“If polling firms were still applying the same techniques they did in 2016 and 2020, we’d probably be seeing a Harris lead in the Electoral College right now. Instead we have a toss-up, more or less”
Why haven’t any outfits released a “here’s what our data would look like given our 2020/2016” approach? Would be very informative to this discussion, but maybe it would highlight the large amount of subjective modeling work at play in a way that pollsters would prefer not to talk about.
I suspect that some of the difference is the methods they are using to get in touch with voters and get them to respond to the poll. This is going to be harder to “undo” than just changing the likely voter model.
In any case, I’m not sure why the modeling is said to be “subjective” here.
That analogy doesn’t work, as other commenters have pointed out. A lot of decisions go into constructing a model, as Nate has made abundantly clear in many many posts. Not only what to include, but in the choice of parameters. Not even remotely like a ruler. And when it comes to pollsters, just look at the whole “likely voter” issue.
Of course, that doesn’t make model construction *purely* subjective. Arguably it’s more objective than subjective. But it was obvious to me what Casey meant. I disagree only with the claim that “pollsters prefer not to talk about” it. I’ve read more articles on these issues in the last few weeks than I can count.
You misunderstand the analogy. The instrument choice is the model choice. There are actually choices to be made about the instrument to use.
We don’t refer to this as “subjective measuring” because, again, that would remove meaning from the word subjective. Further it would imply existence of “objective modeling”, and of course by this standard that’s just not a thing. Even if someone else built the model you’re still choosing to apply it.
I think I understood the analogy just fine. It’s a bad analogy because basically the only relevant variables with measuring lengths is the desired degree of precision, and the range of lengths. Yardsticks made of wood, metal, or plastic will give the same results. That’s far from being the case with these kinds of models.
I think you misunderstood what Casey meant by “subjective”. You’ve constructed a strawman definition. He didn’t say “subjective measuring”, he said “subjective modeling”. If your beef is that he should have said “modeling using models in whose construction subjective choices were made”, okay, fine…
** Actually even laboratory ruler use is slightly subjective as you record a digit more precise than smallest gradient.
Using “subjective” like you are makes everything subjective, word loses meaning. I look forward to ignoring your philosophy 101 essay assignment defense of this
There's a lot of subjectivity in polling. It's not an exact science. Previous methods of reaching prospective voters don't work any longer. Response or non-response biases. Shifting party identification. Overweighting certain sub-groups. Case in point this year most polls still use D+3 to D+7 weighted models, when Gallup shows it is R+3. If this is true, those polls will miss outside the margin of error.
Cool, so let's bet, you and me. I've already put my money where my mouth is by placing multiple bets on Kalshi. But I find your side never has the strength of their conviction to put money on the line.
Simple bet. I bet that all the big polls will again miss election results outside the margin of error (2.5%).
Agreed that it’s likely impossible to truly reconstruct data using past models, but you could certainly experiment to get a bit closer and understand the impact of the change.
For example, some outfits have chosen to include “drop offs” this year; folks who confirm they’re voting for a candidate but hang up before the end of the call. This was theorized as one of the reasons for undercounting Trump-specific support in past elections.
Well, now firms have some actual data; was the theory true? What do the results look like if you include/don’t include drop offs?
One thing I don’t understand is how they actually include drop-offs in the results. Like, if the only information they have about them is who they’re voting for and where they live (and maybe they’re voter registration data in states that make that available), how do they know how to weight those respondents when they do demographic adjustments? I guess they could use the demographic % breakdowns for D/R voters in the area they live as a placeholder, but even has potential for skewing things if Trump/Harris dropoffs in that area tend to have different demographics than those candidates’ voter in general. And if their LV screens are in anyway based on questions in the survey, how do they treat drop-offs on that front if they didn’t answer all the questions?
Other Nate goes into this in the second of the NYT articles linked here (the first is a very interesting summary of the main theories for why that polling error existed in 2020/2016). The most wide-spread method for adjustment, according to him, is "weighing by recalled vote". I only kinda understand the methodology as he describes it, but it boils down to getting info on who respondents voted for in the previous presidential election, then adjusting how those voters are accounted for by demographic. Most polls are using some form of this method
In practice, this means that polls are being adjusted to the results of the previous election (2020, midterms aren't a factor). Cohn goes into the potential pros and cons of using this strategy (Pro: it would account for another pervasive polling error in Trump's favor, Con: it would have made polling results LESS accurate in every 21st century presidential election).
I'd really recommend that article, it's super interesting. There's no "remove recalled vote weighting" button, but Cohn does compare their model to a version of their state averages that only include polls without recalled vote weighting. It aligns a lot closer to 2022 than to 2020: Harris increases her leads in the midwestern swing states, Trump increases his in Arizona and Georgia, while Nevada and North Carolina are basically the same. That results in a slightly lower popular vote advantage for Harris, but a relatively significant electoral college lead
(It's important to note that this article is from 2 weeks ago. The Silver Bulletin model leaned 54% Harris at that point)
I saw that article, and it’s part of why I’ve increasingly wanted more data from some of the polling outfits.
Cohn didn’t do a great job explaining the mechanism, though. Basically, it’s a selection issue: there’s an increased likelihood that political moderates who actually voted for Trump will instead remember themselves voting for Biden. Given their past lean, these folks are also likely to vote for Trump again, giving pollsters the mistaken impression that a Biden 2020 voter is now a Trump 2024 voter. Add that up across the electorate, and you get a tendency to overstate the chances for the loser of the last election.
Appreciate that clarification. He does a much better job explaining it in their most recent newsletter, the older article would've benefitted a lot from a simplified explanation. I do wish individual pollsters in general, especially the really big-shot ones, were more open in showing how different adjustments and weighting affected their results.
I really appreciate that Nate does it here, but that's after aggregating all of those other polls. There's no possibility to see how decisions on the more subjective aspects of polling have changed that final number. More importantly, we're not seeing WHY those decisions are being made in the first place. Other Nate's articles were fantastic not just because they get a bit into how adjustments to methodology have affected the polls, but also they describe pollsters' reasoning for making those adjustments in the first place (which seem relatively sound imo)
The only question I have about this is that, could that bias impacted by the fact that (for the first time since 1956) the loser of the last election is literally on the ballot. Anybody who actually did vote for Biden and now is voting for Trump may see telling the pollster that as admitting a mistake, could they possibly then lie and say they voted for Trump before? And could the same effect mean that there are less people who voted for Trump last time who say they voted for Biden? Or if some people aren’t being intentionally misleading and instead just misremember due to a bias toward the winner (I don’t know how somebody could forget who they voted for in any race, much less a recent one) could the presence of the same candidate again jog their memory so they recall correctly?
Great thread. Thanks for the clarity. Well done. Also agree with comment about how aggregation hides the details, including important ones. I’m more convinced than ever that polling error will likely favor Harris. But I also recognize that I might me confusing my wishes with reality.
Outside the few true scam artists, for those that have even a modicum of professional self esteem, you should count yourself lucky if you can get them to STFU about it. 🤣
Sorry, but that’s not the same thing; that’s just applying the “error” from each state to current results. Instead, I’d like to see the results if pollsters ran their current data through their older models.
That headline is from The Telegraph, which isn't exactly the most reputable news source. Also, FWIW, they recently changed the headline to say "analyst" instead of "pollster", so I suspect several people told them.
Nate knows this happens. He joked about it in another post. He can't be responsible for how media writes headlines. It is funny, though. Nice post, detailing that.
FWIW (nothing, as Nate persuasively argues), my gut is Kamala, because every single data point I have *other* than the polls says its absolutely insane that Trump could win. I realize that this is part of being in the "bubble" of people who are aware of things like how Trump Chief of Staff John Kelly reported that Trump wished he could have "Hitler's generals" but was unfamiliar with who Rommel was.
"every single data point I have *other* than the polls says its absolutely insane that Trump could win"
Another data point: Democratic Senate candidates in PA, MI, WI are **currently** running campaign ads touting how they voted for some Trump bills when they were in Congress. Why would savvy, experienced Democratic politicians start advertising to their voters that they sided with Trump? Costs a lot of campaign funds to run those ads.
I mean, they are responding to the polls, so it's really the same data point.
I meant more things like:
1) enthusiasm I see around town (lots of yards in my neighborhood with signs for the R congressional candidate but conspicuously no Trump sign - which is different than 2016 or 2020)
2) social media enthusiasm levels (although this may in part reflect some self-segregating, as conservatives have adopted Twitter as their platform of choice and liberals have fled it -- but there is still a sizeable conservative faction on Reddit, and I'm just seeing a lot less pro-Trump stuff than I did in 2016 or 2020)
3) Trump's promises v. Trump's record -- it's one thing to believe the guy's rhetoric about helping the little guy in 2016 (which is why I very much thought he could win back then), but it's another thing entirely after he governed like a typical Republican, with a booster shot of chaos and insanity. But again, I recognize that I'm probably in a bubble that is aware of what Trump actually did instead of just having vague nostalgic vibes about the pre-covid times.
What about early voting? The data I've seen looks good for Harris in PA, MI, and WI and more mixed in the other swing states.
But it's hard to take that seriously because there's no point of comparison. There's no Covid to drive a ton of mail in voting like 2020, but both sides are encouraging their voters to vote early (unlike 2016).
Party Identification is R+3 for the first time in a long time. If polls use D+3 to +7, they will miss bigly.
The trend is towards Trump in all polls and betting markets.
Republicans made massive gains in voter registration in key swing states, thanks in part to grassroots orgs like Turning Point, Scott Pressler, Chase PA, 100X, etc.
Early data coming out shows massive turnout in red counties, and those with propensity data show it is 0/4 or 1/4 Rs coming out, so they are not cannibalizing ED votes.
Miami-Dade is R+10 in early voting right now, insane.
Nevada is R+ right now as Jon Ralston admits.
Articles coming out daily on the panic setting in amongst Dems.
There was a significantly larger number of relatively high profile Republicans who endorsed the Democrat this time around, so it's hard to say how many of those +R advantages are actually Harris votes.
Polls consistently show both Harris and Trump are getting 95% of each party's respective voters. There's not going to be a mythical large number of Rs voting for Harris. If so, it would be canceled out by the Ds voting for Trump (RFK, Gabbard supporters, etc.).
The sign thing is obviously anecdotal, but it’s hard not to notice. I’ve seen more Harris signs than I did Biden,l and Clinton—I saw almost no Clinton signs. I still see Trump signs, but not as many as 2020 or 2016. I saw a ton in 2016.
I wouldn't call the words immortal when she never said them. Here's the actual quote: "I live in a rather special world. I only know one person who voted for Nixon. Where they are I don’t know. They’re outside my ken. But sometimes when I’m in a theater I can feel them."
She was perfectly aware that she lived in a vacumn chamber, not oblivious to it like the false quote implies.
The "How did candidate X win, I don't know anyone who voted for him" is real. There were definitely people who said that on 1980 in Berkeley about Reagan.
And I've seen similar lines about Trump 2020 used as "proof" that the election was rigged.
Trump helping the little guy, as in median household income adjusted for inflation (up 12.4% for Black working class people, up 8% for white working class during Trump years. Up 0.8% for Blacks under Biden/Harris, down 1.5% for whites).
Sounds like the little guy did quite well under Trump and got whacked under Biden/Harris
"You and I both know that Trump's policies didn't have anything to do with that "
Restricting illegal immigration to 1/6 of its current numbers which reduces the supply of low-wage workers (ever heard of the law of supply and demand setting prices and wages?).
Replacing NAFTA with new trade pact that has specific targets for US content in manufactures imported from Mexico.
Imposing tariffs and technology restrictions on China to prevent US trade from enriching our most powerful adversary
Inflation at less than 2% for the duration of the term
I understand this may not feel intuitive, but most people who identify as conservative but who are still on the fence about Trump, hate him as a person but like a few of his policies.
The goal for senate / house Dems running in swing states is to pull the few palatable Trump policies they support, while drawing contrast to between them and the far right policies supported by Trump and the Republican running against them.
That would clearly show this ad approach by Kasey in PA as smart and not at all in conflict with Kamala having just as good a chance to win PA as Trump.
This is the problem with the popular stat driven side of politics and the sometimes misguided out of touch understanding of how normal people think and vote. Most voters are complicated and have contradictory view points. You can’t just do one-size fits all approach to reaching the persuadable voter.
Also, if Kasey was so afraid of tying himself with Harris, he wouldn’t have stumped for her numerous times in the last few months. This is the second or third time I’ve seen someone use this as evidence of Dems giving up or acknowledging Trump will win and that’s just not the case.
The polls have been showing downballot Democrats running ahead of Harris. It points to the existence of a good number of Trump/Downballot D voters (or Undecided/Downballot D, edit - or Trump/Undecided). These voters are definitely worth targeting for downballot Democrats. It doesn't point towards an overperformance for Trump because that data is taken directly from the polls as they are.
Fwiw I'm dubious about the size of that gap. It doesn't make a lot of sense to me. I expect either the downballot Democrats will underperform, or Harris will overperform. But it's pretty clear what the polls are saying.
"Why would savvy, experienced Democratic politicians start advertising to their voters that they sided with Trump?"
This is probably hopelessly naive. But one possible answer is, "because it's good for the country".
Partisanship in the US needs to go down. If we were to suppose that people high up in the Democratic party actually *do* care about democracy, then we should expect them to do that. People who are anti-Trump should recognise that America's problems don't all just suddenly go away in the event that he loses the election (and the loss sticks).
I’d be interested to see whether voters do switch parties. I believe I saw somewhere that switching parties between elections was quite a lot higher than I thought.
And because they don't understand that undermining the Harris campaign by advertising they supported Trump policies in Congress could have repercussions for them if Harris wins?
The media made a big deal out of it, but I kind of doubt that their ads are going to make anyone say, “Oh, wow, I was totally going to vote for Harris, but this one guy said he could be bi-partisan, so now I'm going to vote for Trump.”
Yeah, "absolutely insane that Trump could win" is not based on anything happening in the actual election. It just means, "I don't want it to happen! I don't want it! I don't want it!!!"
Serious question - Do you actually think an undecided swing voter 2 weeks before an election, who is likely apolitical by nature, knows or cares who John Kelly is? Or even cares about a Trump = Hitler comparison as if they haven’t heard it for the last 8 years?
Gotta remember that the people deciding this election are not political nerds paying for polling analysis from Nate Silver.
Thing is, anyone who this would have traction with is already in Harris' camp. The whole "Trump admires Hitler" thing has been reported/alleged before. The fact that CNN is hitting it so hard today I see as a sign of panic.
I didn’t say it definitely happened, but I’d pose the question to you of why everyone who worked with Trump has similar stories. Is the likelihood that all these senior military figures, who served our country for forty years with integrity, are all lying about Trump? All of his secretaries of defense, his chief of staff, his vice president, his attorneys general, his first Secretary of State. They’re all lying? Or Trump is? Sincerely what is more likely.
There are different kinds of swing voters. The moderate Republican pro-choice suburbanite, who has always voted Republican in the past, likely knows who John Kelly is. In any case, should anybody run into a report of that interview, they will be told who he is. But it is the former swing voter, who may even be a shy Harris voter, who would be affected by Kelly's comments.
I believe once there was a survey done of when Boris Johnson kept on saying: Get Brexit done. I think only 30% could recall it. The point being that most people don’t listen to politics and messages need to be really quick and easy to remember. I think Trump has done better here (not perfect) with the key points and why the stories and huge campaign slogans cut through. Unfortunately he needs to do them now.
As another person with a meaningless gut feeling that Kamala will pull it out, I tend to agree.
The other big reason my gut says Harris is that pollsters have much more incentive to avoid underestimating Trump than they do to avoid underestimating Harris, which surely affects them consciously and subconsciously.
My gut says Kamala too largely based on what you said in your last paragraph. Nate said in his last sentence that it’s probably equally likely for a pro-Trump miss and a pro-Harris miss. I think with the weighting by recalled vote being relatively abundant that it’s less likely Trump support is underestimated a third time. It’s either pretty accurate or we are underestimating Kamala’s strength IMO
Saying you want "Hitler's generals" is not an endorsement of Hitler himself. Those officers tried to kill him. Instead, it reflects the high regard for German military leadership in many post-WW2 studies. Similarly, studies have shown German soldiers were objectively more efficient than counterparts in other countries.
Although Kelly said that Trump revealed in the conversation that he didn't know that Hitler's generals tried to kill him, and Trump didn't know who Rommel was. Occam's Razor here: the dude likes Nazis.
What does 'efficient' mean? What studies are you talking about?
Everything I've ever read indicates that the American military was superior to the German military in basically every way except for the fact that the Germans overengineered their vehicles (but even this is really another reason the American military was superior, because America made vehicles that could be easily mass-produced).
Sure. Shame about their leadership (generals, Hitler himself) pointlessly over-stretching them and then needlessly trapping a bunch of them in Stalingrad.
I mean, there are data points within the polls that also point in her favor. For example, the younger a voter is, the less likely they are to say they plan to vote early. Yet, those young voters *did* show up in 2020, and they *did* show up in the Blue Wall battleground states in 2022.
The media loves to harp on the fact that young voters didn't show up as expected in 2022, but they never focus on individual states. Young voters vote when the issues are important to them *and* when they think their vote can make a difference.
Combine that with the fact that none of the Blue Wall states restricted voting (not for a lack of trying, Pennsylvania) since 2020, and I fully expect young voters to show up in force on or before election day.
All of that is a valid analysis based on data. It would be absurd to try to use it to predict that Harris will win. However, it's equally absurd to ignore data like that and claim that Trump will win just because polls have had a tiny change and Republican early voting numbers are up! (I'm not accusing Nate of doing that; I'm speaking generally.)
If that was your point, you would have said it right after talking about the big Trump sign. Or after your first comment about the farmer. But you didn’t.
It’s genuinely fascinating to watch the counflip nature of this election open an X-ray into everyone’s soul.
Nate, professional hater and liberal contrarian, is ranting about Kamala’s weak campaign. The betting markets, mostly a certain type of dude, are going further and further toward Trump. Anti Trump Republican Mike Murphy keeps waffling back and forth. I spend half my time in despair and half my time optimistic.
As someone who is basically completely divorced from the American political spectrum, it's becoming increasingly tempting to just move to Switzerland and bunker down for the next few decades.
Ah the "coin flip". I have been following Nate since 2008 and appreciate the scientific rigor. But my question is this. It seems no matter the analysis or commentary or what-if scenario ("take out RFK", "take out partisan polls"), you could replace the headline of every post to be "Race is a coin toss". If, in a polarized electorate, every election is going to be within a a few points, and if you're going to call any win probability between 40% and 60% a "coin toss" (seems reasonable to me) - then what is the point of all this polling and all this analysis?
For this content specifically, the point is entertainment value. Do you enjoy reading the thoughts of Nate Silver, who thinks about elections in a statistical way? And do you enjoy reading it enough to pay for it?
I sure do. And so do many others. And that's the only reason this exists.
The point of the polling at least, is that that's the only reason we *know* that it's a toss up. Elections aren't always close.
Imagine if voters had categorically rejected Harris after Biden dropped out - maybe she was tainted too much by association. It's easy to imagine that having happened, but the polls are reasons to think that that hasn't happened. It's easy to miss the importance of polls in this, because there's tonnes of vibes that will tell you Harris wasn't rejected. But a lot of those vibes are *informed* by the polls. Vibes tend to be given a lot more weight when they align with polls, and that's one of the few good things there is to say at them.
Yeah this ambivalence from blue collar union leaders seems more driven by an attempt to hedge bets than any political ties, and the teamsters in particular seem driven by internal politics. Many teamster bosses who voted in agreement with non endorsement then turned around and had their locals endorse. While they've thrown around a survey of members as justification, the panel of teamsters members that interviewed candidates chose harris.
Sean O'brien in particular seems to be trying to flirt with right wingers that actively despise union labor, in an effort that is likely to only lose political allies than gain any new ones.
“If polling firms were still applying the same techniques they did in 2016 and 2020, we’d probably be seeing a Harris lead in the Electoral College right now. Instead we have a toss-up, more or less”
Why haven’t any outfits released a “here’s what our data would look like given our 2020/2016” approach? Would be very informative to this discussion, but maybe it would highlight the large amount of subjective modeling work at play in a way that pollsters would prefer not to talk about.
I suspect that some of the difference is the methods they are using to get in touch with voters and get them to respond to the poll. This is going to be harder to “undo” than just changing the likely voter model.
In any case, I’m not sure why the modeling is said to be “subjective” here.
“Subjective” because of the subjective judgement calls that go into defining the model.
That’s not how it works. You don’t refer to a length measurement as “subjective” because you chose to use a ruler
That analogy doesn’t work, as other commenters have pointed out. A lot of decisions go into constructing a model, as Nate has made abundantly clear in many many posts. Not only what to include, but in the choice of parameters. Not even remotely like a ruler. And when it comes to pollsters, just look at the whole “likely voter” issue.
Of course, that doesn’t make model construction *purely* subjective. Arguably it’s more objective than subjective. But it was obvious to me what Casey meant. I disagree only with the claim that “pollsters prefer not to talk about” it. I’ve read more articles on these issues in the last few weeks than I can count.
You misunderstand the analogy. The instrument choice is the model choice. There are actually choices to be made about the instrument to use.
We don’t refer to this as “subjective measuring” because, again, that would remove meaning from the word subjective. Further it would imply existence of “objective modeling”, and of course by this standard that’s just not a thing. Even if someone else built the model you’re still choosing to apply it.
I think I understood the analogy just fine. It’s a bad analogy because basically the only relevant variables with measuring lengths is the desired degree of precision, and the range of lengths. Yardsticks made of wood, metal, or plastic will give the same results. That’s far from being the case with these kinds of models.
I think you misunderstood what Casey meant by “subjective”. You’ve constructed a strawman definition. He didn’t say “subjective measuring”, he said “subjective modeling”. If your beef is that he should have said “modeling using models in whose construction subjective choices were made”, okay, fine…
If we were measuring something as concrete as the physical length of an object we wouldn't have so much disagreement between models either
Tool & methodology is subjective, not the tool.**
** Actually even laboratory ruler use is slightly subjective as you record a digit more precise than smallest gradient.
Using “subjective” like you are makes everything subjective, word loses meaning. I look forward to ignoring your philosophy 101 essay assignment defense of this
correct
There's a lot of subjectivity in polling. It's not an exact science. Previous methods of reaching prospective voters don't work any longer. Response or non-response biases. Shifting party identification. Overweighting certain sub-groups. Case in point this year most polls still use D+3 to D+7 weighted models, when Gallup shows it is R+3. If this is true, those polls will miss outside the margin of error.
Read other replies, this has already been addressed.
Edit: Other than your favourite 🤡🚗 talking point you love to bang on. 🤣🤣 No need to take that even remotely seriously
Wait, so you think a poll that's D+7 is representative of the current electorate?
Cool, so let's bet, you and me. I've already put my money where my mouth is by placing multiple bets on Kalshi. But I find your side never has the strength of their conviction to put money on the line.
Simple bet. I bet that all the big polls will again miss election results outside the margin of error (2.5%).
Deal?
Agreed that it’s likely impossible to truly reconstruct data using past models, but you could certainly experiment to get a bit closer and understand the impact of the change.
For example, some outfits have chosen to include “drop offs” this year; folks who confirm they’re voting for a candidate but hang up before the end of the call. This was theorized as one of the reasons for undercounting Trump-specific support in past elections.
Well, now firms have some actual data; was the theory true? What do the results look like if you include/don’t include drop offs?
Etc.
One thing I don’t understand is how they actually include drop-offs in the results. Like, if the only information they have about them is who they’re voting for and where they live (and maybe they’re voter registration data in states that make that available), how do they know how to weight those respondents when they do demographic adjustments? I guess they could use the demographic % breakdowns for D/R voters in the area they live as a placeholder, but even has potential for skewing things if Trump/Harris dropoffs in that area tend to have different demographics than those candidates’ voter in general. And if their LV screens are in anyway based on questions in the survey, how do they treat drop-offs on that front if they didn’t answer all the questions?
What are you going to do with the drop-offs? Outside of postmortem
Other Nate goes into this in the second of the NYT articles linked here (the first is a very interesting summary of the main theories for why that polling error existed in 2020/2016). The most wide-spread method for adjustment, according to him, is "weighing by recalled vote". I only kinda understand the methodology as he describes it, but it boils down to getting info on who respondents voted for in the previous presidential election, then adjusting how those voters are accounted for by demographic. Most polls are using some form of this method
In practice, this means that polls are being adjusted to the results of the previous election (2020, midterms aren't a factor). Cohn goes into the potential pros and cons of using this strategy (Pro: it would account for another pervasive polling error in Trump's favor, Con: it would have made polling results LESS accurate in every 21st century presidential election).
I'd really recommend that article, it's super interesting. There's no "remove recalled vote weighting" button, but Cohn does compare their model to a version of their state averages that only include polls without recalled vote weighting. It aligns a lot closer to 2022 than to 2020: Harris increases her leads in the midwestern swing states, Trump increases his in Arizona and Georgia, while Nevada and North Carolina are basically the same. That results in a slightly lower popular vote advantage for Harris, but a relatively significant electoral college lead
(It's important to note that this article is from 2 weeks ago. The Silver Bulletin model leaned 54% Harris at that point)
I saw that article, and it’s part of why I’ve increasingly wanted more data from some of the polling outfits.
Cohn didn’t do a great job explaining the mechanism, though. Basically, it’s a selection issue: there’s an increased likelihood that political moderates who actually voted for Trump will instead remember themselves voting for Biden. Given their past lean, these folks are also likely to vote for Trump again, giving pollsters the mistaken impression that a Biden 2020 voter is now a Trump 2024 voter. Add that up across the electorate, and you get a tendency to overstate the chances for the loser of the last election.
Appreciate that clarification. He does a much better job explaining it in their most recent newsletter, the older article would've benefitted a lot from a simplified explanation. I do wish individual pollsters in general, especially the really big-shot ones, were more open in showing how different adjustments and weighting affected their results.
I really appreciate that Nate does it here, but that's after aggregating all of those other polls. There's no possibility to see how decisions on the more subjective aspects of polling have changed that final number. More importantly, we're not seeing WHY those decisions are being made in the first place. Other Nate's articles were fantastic not just because they get a bit into how adjustments to methodology have affected the polls, but also they describe pollsters' reasoning for making those adjustments in the first place (which seem relatively sound imo)
The only question I have about this is that, could that bias impacted by the fact that (for the first time since 1956) the loser of the last election is literally on the ballot. Anybody who actually did vote for Biden and now is voting for Trump may see telling the pollster that as admitting a mistake, could they possibly then lie and say they voted for Trump before? And could the same effect mean that there are less people who voted for Trump last time who say they voted for Biden? Or if some people aren’t being intentionally misleading and instead just misremember due to a bias toward the winner (I don’t know how somebody could forget who they voted for in any race, much less a recent one) could the presence of the same candidate again jog their memory so they recall correctly?
Great thread. Thanks for the clarity. Well done. Also agree with comment about how aggregation hides the details, including important ones. I’m more convinced than ever that polling error will likely favor Harris. But I also recognize that I might me confusing my wishes with reality.
Your inane use of “subjective” adjective aside:
> that pollsters would prefer not to talk about.
Outside the few true scam artists, for those that have even a modicum of professional self esteem, you should count yourself lucky if you can get them to STFU about it. 🤣
There is a chart that calculates the position if the polls were as wrong as they were in 2020 and 2022; they show Trump up
substantially in the seven battlegrounds on the 2020 miss but trailing in all but Georgia on the 2022 miss.
Sorry, but that’s not the same thing; that’s just applying the “error” from each state to current results. Instead, I’d like to see the results if pollsters ran their current data through their older models.
No one is going to pay them to do that though along with their polling for the election, so it's wasted effort to even think about.
nyt siena did do this actually
Title of essay: Nate Silver: Here’s What My Gut Says About the Election. But Don’t Trust Anyone’s Gut, even mine.
How it is being reported in other headlines: Nate Silver says his instinct is that Trump will win.
That was written as a joke but after looking it up it doesnt go far enough
https://duckduckgo.com/?q=nate+silver&iar=news&ia=news
"Trump is going to win election, says America's top pollster" is a real headline.
I like Nate a lot but acting as if he isn’t setting the vibes through these posts the past two weeks just as much as the rest of the media is bananas.
I wish I know what people mean when they say "vibes."
Nate isn't a top pollster, because he IS NOT a pollster.
He's an analyst, and one of the best. Why else are you paying for this site?
That headline is from The Telegraph, which isn't exactly the most reputable news source. Also, FWIW, they recently changed the headline to say "analyst" instead of "pollster", so I suspect several people told them.
Nate got 2016 more right than others because he had access to inside information from the Dem campaign.
"One of the best". How?
Media doesn't understand nuance. In other news, the Pope shits in the Vatican.
Nate knows this happens. He joked about it in another post. He can't be responsible for how media writes headlines. It is funny, though. Nice post, detailing that.
FWIW (nothing, as Nate persuasively argues), my gut is Kamala, because every single data point I have *other* than the polls says its absolutely insane that Trump could win. I realize that this is part of being in the "bubble" of people who are aware of things like how Trump Chief of Staff John Kelly reported that Trump wished he could have "Hitler's generals" but was unfamiliar with who Rommel was.
"every single data point I have *other* than the polls says its absolutely insane that Trump could win"
Another data point: Democratic Senate candidates in PA, MI, WI are **currently** running campaign ads touting how they voted for some Trump bills when they were in Congress. Why would savvy, experienced Democratic politicians start advertising to their voters that they sided with Trump? Costs a lot of campaign funds to run those ads.
https://www.axios.com/2024/10/18/senate-democrats-campaign-ads-trump-2024
I mean, they are responding to the polls, so it's really the same data point.
I meant more things like:
1) enthusiasm I see around town (lots of yards in my neighborhood with signs for the R congressional candidate but conspicuously no Trump sign - which is different than 2016 or 2020)
2) social media enthusiasm levels (although this may in part reflect some self-segregating, as conservatives have adopted Twitter as their platform of choice and liberals have fled it -- but there is still a sizeable conservative faction on Reddit, and I'm just seeing a lot less pro-Trump stuff than I did in 2016 or 2020)
3) Trump's promises v. Trump's record -- it's one thing to believe the guy's rhetoric about helping the little guy in 2016 (which is why I very much thought he could win back then), but it's another thing entirely after he governed like a typical Republican, with a booster shot of chaos and insanity. But again, I recognize that I'm probably in a bubble that is aware of what Trump actually did instead of just having vague nostalgic vibes about the pre-covid times.
Social media isn't real life. The data points I see that favor Harris I'm interested in are things like
1. Number of small dollar donors.
2. Number of volunteers (not paid canvassers).
3. The fact that Trump underperformed his polls in the primaries and Biden didn't.
4. The sizeable overperformances of Democrats in special elections all year.
I'm not sure that these are actually meaningful, but Trump doesn't have any real data to counter that.
Early voting.
What about early voting? The data I've seen looks good for Harris in PA, MI, and WI and more mixed in the other swing states.
But it's hard to take that seriously because there's no point of comparison. There's no Covid to drive a ton of mail in voting like 2020, but both sides are encouraging their voters to vote early (unlike 2016).
Democrats typically hold an edge compared to Republicans in early voting.
1. Even absent Covid early voting is breaking records right now--largely driving by increased Republican utilization.
2. Contribution to the vote in GA for black voters is much lower than normal.
3. In Nevada Republicans haven't led in early voting since at least 2008. They're ahead now.
Party Identification is R+3 for the first time in a long time. If polls use D+3 to +7, they will miss bigly.
The trend is towards Trump in all polls and betting markets.
Republicans made massive gains in voter registration in key swing states, thanks in part to grassroots orgs like Turning Point, Scott Pressler, Chase PA, 100X, etc.
Early data coming out shows massive turnout in red counties, and those with propensity data show it is 0/4 or 1/4 Rs coming out, so they are not cannibalizing ED votes.
Miami-Dade is R+10 in early voting right now, insane.
Nevada is R+ right now as Jon Ralston admits.
Articles coming out daily on the panic setting in amongst Dems.
There was a significantly larger number of relatively high profile Republicans who endorsed the Democrat this time around, so it's hard to say how many of those +R advantages are actually Harris votes.
That poll isn't party registration, it's "Do you feel more like a Democrat or a Republican today?"
Why would somebody who's going to vote for Harris say "I'm a registered independent but I feel closer to the Republicans"?
Polls consistently show both Harris and Trump are getting 95% of each party's respective voters. There's not going to be a mythical large number of Rs voting for Harris. If so, it would be canceled out by the Ds voting for Trump (RFK, Gabbard supporters, etc.).
The sign thing is obviously anecdotal, but it’s hard not to notice. I’ve seen more Harris signs than I did Biden,l and Clinton—I saw almost no Clinton signs. I still see Trump signs, but not as many as 2020 or 2016. I saw a ton in 2016.
In the immortal words of Pauline Kael, after Nixon's 1972 landslide victory (49 states, only Massachusetts went for McGovern):
" I can't believe Nixon won, I don't know anyone who voted for him", spoken from Manhattan's Upper East Side.
I wouldn't call the words immortal when she never said them. Here's the actual quote: "I live in a rather special world. I only know one person who voted for Nixon. Where they are I don’t know. They’re outside my ken. But sometimes when I’m in a theater I can feel them."
She was perfectly aware that she lived in a vacumn chamber, not oblivious to it like the false quote implies.
The "How did candidate X win, I don't know anyone who voted for him" is real. There were definitely people who said that on 1980 in Berkeley about Reagan.
And I've seen similar lines about Trump 2020 used as "proof" that the election was rigged.
Right, but I’ve been three states in the last two months, and in a mix of conservative/liberal locals within those states.
A few months ago I was in farm country chatting with a farmer about how tough it was to slaughter his pet cow. That the kind of place you visited?
Trump helping the little guy, as in median household income adjusted for inflation (up 12.4% for Black working class people, up 8% for white working class during Trump years. Up 0.8% for Blacks under Biden/Harris, down 1.5% for whites).
Sounds like the little guy did quite well under Trump and got whacked under Biden/Harris
https://fred.stlouisfed.org/series/MEHOINUSA672N
You and I both know that Trump's policies didn't have anything to do with that -- which is probably why you didn't talk about any policies.
"You and I both know that Trump's policies didn't have anything to do with that "
Restricting illegal immigration to 1/6 of its current numbers which reduces the supply of low-wage workers (ever heard of the law of supply and demand setting prices and wages?).
Replacing NAFTA with new trade pact that has specific targets for US content in manufactures imported from Mexico.
Imposing tariffs and technology restrictions on China to prevent US trade from enriching our most powerful adversary
Inflation at less than 2% for the duration of the term
https://www.pewresearch.org/short-reads/2024/10/01/migrant-encounters-at-u-s-mexico-border-have-fallen-sharply-in-2024/
Those all sound like inflationary policies to me. But what do I know, I’m just an economist who works in finance.
Why does immigration today matter for setting wage levels in 2017 to 2019?
At the time, immigration into the US was an all time high.
Early voting.
It's still early (nyuk nyuk nyuk) but Republicans seem to be happier than Demkcrats with the early voting tabs.
At least in Texas, I've seen R campaign offices making yard signs available free for all on the ballot except Trump. You have to pay for one of those.
I understand this may not feel intuitive, but most people who identify as conservative but who are still on the fence about Trump, hate him as a person but like a few of his policies.
The goal for senate / house Dems running in swing states is to pull the few palatable Trump policies they support, while drawing contrast to between them and the far right policies supported by Trump and the Republican running against them.
That would clearly show this ad approach by Kasey in PA as smart and not at all in conflict with Kamala having just as good a chance to win PA as Trump.
This is the problem with the popular stat driven side of politics and the sometimes misguided out of touch understanding of how normal people think and vote. Most voters are complicated and have contradictory view points. You can’t just do one-size fits all approach to reaching the persuadable voter.
Also, if Kasey was so afraid of tying himself with Harris, he wouldn’t have stumped for her numerous times in the last few months. This is the second or third time I’ve seen someone use this as evidence of Dems giving up or acknowledging Trump will win and that’s just not the case.
So why has Baldwin declined to appear with Harris in WI?
Casey hasn't campaigned with Harris yet either. And then there's this:
https://www.axios.com/2024/10/18/senate-democrats-campaign-ads-trump-2024
The polls have been showing downballot Democrats running ahead of Harris. It points to the existence of a good number of Trump/Downballot D voters (or Undecided/Downballot D, edit - or Trump/Undecided). These voters are definitely worth targeting for downballot Democrats. It doesn't point towards an overperformance for Trump because that data is taken directly from the polls as they are.
Fwiw I'm dubious about the size of that gap. It doesn't make a lot of sense to me. I expect either the downballot Democrats will underperform, or Harris will overperform. But it's pretty clear what the polls are saying.
Or the pollsters are adjusting their numbers so they don't release outliers
"Why would savvy, experienced Democratic politicians start advertising to their voters that they sided with Trump?"
This is probably hopelessly naive. But one possible answer is, "because it's good for the country".
Partisanship in the US needs to go down. If we were to suppose that people high up in the Democratic party actually *do* care about democracy, then we should expect them to do that. People who are anti-Trump should recognise that America's problems don't all just suddenly go away in the event that he loses the election (and the loss sticks).
I’d be interested to see whether voters do switch parties. I believe I saw somewhere that switching parties between elections was quite a lot higher than I thought.
Because they're trying to pick up whatever "leans Republican but are undecided" voters are left.
And because they don't understand that undermining the Harris campaign by advertising they supported Trump policies in Congress could have repercussions for them if Harris wins?
The media made a big deal out of it, but I kind of doubt that their ads are going to make anyone say, “Oh, wow, I was totally going to vote for Harris, but this one guy said he could be bi-partisan, so now I'm going to vote for Trump.”
To get Republicans who are sick of Trump to vote for them, obviously.
Yeah, "absolutely insane that Trump could win" is not based on anything happening in the actual election. It just means, "I don't want it to happen! I don't want it! I don't want it!!!"
Serious question - Do you actually think an undecided swing voter 2 weeks before an election, who is likely apolitical by nature, knows or cares who John Kelly is? Or even cares about a Trump = Hitler comparison as if they haven’t heard it for the last 8 years?
Gotta remember that the people deciding this election are not political nerds paying for polling analysis from Nate Silver.
I don't think anyone knows. That's why said I recognize I'm in a bubble.
I think if they knew, some might care - this isn't someone else saying Trump is Hitler, this was Trump saying he wishes he *was* more like Hitler!
Thing is, anyone who this would have traction with is already in Harris' camp. The whole "Trump admires Hitler" thing has been reported/alleged before. The fact that CNN is hitting it so hard today I see as a sign of panic.
Snopes debunked such a claim back in 2019:
https://www.snopes.com/fact-check/trump-hitler-mein-kampf/
someone specifically stated that this happened on the record so the snopes "debunking" is irrelevant lol.
Ohhh so they stated it happened! Then it must have happened, because it's not like people lie or anything.
I didn’t say it definitely happened, but I’d pose the question to you of why everyone who worked with Trump has similar stories. Is the likelihood that all these senior military figures, who served our country for forty years with integrity, are all lying about Trump? All of his secretaries of defense, his chief of staff, his vice president, his attorneys general, his first Secretary of State. They’re all lying? Or Trump is? Sincerely what is more likely.
There are different kinds of swing voters. The moderate Republican pro-choice suburbanite, who has always voted Republican in the past, likely knows who John Kelly is. In any case, should anybody run into a report of that interview, they will be told who he is. But it is the former swing voter, who may even be a shy Harris voter, who would be affected by Kelly's comments.
I believe once there was a survey done of when Boris Johnson kept on saying: Get Brexit done. I think only 30% could recall it. The point being that most people don’t listen to politics and messages need to be really quick and easy to remember. I think Trump has done better here (not perfect) with the key points and why the stories and huge campaign slogans cut through. Unfortunately he needs to do them now.
100% this.
As another person with a meaningless gut feeling that Kamala will pull it out, I tend to agree.
The other big reason my gut says Harris is that pollsters have much more incentive to avoid underestimating Trump than they do to avoid underestimating Harris, which surely affects them consciously and subconsciously.
My gut says Kamala too largely based on what you said in your last paragraph. Nate said in his last sentence that it’s probably equally likely for a pro-Trump miss and a pro-Harris miss. I think with the weighting by recalled vote being relatively abundant that it’s less likely Trump support is underestimated a third time. It’s either pretty accurate or we are underestimating Kamala’s strength IMO
Saying you want "Hitler's generals" is not an endorsement of Hitler himself. Those officers tried to kill him. Instead, it reflects the high regard for German military leadership in many post-WW2 studies. Similarly, studies have shown German soldiers were objectively more efficient than counterparts in other countries.
Although Kelly said that Trump revealed in the conversation that he didn't know that Hitler's generals tried to kill him, and Trump didn't know who Rommel was. Occam's Razor here: the dude likes Nazis.
What does 'efficient' mean? What studies are you talking about?
Everything I've ever read indicates that the American military was superior to the German military in basically every way except for the fact that the Germans overengineered their vehicles (but even this is really another reason the American military was superior, because America made vehicles that could be easily mass-produced).
Sure. Shame about their leadership (generals, Hitler himself) pointlessly over-stretching them and then needlessly trapping a bunch of them in Stalingrad.
I mean, there are data points within the polls that also point in her favor. For example, the younger a voter is, the less likely they are to say they plan to vote early. Yet, those young voters *did* show up in 2020, and they *did* show up in the Blue Wall battleground states in 2022.
The media loves to harp on the fact that young voters didn't show up as expected in 2022, but they never focus on individual states. Young voters vote when the issues are important to them *and* when they think their vote can make a difference.
Combine that with the fact that none of the Blue Wall states restricted voting (not for a lack of trying, Pennsylvania) since 2020, and I fully expect young voters to show up in force on or before election day.
All of that is a valid analysis based on data. It would be absurd to try to use it to predict that Harris will win. However, it's equally absurd to ignore data like that and claim that Trump will win just because polls have had a tiny change and Republican early voting numbers are up! (I'm not accusing Nate of doing that; I'm speaking generally.)
I don't believe that. Everyone knows who Rommel was. I wish we DID have another Rommel for the USA!
Wait, that doesn't make me a fascist, does it??
If that was your point, you would have said it right after talking about the big Trump sign. Or after your first comment about the farmer. But you didn’t.
It’s genuinely fascinating to watch the counflip nature of this election open an X-ray into everyone’s soul.
Nate, professional hater and liberal contrarian, is ranting about Kamala’s weak campaign. The betting markets, mostly a certain type of dude, are going further and further toward Trump. Anti Trump Republican Mike Murphy keeps waffling back and forth. I spend half my time in despair and half my time optimistic.
God, please end this hell!
As someone who is basically completely divorced from the American political spectrum, it's becoming increasingly tempting to just move to Switzerland and bunker down for the next few decades.
Sounds good --- send us back some chocolate. Yodel when it's coming.
Can you get a 4-year Swiss tourist visa? 😁
Ah the "coin flip". I have been following Nate since 2008 and appreciate the scientific rigor. But my question is this. It seems no matter the analysis or commentary or what-if scenario ("take out RFK", "take out partisan polls"), you could replace the headline of every post to be "Race is a coin toss". If, in a polarized electorate, every election is going to be within a a few points, and if you're going to call any win probability between 40% and 60% a "coin toss" (seems reasonable to me) - then what is the point of all this polling and all this analysis?
For this content specifically, the point is entertainment value. Do you enjoy reading the thoughts of Nate Silver, who thinks about elections in a statistical way? And do you enjoy reading it enough to pay for it?
I sure do. And so do many others. And that's the only reason this exists.
Busted. I enjoy it. I pay for it.
I had kinda hoped for greater societal value too ha. Maybe Nate or Eli could tell us some time in a post.
The point of the polling at least, is that that's the only reason we *know* that it's a toss up. Elections aren't always close.
Imagine if voters had categorically rejected Harris after Biden dropped out - maybe she was tainted too much by association. It's easy to imagine that having happened, but the polls are reasons to think that that hasn't happened. It's easy to miss the importance of polls in this, because there's tonnes of vibes that will tell you Harris wasn't rejected. But a lot of those vibes are *informed* by the polls. Vibes tend to be given a lot more weight when they align with polls, and that's one of the few good things there is to say at them.
Or simpler, imagine if Biden hadn't dropped out.
Biden would then have gotten the endorsement of the Teamsters union.
Union voters are about 14% of the electorate in PA, MI, WI.
That's what I thought, too! Might have saved the Dems the election, having the unions in their pocket.
I’m not at all sure he would have but I’m pretty sure it wouldn’t have saved him regardless.
Yeah this ambivalence from blue collar union leaders seems more driven by an attempt to hedge bets than any political ties, and the teamsters in particular seem driven by internal politics. Many teamster bosses who voted in agreement with non endorsement then turned around and had their locals endorse. While they've thrown around a survey of members as justification, the panel of teamsters members that interviewed candidates chose harris.
Sean O'brien in particular seems to be trying to flirt with right wingers that actively despise union labor, in an effort that is likely to only lose political allies than gain any new ones.