How The Economist’s presidential forecast works | Hacker News Reader

How The Economist’s presidential forecast works

economist.com

86 pointstosh82 comments

It's great to see the Economist explain statistical concepts like parsimony, regularization, cross-validation, and MCMC methods in a way that is both accessible and completely accurate. I can't think of another mainstream publication that would bother to explain the mathematical techniques behind their model in such detail. Kudos to the Economist!

This is great statistics, but it avoids the problem that moving to online polling has made it very difficult to get representative populations, so the data itself is biased in ways that cannot be counteracted by methods (only by assumptions, priors, etc). Which makes this misleading because it gives the forecast air of confidence that is unjustified.

> makes this misleading because it gives the forecast air of confidence

The entire first half of the article is about why polling is not only unreliable at this point, but becoming increasingly unreliable in biased ways.

True, I suppose my disagreement is that I believe it doesn't go far enough to explain how big of a deal it is, and how there _aren't_ ways to deal with it without substantial, subjective intervention from the forecasters.

I've worked on weighting code for online polls, they literally rely on dozens of hand picked decisions to stay "reasonable". These decisions aren't factored into the error bars, making them appear smaller.

And as far as the fundamental style predictions, how can you use a single GDP number when Fox tells its viewers one number and MSNBC tells its viewers another?

This article does describe a faithful statistical effort, but to me it doesn't emphasize the risk of a "black swan" event enough.

Absolutely. But we’ve been past the “golden age” of polling using live callers on landlines for more than 20 years now. We now have a reasonable corpus of polling data that we can use to evaluate how good pollsters are at making the corrections (often educated guesses) that they use to adjust their polls.

The justification for Bayesian inference, that the posterior will eventually converge to the true distribution, breaks down unless your prior has support for a good approximation to the data-generating model. So without a good model of how polling results map to actual voter distributions, the Economist model is guesswork, to a large extent.

https://abcnews.go.com/538/538s-2024-presidential-election-f...

It is going to be very close and decided by key swing states , as has typically been the case. It will not be at all like 2008 or 2012 or 1996. This is why forecasting models are not that useful ,as the swing states are such big and unknown variables. It's like "I have 95% confidence that either candidate will get 48-52% of popular vote." great, real helpful .

Knowing that a candidate will get 48%-52% of popular vote is indeed more helpful than not knowing what’s going on at all.

Aleatory, epistemic and ontological uncertainties are three different things.

Honest question, what different decisions could you make with a 48-52 interval than a completely uninformed interval of 0-100?

You know where to spend your resources.

If you can move the needle in 5 states by a total of 25 points, you aren't going to waste them in a states where it's 60-40.

Agreed, and the subversion tactics tried before have gone through refinement and people put into key positions to make 2020 look like a garden party.

FWIW, Trump was given worse odds than Hilary much closer to the election using the 538 model (which is similar in principle).

Not a knock on these projections, but just a reminder that even if a lead is statistically significant, the size of the lead matters.

66% is hardly a fait accompli, especially five months out. That said, as Nate Silver has said (paraphrasing), 16% probability sounds unlikely until you're playing Russian Roulette.

At some point the current incarnation of the GOP will have power - with or without Trump, 2024 or later. It’s just sadly something that’s going to happen.

And they’ve literally told us everything they’re going to do. That makes watching this all the more painful.

Agreed. Every election it loses in the meantime will amp up resentment and aggravate the USA's political dysfunction. I think of it as a counterpart of Brexit, which had similar populist drivers, and was something the country just had to get out of its system, despite the steep attendant economic and social costs.

The brexit referendum was supposed to answer the question once and for all, and stop the tory party split. Instead we have a split down the middle, and the split is far further right than it was in the Major/Redwood days of the 90s.

8 years on we're seeing the tory party massively damaged, an outside chance of FPTP meaning they're neither the second largest party in seats or in votes. There's a crazy situation where half the tory voters are literally voting for a company run by Nigel Farage. I suspect 90% of the voters would struggle to name anyone else in that company (it's not a party the way the tories or greens are).

Brexit didn't solve the problem, there's always something else to blame.

One of the most pivotal moments in UK political history was Farage surviving the 2010 plane crash. There's just nobody else as large on that side of the aisle that can operate outside of the party structure (including Johnson)

Also FWIW the Keys To The White House guy who predicted Trump's win and most elections thinks Biden will win this time. https://www.theguardian.com/us-news/2024/apr/26/allan-lichtm...

It seems that they no longer publish an "election day forecast" regarding popular vote - something that they included in 2020: https://www.researchgate.net/profile/Rachael-Meager/publicat...

I gotta give the economist credit for acknowledging that Trump has a clear lead. https://www.economist.com/united-states/2024/06/12/five-mont...

Biden’s only realistic shot is to sweep Michigan, Wisconsin and Pennsylvania. Biden is polling 4-8 points behind where he was in June in those states in 2020, and where Clinton was in 2016.

As an American so many of my fellow citizens have been duped and will now be fleeced.

“Duped” how? Say what you want about Trump, but “hiding the ball” isn’t exactly his thing.

The man who promised a healthcare plan in "two weeks" for about four years straight?

Well, they had that for four years so they know what they are getting I guess.

I suppose they do deserve credit, but isn’t that degree of objectivity the very least we should expect from our news sources? I’d be shocked to learn that the economist falsified their forecast results.

>Biden is polling 4-8 points behind where he was in June in those states in 2020

Further, Trump consistently did better than the polls in almost every state in both 2016 and 2020.

To have him actually be up in the polls—something he never did nationally in the aggregate in 2016 or 2020—is signaling landslide.

>and where Clinton was in 2016.

I highly recommend this excellent Politico article on 10 key moments of the 2016 election campaign <http://www.politico.com/magazine/story/2016/12/2016-presiden...>. One is how Trump campaigned like crazy while Clinton took weeks off just loafing in Chappaqua.

Eight years later Trump is still campaigning like crazy while Biden is hiding in his basement. I was amazed to learn recently that Biden as president has never done an interview with a major newspaper <https://www.natesilver.net/p/its-time-for-the-white-house-to...>; only podcasts like Conan O'Brien's, where he knows he'll only get softball questions.

> To have him actually be up in the polls—something he never did nationally in the aggregate in 2016 or 2020—is signaling landslide.

That’s certainly possible, but another explanation is that the polls are trying to correct for this and have over corrected.

If they were making such changes I’d suspect Nate Silver or Nate Cohn would write about that.

I think the polls will still possibly be off, but I wouldn’t bet money that they will overestimate Trump’s support this time. I think this time Trump’s growing support among Hispanic, black, and young voters will mess up the demographic adjustment again (but in a different way than in 2016 and 2020).

Most of the polls have changed their methodologies: https://www.pewresearch.org/methods/2023/04/19/how-public-po...

It's possible they still have major problems, but they've had multiple elections to learn from their mistakes at this point.

The problem is they’re shooting and a moving target. Polls all perform various demographic adjustments. In 2016 they underestimated Trump’s support among white non-college educated folks. They fixed that for 2020, but still underestimated Trump’s support. Now, polls are showing a large drop in support for Biden among black, hispanic, and young voters, and a significant shift to Trump. A recent poll showed Biden’s support among black people in Michigan has dropped 20 points since 2020, with Trump picking up some those disaffected voters. Polls have also consistently shown Trump pulling even with young voters. Do the top-line results fully reflect those shifts?

I think various scenarios are possible. But I think the outcome that’s somewhat more likely than other possible outcomes is that pollsters underestimate the degree to which Trump-leaning minorities and young people will turn out, and will underestimate Trump’s support again by a few points.

It’s way too early for any sort of prediction.

This is a rematch election between two former presidents. Everyone already has hardened opinions of both men. And Biden hasn’t led Trump in the RCP poll average since September 2023. It’s like the mirror image of 2020. Biden led Trump the whole time from spring 2020 and although the polls overestimated Biden’s margin they called that it wasn’t really a horse race.

https://archive.ph/tnC0e

This kind of prediction is meaningless and will likely change drastically throughout the campaign. Instead of predicting who will win, the media should always be explaining the stakes to us.

I feel like it’s worse than meaningless. It distorts the coverage—newspapers start covering the election as if it were sport, and as the number goes up or down, it is treated as one side winning or losing, with substantive coverage falling by the wayside. Political actors act differently because of how they think it will make the number move, instead of what they think is best for the country. And it's so far out from the election that the number in its current state is not very meaningful—if anything it convinces people, incorrectly, that it is a "reality", when the only real measure is who will show up to vote on election day.

I've worked as a data scientist for political campaigns (but no longer). The elephant in the room is that there are so many forces that make this election different from historical ones, from polls moving online away from phones (making them much much less reliable) to Trump's conviction to a completely unprecedented/fractured media landscape. Even if these forecasts were accurate, their usefulness for anyone other than people spending ad money is absolutely zero. Now that these forecast have so so much bigger error bars, I think public forecasts are actually harmful and just a scummy money grab from the media that posts them.

EDIT:

Why are they harmful? At best they make lots of people angry about something that hasn't happened. At worst they skew the election by making people not vote, or afterward serve to justify election fraud complaints ("my candidate wasn't supposed to lose based on forecasts, so fraud occured").

I think we all need to take a deep breath, accept we have little understanding what will happen, _vote_, and hope for a sane outcome.

Totally agree. Journalists should be explaining the stakes, not trying to predict the outcome. Predicting the outcome of an election is mostly a fool’s game.

Pretty cool site reminiscent of the 538 analysis of elections past. I wonder how the model will react if Biden drops out before the election, though, since the blue polls would suddenly be "obsoleted" (wrong word, I know) with a new incumbent.

It is too late for Biden to bow out. Although the chance of either Biden or Trump dying before the election is uncomfortably high.

They're both supposed to be in good health though, we have Biden's public yearly checkup report. I don't think many old people get scans and full-body checkups every year and then die of something catastrophic with no warning?

My priors would go the opposite way: a full-body checkup cannot clear out the plaque in my arteries, lower my blood pressure, strength my vascular walls, and there's no miracle drugs that definitively address those. I think of a checkup that says "good to go, won't die in the next year!" is more like an auto checkup, where we do diagnostics then can take definitive action.

That being said saying odds are one of them dies before the election is uncomfortably high isn't something I'd subscribe to. Quick check here, using N=age-google-told-me and N+1, gets 1 in 18 and 1 in 13, 6% and 8%. If they were 60 it'd be 1%, 70, 2%.

https://www.candidmoney.com/calculators/death-probability-ca...

Actuarial tables tell you basically nothing about a specific individual. For that you would want to look at some personal. Like, say, a full body checkup.

Yeah thats a good point, I agree, a full body checkup tells us stuff about an individual, but it doesn't heal them (or at length: meaningfully rule out, or affect, odds of dying in the next year)

Apparently [1], Biden at least has "controlled" cholesterol, a normal EKG, some asymptomatic and controlled fibrillation. The number of experts who consult the guy every year is actually impressive.

The usual 70-year-old is probably a risky bet, but surely 10 specialists plus a team of generalists would make a more predictive yearly checkup. I think he's going to lose through normal politics and not by way of a heart attack.

[1] https://www.whitehouse.gov/wp-content/uploads/2024/02/Health...

I didn’t realise his checkup was public [1]

I thought there would be more around neuro in there. Interesting that both recent presidents are (purportedly) non drinkers.

[1] https://www.whitehouse.gov/wp-content/uploads/2024/02/Health...

Bush II was tee-total too wasn't he?

I wonder if he does CT/MRI/angiograms to find any vascular abnormalities. Seems to be a big cause of sudden death, but they don't talk about it. One CT for a 70-year-old head of state doesn't sound like too much?

Not an expert, but my experience is old people can decline very fast.

It’s literally the best possible outcome of old age: being energetic and in full capacity, declining quickly and passing on.

The alternatives are dying early, or a long period of dependency.

But yeah, that’s how old people die.

Still, their actuarial chance of dying within a year is >5%, given their age and sex.

Even with scans and checkups, at that age, a bad case of the flu can kill you, or start you on that path. And I don't think we have any reliable non-invasive ways of detecting heart attack risk, or aneurism risk, those kinds of things.

> heart attack risk

Not sure, but don't ECGs find cases of severe plaques or narrowed arteries? Or most cases of heart disease.

> aneurism risk

An angiography should find most cases, from some quick research. "The CTA sensitivity, specificity and accuracy was 80%, 43% and 73%, respectively. The CTA sensitivity for aneurysms < 3 mm and 3 mm – 5 mm in size was 30% and 81.5%, respectively (p = 0.024)." [1]

Agree on the rest though.

[1] https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10244944/

We haven't had an election like it in over 50 years, but in 1968 the democratic candidate was Hubert Humphrey, who entered the race too late to participate in any primaries.

Can't Biden bow out any time before the Democratic National Convention in mid-August? I don't expect that he will, but he could.

It would be political suicide for the Democrats not to go with a sitting incumbent President as their candidate, especially against a candidate who was formerly President.

Different states have different deadlines for getting on the ballot. The DNC is already going to have to a pre-convention virtual convention where they legally nominate Biden so he can get on the ballot in Ohio. The physical convention will be for rallying the party behind him. If they were to switch to someone else at the convention, at the very least Ohio would retain Biden on the ballot. Ohio does allow for 'faithless' electors, so Biden winning Ohio while the official nominee won other states might not be the worst thing provided the electors could be trusted to switch their vote to whoever the DNC put up in place of Biden.

When is the Ohio deadline?

The chance of them both dying is uncomfortably low.

Seriously, the best two candidates the entire USA has to offer are these washed up octagenarian politicians?

They are a symptom of a gridlocked and extortionist political system that serves the elite.

Since either of them winning would mean a term limited incumbent in 2028, my guess is that both parties will nominate a Millennial candidate for the next election, both messaging that they're the party of fresh ideas for the future.

The parties barely have control over who gets nominated, no one in the Republican Party thought Trump was a good idea in 2016 at the start of the primaries. It only matters who gets votes, and the Republican side skews old enough that a millennial would have problems.

Every year they get older, they have a higher chance of surviving until the next year. It’s a weird thing about aging. Of course until you die that is.

How do you figure? As I read the actuarial tables, I do not see any way of interpreting the data that makes your statement true.

I think what they're saying is an 82 year old is more likely to make it to 83 than an 81 year old is

I would bet large amounts of money that Biden doesn’t drop out before the election. It would be unprecedented and there’s no indication it would happen.

This sounds like a more thorough update to the 2020 model they published here:

https://github.com/TheEconomist/us-potus-model

Nate Silver was both the best and worst thing that happened to these predictions. The modus operandi of people in media (New York and Hollywood) is "gut checks" - Dean Baquet is only going to publish a model, no matter what the math says, that agrees with his "gut" that Clinton was going to win 2016. He was so crazy wrong but like, he's not getting fired. Laypeople and cranks conflate making a mistake that everyone in media made with a conspiracy between Democrats and the media, unable to grasp incompetence. Of course, if Dean Baquet had spent five minutes talking to a political scientist of some merit, asking, "is the voter turnout of traditional demographic groups like women and old people is really as predictive of elections as Nate Silver said it was?" He would have learned that Australia has had compulsory voting for decades and it is as conservative as ever.

Now that "gut checks" have disentangled themselves from New York graphics departments - can you believe its the graphics department, masquerading as a "data science" department, that runs election forecast stories?! - maybe forecasts will be better everywhere.

If I were doing this, I'd frame it as a Kaggle competition built on PyroPL. And I'd require top performance in democratic election forecasting everywhere, not just the US. The Economist is right, it's all about fundamentals, but which fundamentals, and can you prevent the choice of fundamentals from reflecting the publisher's political aspirations? Personally I'm skeptical that the Economist's choice of GDP growth makes any sense, and net approval rating sounds like a poll, so... there must be something better.

> Nate Silver was both the best and worst thing that happened to these predictions.

Can you elaborate on this opinion further? I have a hard time aligning the rest of your post with what looks like your thesis.

Now if the Economist wasn't a warren of conservative talking points how might it land?

I thought the economist was the left one and wsj was the right one, relatively speaking?

The economist is probably center right.

Money oriented right is never center.

Economist is British, so doesn’t really compare to WSJ. In the UK, it is center-center right, WSJ is for rich American conservatives compared to FoxNews, which is for poorer American conservatives (they don’t read papers, maybe tabloids, but that is basically what FoxNews is). NyTimes is the left side of WSJ, but isn’t so business focused (not so many liberal rich business people in NYC).

looks like trump will win :(

Honestly, if the US still wants Trump badly enough to vote him in, we deserve what we get. Fuck it. At least I'll enjoy watching this place burn.

Trump was already president for 4 years.

Yes, I'm very much aware.

You mist not be up on the plans for the second season. It’s not going to resemble his first term hardly at all.

> still wants Trump badly enough

IMHO it's less about "badly enough", I know very few people who are die-hard Trump supporters, but do know many centrist types who just aren't very enthusiastic with Biden and Harris in the past four years.

The Democrats should really pick someone more desirable for the majority of Americans. I know there are decent options, even if not the level of exciting as Obama was in 2008.

It's similar to 2016, and many people could not understand that much of the Trump vote was basically just not thrilled about Hillary Clinton, and was hardly a rally behind Trump.

I mean, it looked a lot like Hillary would win, too. 538 was an outlier in only giving her a 2/3 chance of winning when everyone else was saying 90%+.

Reply on news.ycombinator.com