be a bigger effect than the 55/45 thing.
Agh, yes, this is so annoying. A few weeks ago, when you had Trump as 62% likely to win briefly, the Daily Mail (which is admittedly only slightly more reliable a news source than Fox News) had a headline saying you were predicting a "blow-out win" for Trump. I suspect they do know this is not what probabilities mean, but pretending they don't gives a far better headline
Nate we need your take on the Detroit Tigers!
Also, re the election, when Harris smoked and depantsed Trump in the debate and THAT did not move the needle… unfortunately feels like people just want Trump back
The debate did not move the needle because while Harris voters were ecstatic at her performance and Trump voters disappointed that he was totally by her bait , I did not hear anyone say that she provided any substance or clarification of her views. In fact, most undecided voters felt that she disappointed them. Style points alone never get you a real win in any sport of which I am aware, even if they make you look good. Her interviews, even with sympathetic interviewers, continue to be content free and/or word salads.
I don't think you know what word salad means. You can say you don't like what she's put out there so far. You can say her plans aren't substantive enough for you. But only one candidate is spewing incoherent nonsense. Just because you're accustomed to unhinged language from him doesn't mean it isnt much closer to word salad.
It's not word salad, technically --- I've seen that, idiopathic schizophrenic talk. But she does have pretty bad perseveration, a brain damage symptom: all that repetition of the same word or phrase that she catches onto and then repeats throughout the sentence and the paragraph, making no sense. She is 60 --- it could be mini-strokes, I suppose.
How many people do you know who do that perseveration on one word? Right, none.
Lol most women with jobs ranking higher than manager whom I know talk this way. Me included. It's how we're trained not to appear too threatening to men.
https://m.youtube.com/watch?v=r1RmWKA5Vaw
That isn't a word salad. If it is "unintelligible" to you, it indicates your comprehension problems.
As Crocodile Dundee would say, "You think that is a word salad? Now this is a word salad".
Q : "If elected, what specific legislation would you seek to pass to lower childcare costs in America? "
Trump : "Well, I would do that. And we’re sitting down – you know, I was – somebody we had, Sen. Marco Rubio, and my daughter Ivanka was so impactful on that issue. It’s a very important issue. But I think when you talk about the kind of numbers that I’m talking about, that – because, look, childcare is childcare, it couldn’t – you know, there’s something – you have to have it. In this country you have to have it."
Nate is absolutely right that there's no meaningful difference between a 40-60 probability vs a 60-40 probability for a one time event (unless you're a gambler. Unfortunately none will need Nate's advice.
Nate's point on the precision of polls is also important. If you wanted to know what margin would be rounded to the nearest 10th (Trump +10, tied, Kamala +10, etc) polls would rarely be wrong! Polls are even useful if you want to round to the nearest 5th (Trump +5, tied, Kamala +5, etc). But happens when the race is Kamala +2.5? If you trust RCP's average of Kamala +2, you likely think Trump is the favorite. If you prefer Nate's Kamala +3, you likely think she is winning. But those are arbitrary and self motivated reasons for preferring an average that are subject to high degrees of error.
The race is 50-50, it's always been 50-50, it always will be 50-50.
"Nate is absolutely right that there's no meaningful difference between a 40-60 probability vs a 60-40 probability for a one time event"
Nice!! I love that. Never thought of it that way --- probabilities with just one throw.
Absolutely love this, and why I’m a paid subscriber (Trump voter, btw…holding my nose).
@Nate - Question: If you were to frame the race in sports betting odds (-110, +215, etc), what would it look like? For example, something like, Harris -115, Trump +105…?
https://www.aceodds.com/bet-calculator/odds-converter.html
size of the vig matters, but this chart has 55% at -125
Thank you! Excellent info!
One thing that Nate is leaving out is MOMENTUM. In sports, this plays a big part in who winds up winning a particular game. I have no doubt that TRUMP has that momentum at the present time. When a candidate can draw a crowd of 80,000 people to his rallies compared to 5,000 for Harris (many of which are bused in by the candidate) I have to believe that Trump's base is fired up. The Sportsbooks now have Trump as the slight favorite in the betting odds. They are the best indicators of who is going to win or lose, IMO. When I see Harris answering questions with a bunch of word salads & that ridiculous habit of shaking her head yes, at the end of every sentence, I see a person who is not being authentic, & quite frankly not fit to be the leader of the free world. But, of course, we have to factor in the elephant in the room. The legacy media & multinational Corporations all want Harris to win. And make no mistake, they are the ones that are both manipulating & counting the votes. That may be an obstacle that Trump will not be able to overcome. In a free & fair election my money would be on Trump, however in today's corrupt political process, I'm probably going to bet on Harris. Sad but True.
Have you heard a 5 minute slice of anything from Trump? If it weren't for the sanewashing in the media he would have been dropped from the election before Biden.
Harris on the other hand is only unintelligible if you have problems with complex concepts and rhetorical devices.
This is the typical Trump Derangement Syndrome response given by the average leftest loony tune. Makes absolutely No Sense.
🤣🤣🤣🤣🤣🤣
Every single time I’ve seen someone mention TDS they are the most unhinged MF in the room. Your every accusation is a confession 🤷♂️
Very thoughtful response. Now, go back to mommy.
🦗🦗🦗
If you’re having trouble finding your mommy, letting you know I've got her out working her regular corner. 👍🏼
It was well above the standard your “thoughtful” post warranted you hypocrite. 🤣🤣🤣
But please, try spin this as anything other than word salad gibberish: https://m.youtube.com/watch?v=XCzF0Qg4M4E
I’ll wait here….
I have no interest in debating someone who has TDS. NONE!
The Harris people are trying to say her crowds are also big ---- doesn't look like it, though, and this enthusiasm gap is VERY reminiscent of the problem Hillary had. Both Kamala and Hillary have small inside groups of bused-in audiences ------ huh. Probably not a good sign.
🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣
https://ash.harvard.edu/articles/the-real-numbers-tracking-crowd-sizes-at-presidential-rallies/
As a former Democrat, let me also state that everything that the Democrat Party accuses Trump of either doing or going to do, is something that they are either doing now (at this very moment), or plan to do in the future, if they remain in power. The only End To Democracy that will happen is when the federal government controls what we are allowed to say, either in a private or in a public venue. George Orwell predicted what was going to happen. The only thing he got wrong was the year it would take place. Be warned!
🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣🤣
🤫
The highest rated legacy media company in the US is Foxnews, last time I checked they weren't exactly rooting for Kamala Harris.
Fox is not a legacy media company. CBS, NBC, ABC CNN, MSNBC, all push the legacy corporation party line. Legacy media refers to traditional forms of mass communication that have been established for a significant period, such as newspapers, television, and radio. These mediums were dominant before the rise of the internet and continue to shape public discourse and information dissemination today. Get a clue before you respond.
Ok if you say so. I'm sure Rupert Murdoch would be surprised hear that but you know whatever.
My friend = Google "Legacy Media" before you respond in a way that will make you look foolish.
🪞🤡
https://en.m.wikipedia.org/wiki/Old_media
“ These conglomerates are often owned and inherited between families, such as the Murdochs of NewsCorp.”
Fox is far from being "Old Media" such as CBS, NBC or ABC.
I just saw a video of a nearly empty Trump rally. Don't blame the media, it was a video recording, and Trump was talking to a mostly empty room. They panned across the room, and most attendees were behind Trump. Nobody really has momentum at the moment.
PS - There is no factor in the model for momentum. Nate addressed this a while back when Harris had the momentum. If somebody has momentum, it needs to show in polling.
If someone has momentum, it WILL show in polling, unless the polls are corrupt. Which they may be: keeping it close to keep us all clicking. I remember George H.W. Bush used to call it "the Big Mo."
There can be a Big Fade, too, as when Comey interfered with the Hillary election by tying her to that Anthony Weiner character, because he had one of her old laptops and was sexting on it.
Or when little Dukakis wore the oversized helmet in the oversized tank and looked like an idiot.
Right. Harris had true momentum at the start of her campaign, and it clearly showed in the polling at the time.
Right! The Butler, Pennsylvania crowd was very small??
Did I say that? I don't think so. Butler v2 was obviously a special case.
Rally size really doesn't mean much. We all have no doubt that Trump has a deeply devoted base of followers who will show up, even if it kills them (see 2020 superspreader rallies, for example). That doesn't win elections.
Big crowds, however, HAVE to be a better sign than bussing in a hundred people because nobody will come otherwise. I think Trump has always been correct to focus on that and publicize it. It's impressive. Dems often sneer at Trump's "deeply devoted base of followers" -------------- but I bet Harris' campaign wishes they had that.
Harris campaign definitely has that. I live in a purple part of a blue state and am experiencing it directly. It's not the same as Trump's base, but it is there. By the way, it includes a number of Republicans who voted for Trump in past elections.
PS - Note that Ralph Nader had multiple "super rallies" with well over 10,000 attendees - and he got 3%. Rallies have value for getting media attention and volunteers, but mean little otherwise.
Nate can you talk about weighting on recall vote, as per the Nate Cohn articles?
What's the state of the race/model if you only look at polls that avoid weighting on recall?
Which is more misunderstood, Nate's EC win probabilities or Rotten Tomatoes' Tomatometer?