In the Network of the Conclave | Hacker News Reader

In the Network of the Conclave

unibocconi.it

103 pointstaubek74 comments

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.

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

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

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?

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."

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…

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.”

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

Was this published before the Pope was elected?

The article byline indicates 08 May 2025 but response header shows Last-Modified: Fri, 09 May 2025 13:39:02 GMT and the earliest entry in the Internet Archive is Fri, 09 May 2025 12:28:01 GMT.

The white smoke emerged from the Vatican Thu, 08 May 2025 16:07 GMT and Pope Leo XIV was announced shortly thereafter.

Interesting observation!

One explanation:

There’s a “Research” heading at the bottom that links to an article from today: “The Long Hand of Brussels on U.S. Businesses”, 09 May 2025 by Barbara Orlando.

Maybe they have a static site generator or even dynamic with caching that piled this in?

And even if. Just publish multiple versions and keep the one that is right.

Linkedin post: https://www.linkedin.com/posts/universita-bocconi_a-new-way-...

Posted at Thu, 08 May 2025 08:24:51 GMT according to https://trevorfox.com/linkedin-post-date-extractor.html

Thanks! Also, according to that LinkedIn post date extractor, this post by first author Giuseppe Soda was made on Thu, 08 May 2025 06:22:28 GMT:

https://www.linkedin.com/posts/giuseppe-beppe-soda-414749b0_...

Prevost was at the top of the "status" list. They also published guesses by "information control" and "coalition building".

Totally. I'm not super impressed.

Thank you! That answers my question definitively.

It’s notable, of course, that Robert Prevost was highlighted in this post but left off of many other lists of papal candidates.

We’re only reading this particular list because Robert Prevost was at the top of it!

Good question!

Leonardo Rizzo, one of the researchers, claimed on X.com that they published before the Pope was elected.

An X user commented:

> “Guessed” after the fact. Interesting nonetheless and worth sharing before the event next time!

Rizzo replied:

> Thanks a lot! We shared it the 8th morning on linkedin, the university website and few other sources (italian press). Next time I’ll also share it on X

https://x.com/LnrdRizzo/status/1920841806096343409

https://www.linkedin.com/posts/universita-bocconi_a-new-way-...

It is not that unlikely that some model predicts the right outcome. The only question is: can it be reproduced?

Forcing the conditions to allow this to be reproduced quickly would probably be unethical.

I know (of) a guy...

Why is "guessed" in quotes in the HN headline. That word does not appear in the article. They even say the following:

>The Bocconi team is the first to point out the limitations of the model. “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.”

Trying to take a victory lap on something like this seems to fly in the face of the statistical thinking that goes into creating a model like this.

Also the explicit survivorship bias... this would not be near the front page if their predictions were all wrong.

Polymarket had this guy at something like 1%. Enjoy your winnings, modellers.

I took that “guessed” from their twitter post at https://x.com/lnrdrizzo/status/1920783054181728701?s=46

Fair enough, but I think that makes this a lesson in the importance of context.

When someone uses quotes in their own informal original writing, they will often be received as scare quotes[1]. Knowing nothing about that author, I would assume he is using the word with some detachment. He knows the analysis wasn't trying to guess the pope, but he is having fun with the fact that the analysis pointed in the right direction.

When someone uses quotes to summarize something someone else wrote or said, especially when it is in a more formal context like a headline, it generally comes across as a direct quote. The headline therefore implies that the goal of this exercise was to predict the pope, which the article directly refutes.

The quote in the context of the headline wasn't "guessed" it was "How we 'guessed' the Pope using network science".

[1] - https://en.wikipedia.org/wiki/Scare_quotes

The real lesson here is that you can’t please everyone.

Quote it literally, some people say you missed the context.

Edit the quote, some people say you editorialized away the true meaning.

Summarize the situation yourself, some people say you took away the essence when there was a perfectly good quote available.

Which is part of the reason that the HN guidance is to use the original headline, which OP didn't do here.

I would add, this wasn't really a prediction anyway, they listed 14 total people across three separate top 5 lists. That's a pretty sizeable chunk of the total people who had a chance of being elected pope and they didn't try to make a unified list or assign percentages of getting elected.

I have tried to estimate when the Pope will be elected with a bayesian model, but it's failed predicting that the Pope will be elected at 7th ballot.

Proof: https://old.reddit.com/r/Catholicism/comments/1kgst9c/concla...

Maybe I can make a blog post, just for the sake of whom that are curious

Why would a bayesian model be better than, say, astrology in pricing that?

Bayesian models are working fine when you have few datas, and I have used the last conclaves ballots number to "train" that. It was mostly a toy project to understand better how a Bayesian model is working.

You can build a model, but is there any reason to assume it can predict what you want it to predict?

A very rudimentary process for a conclave election can be modeled by a geometric distribution. So using a bayesian model to estimate that can be appropriate. Of course, if you want a better estimation you need a more complex model, but I repeat, for a toy project it would be interesting.

Anyway, at least the real election at 4th ballot was inside the confidence interval predicted by my model.

Their model had 15 slots spread across three lists, with Prevost appearing on one list in the top spot (and not in the other two lists at all). I am not sure we can conclude a ton about their predictive power.

Look at the graph at the end. Prevost is the largest circle.

Shoutout to the Pope Crave (@ClubConcrave) account on Twitter/X. They somehow went from a fandom account posting yaoi/BL content for the movie Conclave to an actual journalistic outfit who posted the results of the actual conclave before mainstream news outlets did

Did they post it before the cardinals came out onto the balcony to make the announcement? If so, that’s impressive. If not, I assume they were just watching the livestream like everyone else.

For anyone interested in more serious journalism about the Church, I can't recommend The Pillar enough. The guys who run it have a lot of connections and know how the Church works inside and out. They had identified Prevost as being one of the top contenders in the days leading up to the conclave [1].

In their final "gut check" analysis, one of the two editors said his gut was going with Prevost. (Unfortunately behind a paywall, but here for reference [2].)

[1]: https://www.pillarcatholic.com/p/why-prevosts-papal-prospect...

[2]: https://www.pillarcatholic.com/p/conclave-day-1-head-check-h...

This article lists likely candidates for three categories and only one even contains Prevost.

I think the value lies in the mapping of heterogenous sources to one network that can be queried. How you arrive at conclusions derived from the information is human interpretation.

I’d be interested in how the facts were validated against the sources. Like debugging the mapping process.

> Informal relationships: mapped through authoritative journalistic sources, these include ideological affinities, mentoring relationships, and membership in patronage networks.

So a key part of this is impressionistic stuff: labels like "soft conservative," "liberal," and so on. Doesn't sound very rigorous.

It doesn’t need to be for this type of analysis. Given the information available, you’re looking for ways to bend the odds, not necessarily “know” with precision. Facebook makes billions with similar techniques.

This is textbook survivorship bias. Out of 133 electable cardinals, someone was bound to guess Robert Prevost. If they were wrong, no one would remember. You could probably find 132 others who guessed wrong.

This is particularly impressive because polymarket failed harder than I can remember it ever doing at predicting the Pope https://www.reddit.com/r/slatestarcodex/s/PRqb1nBVhA

The person who was elected Pope is meaningless if he does not clean house of sex abusing priests - past and present - as well as the cardinals and bishops who covered it up.

Should have hit the Polymarket. tbh good predictive models give you money nowadays and money lets you do more science. So if you have good model, you should use it.

I wonder if predictions of various models are spread more or less evenly across the candidates. Like one out of then knows the last digit of pi.

All they said about Prevost is that he had the highest status, which is just reporting a fact.

It will be interesting to add Pope Francis to this graph OR study similar graph as of 2013.

It would be interesting to backtest this to see if it can predict previous popes.

Reply on news.ycombinator.com