Hacker News Reader

Comment by tyleo | Hacker News Reader

It’s interesting but also reminds me of US presidential predictors. All the models that guess right come out saying they have the magic formula but are often refuted by future elections.

This model needs a few more popes under its belt to build confidence in it.

View on HN

The authors do not suggest using this model for prediction. FTA: “We do not claim to predict the outcome of the Conclave,” Soda points out. “As the great statistician George Box said: ‘All models are wrong, but some are useful.’ Ours is intended to be a tool for reading the context, not an oracle.”

The best ones would show how they mopped up in the prediction markets with their brilliant model.

This is not different from any exercise in seeing the future: some people will get it right, and they will be hailed as having some secret sauce when in fact it is mostly based on luck.

I do appreciate that, on HN at least, the headline writer had the decency to honestly say "guessed" instead of a more certain-sounding word like "predicted."

sociology has same weird data showing even the weather can change for who people vote [0] https://www.sciencedirect.com/science/article/abs/pii/S09626...

and this is only an example of the multiple and more extensive studies made on this field

the moment people mock data from AI crawlers, this technology is useless?

Every prediction, once divulged, affects its own outcome. Some prophecies are self fulfilling while others are self preventing (and, therefore, “antiphecies”)

If it's so good it could at least predict more past popes already.

Related recent podcast interview from the Star Spangled Gamblers: https://starspangledgamblers.libsyn.com/the-trader-who-guess...

> In 2016, Blitz (@blizzythegoat24) bet on Donald Trump to win the general election.

> In 2020, Blitz not only bet on Biden to win the election, he guessed the outcome of every state correctly.

> In 2024, he managed to do the same. He bet on Trump to win the election and guessed every state correctly.

There are people who've won the lottery more than once, too, out of pure chance. It doesn't necessarily mean there's skill involved.

Also most states are foregone conclusions so there's really only a handful in doubt.

You have a 12.5% chance to correctly predict 3 coin flips in a row. You need a lot more than 3 examples before we can actually be confident that a successful prediction is anything more than luck.

To be fair "guessed every state" carries significantly more bits of information than 3 coin tosses.

What are the chance of guessing 50 coin flips in a row?

What are the odds of doing it again four years later?

This question shows a fundamental misunderstanding of predicting presidential elections. There is no value in predicting New York would go blue or Alabama would go red. Most states are easy to predict and the states that are more difficult are correlated.

Especially in an election like 2024 when all the swing states went the same way, getting them all right isn't much of an accomplishment. These weren't 51 independent coin flip predictions. There was one prediction, Donald Trump would do slightly better than polling indicated (or really more accurately Harris would do worse), and the results all flowed logically downstream of that.

Your hubris would be justified had every media outlet and pundit been talking ad nauseam about how Trump was absolutely going to sweep the field, but they didn't. Harris losing was a shock to many, many informed people.

Claiming "guessing every state right isn't much of an accomplishment" is a joke and absolutely, categorically not true. Otherwise I hope you made a lot of money on the prediction markets!

I knew the minute I saw Obama pull frozen Biden off a podium at some random event in early '24. It was no surprise he bowed out later, and it was downhill from there doubling down on the 2016 mistakes. Oh well.

Once again, this is not an accurate description of presidential elections or even what I said for that matter. I did not say that everyone knew Trump would win in the manner in which he did. But once we take Trump winning as a prediction, the results of the individual states mostly fall in line.

For example, a standard path to victory for Harris was to win Wisconsin, Michigan, and Pennsylvania. But those states are pretty similar to each other and their results are correlated. If you think Trump is going to win, you could have just awarded him one of those states, but that isn't actually a likely outcome due to their similarity. Whatever hypothetical reason that would explain Trump being underestimated in Michigan likely also applies to Wisconsin. The reverse is also true, if Harris won one of those, it was likely she would win the other. Therefore predicting Trump winning both Wisconsin and Michigan is not nearly as bold of a prediction as winning one and not the other. Considering the way that every swing state went to Trump makes it very easy to predict all 50 states, because the only actual prediction you need to reach that conclusion is that Trump will outperform his polling (or like I said before, Harris to underperform).

It seems like ABC has wiped the full 538 predictions from the web, but look at this chart of their predictions[1]. They have Trump winning by roughly his margin of victory as the most likely of any individual outcome. It doesn't mean he was favored, it is just the way all these state results are correlated to each other. If Trump were to win, the manner in which he did is not surprising.

[1] - https://i.abcnewsfe.com/a/ff20d77c-0734-48d0-b9b3-72eb9bec8a...

> Your hubris would be justified had every media outlet and pundit been talking ad nauseam about how Trump was absolutely going to sweep the field, but they didn't. Harris losing was a shock to many, many informed people.

My own sense is a fair number of people were predicting Harris to win because that's what they wanted to be true – people on the centre-left who said she was in trouble would be attacked for dragging their own side down rather than taken seriously – and other people saw that and didn't want to be attacked so they kept quiet. And it was a "shock to many, many informed people" because journalists and pundits were subject to the same phenomenon. Polling averages showed a race which was either 50-50 or with Trump slightly ahead, yet many Democrats were confident Harris would win.

I don't think it was always like this. I think if someone on the centre-left said in 2008 or 2012 that they thought Obama was in trouble, people may well have disagreed (correctly, as it turned out), but their opinion would not have received the same strong negative emotional reaction.

Most of them aren't coinflips, and all of their outcomes are strongly dependent, and you don't have to call it until the coin is about to land.

It does boil down to predicting a coinflip of 'Which party will outperform the polls (and by roughly how much).'

And getting three of those in a row is, while not trivial, does not in itself a dynasty make.

They're not fair coins, so the odds are pretty good.

The presidential election is 50 independent events, 20% of which are moderately variable. It’s impressive, especially 2024.

>The presidential election is 50 independent events

This is just fundamentally not true. See my other comment, the results are heavily correlated making prediction of them all easier, especially in 2024 when they all moved in the same direction.

So was it an easy prediction or was it luck? Anyone can say that about any prediction, if not these then what kind of prediction would pass the bar for both?

Lucky and easy are not mutually exclusive. See my other comment below about the 538 predictions[1]. We basically got the most likely individual outcome (making it easy), which itself wasn't that likely because there were so many possible outcomes (making it lucky).

I just don't think the 2024 election was a particularly good one to demonstrate predictive skill due to the results all being relatively uniformly off expectations. You would need to accurately predict different states swinging in different directions to show an actual interesting insight. But that wasn't possible this election because Trump mostly just beat his poll numbers by a relatively consistent amount across the board, not giving any opportunity to hit multiple different uncorrelated outcomes.

[1] - https://i.abcnewsfe.com/a/ff20d77c-0734-48d0-b9b3-72eb9bec8a...

From that podcast, it sounds like his predictions were very reliant on his judgement calls: "you could sense that people really wanted a change", "if you could call it a formula, sure", etc.

Clearly it's worked for him, but I'm not sure you can call that a model.

My understanding is that „model“ in that field generally means things as simple as an Excel table with some formulas.

Most likely the pope will be replaced by AI before there is a new election.

No need to predict.

That is a prediction.

This is a pretty wild thing to say. Pope will probably never - or at least for a very long time - be replaced by AI, simply because he's a symbol first and foremost. Actual decisions may be hypothetically made by AI, but you still need the pope to announce them.

It’s a common theme in science fiction that an AI becomes something divine, because of the appeal to authority. The AI knows best. Who are you to question the AI? The AI wants the best for us.

But all the fun is of course in that turning out not to be true :-)

I actually thought people would understand that this was meant as joke…

Reply on news.ycombinator.com