An accurate model can output a 50-50 prediction. Sure, no problem there. But there is a human bias that does tend to make 50% more likely in these cases. It is the maximum percentage you can assign to the uncomfortable possibility without it being higher than the comfortable possibility.<p>538 systematically magnified this kind of bias when they decided to rate polls, not based on their absolute error, but based on how close their bias was relative to other polls' biases.(<a href="https://x.com/andrei__roman/status/1854328028480115144" rel="nofollow">https://x.com/andrei__roman/status/1854328028480115144</a>) This down-weighted pollsters like Atlas Intel who would've otherwise improved 538's forecast.
Agreed with the points in OP. Though we did have the story that came out shortly after the election that apparently internal polling for the Harris campaign never showed her ahead [1].<p>Obviously it says in the article that they did eventually fight it to a dead heat, which is in-line with a 50-50 forecast, but I do wonder what, if anything, failed such that this key detail was never reported on publicly until after the election.<p>As the article notes, public polls started appearing in late September showing Harris ahead, which they never saw internally. Are internal polls just that much better than public ones? Is the media just incentivized to report on the most interesting outcome (a dead heat)?<p>[1] <a href="https://www.usatoday.com/story/news/politics/elections/2024/11/27/kamala-harris-advisers-internal-polling/76626278007/" rel="nofollow">https://www.usatoday.com/story/news/politics/elections/2024/...</a>
There's a pretty clear test of whether a pollster is about reporting or influencing: the partisanship of their errors. Neutral pollsters have differences with the results randomly distributed across parties. Propagandists have errors that favor their patrons. This essay leans on the magnitude of the errors, but that's less probative than their distribution.<p>Does any poll aggregator order by randomness of polling error?
Suppose you want to know whether it will rain on your wedding day. Without any forecast, there are two scenarios you need to prepare for: it will rain, or it won't.<p>A 50-50 forecast doesn't change anything. You still need to prepare for both scenarios. There's a sense in which it's "useless" even though it's a valid forecast.<p>Forecasts are most useful when they allow you to eliminate a possibility, so you don't need to prepare for it. This means even something like an 80% chance might be of limited use, if you still need to prepare for the 20% case.<p>(Gambling and investing are different, because you can accept losses and make up for them with repeated bets. You still need to be prepared for losses, but that's a matter of not betting too much compared to your bankroll.)<p>Even if the forecasters don't do anything wrong, for many people, presidential election forecasts are of limited use. Either they don't have anything they need to do to prepare, or they still have to prepare for either candidate to win.
Highly recommend this video by NNT about why prediction markets tend towards 50-50.<p><a href="https://www.youtube.com/watch?v=YRvPF__du9w" rel="nofollow">https://www.youtube.com/watch?v=YRvPF__du9w</a><p>Prediction markets are usually implemented as binary options. Like vanilla options, their price depends not just on the most likely outcome, but the whole distribution. When uncertainty increases (imagine squishing a mound of clay), you end up pushing lots of probability mass (clay) to the other side of the bet and the expectation of the payoff (used to make a price) tends towards 1/2.
Seems like there is a hidden function when 50-50 polls exist: 100% uncertainty with 100% certainty. This is the equivalent to having no data at all.<p>Maybe we should just start saying "This is the equivalent scenario/situation to having no data at all." instead of 50-50.
A close race, creates more excitement and more cash for the media.<p>If you get on a horses and your horse keep losing and losing ground,
watching the race becomes quite boring.
If the horses are neck on neck you will be captivated watching it
till the very end.<p>The media has a strong incentive to sell the elections as close to
keep more people seeing their ads and buying subscriptions.
The odds on Polymarket (and other betting markets) started diverging at the start of October; by the end of the month, the odds were 65/35 in favour of Trump.<p>At the time, many commentators were arguing that this was obvious market manipulation. It turned out that the trader who had supposedly manipulated the market had in fact commissioned private polls using an alternative methodology - rather than asking people how they would vote, it asked how they expected their <i>neighbours</i> to vote.<p><a href="https://polymarket.com/event/presidential-election-winner-2024" rel="nofollow">https://polymarket.com/event/presidential-election-winner-20...</a><p><a href="https://www.ft.com/content/e8b2ff85-a4dd-40cd-a4c1-e3d70686917f" rel="nofollow">https://www.ft.com/content/e8b2ff85-a4dd-40cd-a4c1-e3d706869...</a><p><a href="https://www.ft.com/content/4b302ab8-7e40-4d1a-bbe6-3d46f6811d85" rel="nofollow">https://www.ft.com/content/4b302ab8-7e40-4d1a-bbe6-3d46f6811...</a>
There is the statisticians aphorism often attributed to George Box that "All models are wrong but some are useful". This goes against the core argument of this essay that pretty much says that "The models were mostly right, but not useful" which not a useful argument itself, even if its right
I always get annoyed when people look at election outcomes and say “the polls were wrong” when the most likely outcome didn’t happen. It’d be like saying there’s a 5/6 chance that a 6-sided dice will not roll a 4, rolling a 4, and then concluding that the initial proposed probability was wrong.
I don't know about the framing of "giving up." But I think anyone who's been following election models since the original 538 in 2008 has probably gotten the feeling that they have less alpha in them than they did back then. I think there's some obvious reasons for this that the forecasters would probably agree with.<p>The biggest one seems to be a case of Goodhart's Law, leading to herding. Pollsters care a lot now about what their rating is in forecasting models, so they're reluctant to publish outlier results, those outlier results are very valuable for the models but are likely to get a pollster punished in the ratings next cycle.<p>Lots of changes to polling methods have been made due to polls underestimating Trump. Polls have become like mini models unto themselves. Due to their inability to poll a representative slice of the population they try to correct by adjusting their results to compensate for the difference between who they've polled and the likely makeup of the electorate. This makes sense in theory, but of course introduces a whole bunch of new variables that need to be tuned correctly.<p>On top of all this is the fact that the process is very high stakes and emotional with pollsters and modellers alike bringing their own political biases and only being able to resist pressure from political factions so much.<p>The analogy I kept coming back to watching election models during this last cycle was that it looked like an ML model that didn't have the data it needed to make good predictions and so was making the safest prediction it could make given what it did have. Basically getting stuck in this local minima at 50-50 that was least likely to be off by a lot.
I think polling these days is all but useless just because a very small proportion of people actually respond to pollsters and it's next to impossible to statistically account for "what would the other 80% say if they did't tell the pollster to fuck off". There's no way to build a "representative sample" to account for that. People are way more likely to respond to a "friendly" pollster and if they don't know whether pollster is "friendly" (works for the organisation aligned with their party), they assume they are from the other camp. It's a mess.
Hyperaccurate voter profiling and predictions is a tool of oppression in democracy.<p>It highlights precise means for the rich to use targeted propaganda to maximum effect for wedge issues and electioneering to serve their ends, both in forming the winning coalition, and it neutering the opposition.
Nate Silver also had an article saying that the most likely scenario would be a blowout. There was a 25% chance that Trump would win all 7 battleground state, a 15% chance Harris would win all 7. No other permutation of the 7 states was anywhere close.
I’m gonna take this opportunity to clarify a comment I wrote earlier this week: Sure, polling may be difficult to interpret when uncertainty is high and margins are small, but there are many elections every year that are easily predictable blowouts, but which are often not portrayed that way depending on your information diet.<p>Like, I can’t count the number of times someone will show me an awesome political ad for an awesome person who’s going up against an awful politician I hate, and there’ll be all this hope and the person showing me the ad will be thinking this person has a chance of winning. And then I’ll look at the polling and realize they’re down by 20 points in a state Trump carried by 30 and the error bars are only 3 points in each direction.<p>So even if polling close races can get messy, polling overall provides an extremely powerful antidote to a lot of propaganda and echo-chamber thinking.
It is insufficient to make a prediction like 50-50 in a US election unless you also can explain turnout. Anyone can throw out numbers like "51-49" or "55-45" but how did you get to those numbers? How are different demographics voting? How is there turnout changing? I've seen people laud their own accuracy in 2020 while being off by about 30 million in predicted turnout.<p>The way different demographics vote in US elections doesn't change much from election to election. What does change is turnout and turnout is a function of many things: enthusiasm for the candidate, voter suppression, ease of access to voting and so on.<p>2020 was unprecedented because of the pandemic. We greatly expanded early voting and mail-in ballots, which greatly increased participation.<p>A perfect example of this is Arizona. In 2020, Native Americans were crucial to flipping the state to Biden. Arizona state lawmakers responded to this by essentially punishing them and making it way more difficult to vote. Voter ID requirements, birth certificates and even having a physical address are all impediments to people who were born on and/or reside on reservations. There were fewer voting places and voting options. A rural voting place might randomly close early too after being an hours long drive.<p>Some looked at this and said Native Americans in Arizona swung hard to Trump. No, they were simply largely prevented from voting such that the only Native Americans who could reliably vote were more affluent and thus more likely to be Trump voters.<p>My point is: what polling model captured this prior to the 2024 election? I guarantee you it's none.<p>What really happened in 2024 was:<p>1. Biden voters swung to the couch in the millions;<p>2. Trump basically didn't lose white women, despite the abortion issue; and<p>3. Trump activated a previously low-propensity voter demographic: angry, young, terminally online white males, basically the Andrew Tate and 4chan crowd.<p>Any model has a difficulty with low-propensity demographics. Did any model capture this? I think it only started to become apparent with early voting exit polls.<p>I don't think the 2024 polls were particularly accurate. I do think they threw their hands up and simply converged to 50-50. Small differences in turnout predictions for different demographics can massively impact the result.
My pet conspiracy theory is that the monetary influence in US politics is a deciding factor in keeping it close to 50/50 as this allows for the biggest footballification of politics and thus allows politicans/donors to extract the most from people who vote against their interest just to see the other side lose.<p>The only thing I am unsure about is whether this needs coordination (a conspiracy) or is just a systemic dynamic that emerges from the rules of the system.
The whole question of using probability here is philosophically fraught. If you are saying "this has a probability of 50%" then you're saying "this will happen half of the time if you repeat the test". I don't think Harris would win 5 times if you ran that election 10 more times. From that regard, the 50% guess is quite inaccurate, and does just seem to be the pollsters "giving up" in the sense that they're pretty clueless and so ended up going with an even split. This is speculation on my part, but I think that election was decided in Trump's favor pretty decisively long before the publishing date of the last 50/50 poll.<p>Edit: to elucidate, suppose it rains on 50% of days. One forecaster gives a 50% chance of rain every day. Another gives a 90% chance of rain/not rain and is "wrong" 10% of of time. Both stations are giving you accurate information from a probability perspective (when station A says a 50% chance, you know there's a 50% chance of rain tomorrow, when station B says 90%, you know there's a 90% chance of rain tomorrow) but the 50% chance station is less useful as the number is lower. They have effectively given up in the same way pollsters have given up; they're saying there is no information that could give them a higher probability. The reason a single event can have many different probabilities is because probability is about repeated events. Both stations are predictions accurately reflect the distribution of rain/not rain days.
I guess the article says that election polling in the US
given the current landscape, or rather the landscape
during the past election, is useless.<p>Today Biden is up 2%, please remember that the polls have a
+-3% uncertainty built in, so we have nothing to report
>Trump and Harris are both a normal polling error away from a blowout<p>If this is the conclusion from all the data gathering and analysis then what's even the point? If the conclusion is: it could be a blowout either way, we have no idea. Seems like a waste of a lot of time and resources, for what gain exactly?
It seems to be a matter of indifference to many that a persistent societal division of fifty-fifty is symptomatic of a profound societal schism. This is because such a division, in any configuration, maximizes the number of those who are discontented. As it is written, "A house divided against itself cannot stand." To my mind, this matter is of far greater significance than any secondary, tangible processes that may be occurring in consequence.<p>Furthermore, the mechanisms that ensure the mathematical expectation of electoral outcomes to hover around a fifty-fifty split — a phenomenon observable in many nations — are fundamentally economic in nature. Both factions commit resources to the electoral contest to secure a mere one percent advantage, as such is the foundational principle of democracy: a majority of fifty-one percent prevails.<p>Thus, economic factors — for an electoral campaign is, in essence, a contest of capital—having, in effect, subverted the very system of democratic elections, inevitably lead to the decay of nations that religiously adhere to the mathematics of a single percentage point as the sole criterion of legitimacy. In optimizing for democratic representation, social stability and equilibrium have been forfeited.<p>It is akin to the psychological paradox: "I am correct, and all acknowledge it, yet why do I not experience contentment?" It is because one has optimized for correctness — or in the context of elections, for fairness and representativeness — rather than for overall well-being. Such is the predicament inherent in the pursuit of a mere fifty-one percent majority.
Rambling about why a 50-50 forecast on a blowout election is "fine" and actually, like, totally cool and actually you're the dumb one for thinking it isn't, holds no water in my book. Election forecasters are not scientists discovering the nature of the universe, where it might be ok to say "yeah there's nothing to discover here" / "we don't know what we don't know". They're businesspeople selling a product. If AWS says "well, actually, hosting servers is just really really hard and its totally normal that 10% of your requests are failing!" they'd go out of business, full stop. No one is going to buy your product.<p>Trump did not win by a little bit. They missed something about the 2016 election, and it is <i>wild</i> that they missed it again EIGHT years later. Don't let them hide behind the veil of "we said it was 50-50!"
For each of the 2024 7 swing states, the winner was <1% ahead on average, so what good are these polls if the results are going to be within their margin of error?<p>They need to either find a more accurate way, or... give up!
At the end of the day it was a 312 vs. 226 in the Electoral College. Seems a bit odd that this is supposed to be impossible to be predicted with any useful amount of certaintly. But perhaps that says more about the nature of the Electoral College than it says about pollsters.<p><a href="https://en.wikipedia.org/wiki/Efforts_to_reform_the_United_States_Electoral_College" rel="nofollow">https://en.wikipedia.org/wiki/Efforts_to_reform_the_United_S...</a>