As I read the [write up](https://projects.economist.com/us-2020-forecast/president/how-this-works) of Gelman and Heidemanns' model for The Economist their "fundamentals" model is likely overfit. Nate Silver makes the [general point](https://twitter.com/NateSilver538/status/1284593671502012418) and it's worth reading the entire [thread](https://twitter.com/NateSilver538/status/1284565416036245507), but Silver doesn't directly address how Gelman and Heidemanns train the model. Specifically, they claim to avoid overfitting by tuning the regularization param...

@Anthony you cannot change the question after it was supposed to resolve. Seems like it should resolve ambiguously because there was not "on or about the close/resolve date of May 1, 2021" a council. I also think if you had such a council it would still resolve ambiguously anyways given a lack of unanimity.

I've decided that reporting on Afghanistan right now seems a little bit like coloring counties by whether they voted for the Democratic candidate or the Republican candidate. If you look at a normal map it looks like practically the entire country voted Republican. But if you understand the political geography you know that the population centers that cover only a small portion of the map vote overwhelmingly Democratic. See for example this [visual](https://www.fastcompany.com/90572489/u-s-election-maps-are-wildly-misleading-so-this-designer-fixed-them)....

This question is a great example of why we should have a little bit more time between the close of a questions and the resolution. I don't think I did very good job forecasting this question so much as updating over time, particularly at the end. Yes, updating is part of forecasting, but should half of the task really be will he resign/be ousted in the next day?

I read up on this and it's fascinating. When I think of a stablecoin backed by cryptocurrency I think of a guarantee that at any given point 1 stablecoin can be exchanged with the backer for X cryptocurrency where X cryptocurrency = $1 dollar at that point in time. MakerDAO does not actually make any such central guarantee! So if you buy DAI on the open market you actually have no right to redeem it for ethereum. How then do they maintain a peg? Well you cannot actually directly buy DAI from the central org MakerDAO either! What you can do is open a Col...

For a base rate I looked at how often non-state groups win intra-state wars in the post World War II period in the Intra-State Wars data [here](https://correlatesofwar.org/data-sets/COW-war). Base rate of rebel groups against a non-major power state is 19% in 180 conflicts. The state wins outright 41% of the time, there is a compromise 19% of the time, there's a stalemate 8% of the time, and 7% of the time the conflict continues below what's deemed at a war level. The small remainder were either still ongoing at the end of the dataset or morphed into ano...

There are 57,000 dams [in the world](https://www.internationalrivers.org/questions-and-answers-about-large-dams), Wikipedia's list of [major dam failures](https://en.wikipedia.org/wiki/Dam_failure) for 2019 lists only three incidents (there probably were more but still). So the base rate here is **incredibly** low. Of course it's not just any dam, but they're all built to withstand far more pressure than their supposed maximum water height. Imagine building a dam that wasn't, you'd be nuts! I tried googling for the equivalent amount of energy in nuclear ...

[Why these scientists still doubt the coronavirus leaked from a Chinese lab](https://www.vox.com/2020/4/23/21226484/wuhan-lab-coronavirus-china): > “If you do the math on this, it’s very straightforward. ... We have hundreds of millions of bats in Southeast Asia and about 10 percent of bats in some colonies have viruses at any one time. So that’s hundreds of thousands of bats every night with viruses,” Daszak says. “We also find tens of thousands of people in the wildlife trade, hunting and killing wildlife in China and Southeast Asia, and millions of pe...

@(casens) this is a significant improvement, but the title question still does not match the resolution criteria. In particular, they can rule that the MS law does not pose an undue burden under Roe and Casey, which would not overturn Roe from a legal perspective (it would not not be "formally overturned"), but the question would still resolve positively. If the Court does that in this case, people are going to want another question for whether at a later date they overturn Roe. The title question should be about abortion before fetal viability if that's...

Sometimes this site is completely ridiculous: "Also resolves ambiguously if Navalny is still alive in 2500 but has not assumed either office." I find some of the questions on the site around extending human lifespans fascinating, but if this is a less than 1% chance thing, maybe leave your eccentric views on someone alive today living to 2500 out of questions on other topics? More generally, it's best if questions have more proximate end dates so that we can actually learn about whether our predictions come true in less than a lifetime (spoiler alert: y...

Take chess as a barometer. Computer programs did not prevent people from playing online, they just play against people with similar ratings. Of course people don't typically bet on chess games but there's no reason why in the event of rampant cheating you couldn't implement similar ratings for poker ability and kick people out whose rating suddenly increased by too much. Yes, there would still be an incentive to cheat, but there are plenty of ways to make it harder to profit from cheating, e.g. platform fees are already a barrier. Perhaps AI might make i...

I strongly encourage you to delete "An ambiguous resolution will result if a candidates sexuality is brought forth during their campaign from a source other than themselves and they continue to win the presidency." This will result in an ambiguous result for any straight candidate, but particularly for any female candidates. I refer you to the Internet. More to the point, the question should resolve around a winning candidate's open self identity. So I would also change "The candidate must be" to "The candidate must publicly self-identify as" and replace...

I reread the question and suggest that it be rephrased. The question currently reads: "Resolution is ambiguous if a major media call has not been made as of one week after the election." While this might be an interesting additional conditional question, this is currently the MAIN question on Metaculus about the outcome of the 2020 US presidential election. People are coming to the site and looking at is an assessment of what is the probability that Trump will be reelected, but that's NOT what it actually asks. It asks, conditional on there being a clear...

If you think his probability of death in a year is 2%, @Tamay's base rate, then in the month this question covers his prior probability of death is only 0.17%. @shibboleth provides a [great breackdown](https://www.metaculus.com/questions/3445/will-kim-jong-un-still-be-the-de-facto-leader-of-north-korea-on-1-january-2022/#comment-29467) of just how common it is for Kim Jong Un to not appear in picture in the the North Korean newspaper and other posts have detailed how often some rumor like this about him starts. He may very well be having health problems ...

This question is a great illustration of the need for more control over the distribution, like a one sided interval or the ability to drag the density in a segment. With the current interface it's impossible to make a prediction that does not cover the current number of already reported cases in the 25%-75% interval.

This is a great start. There are a few things that adjust the probability. First, something that might increase the probability is to just look at after 1900 when recording is better as @zc points out. Second, housing is incredibly expensive in California so whether or not the three most expensive California quakes in the Reuters list were $10 billion "adjusted for inflation," they would be $10 billion today given California's housing costs. On the other side of the ledger, an earthquake in Alaska is unlikely to cause 10 billion dollars in damage, but th...

Democrats are [floating](https://www.nytimes.com/live/2021/01/10/us/joe-trump-biden#the-house-could-vote-as-soon-as-tuesday-on-an-impeachment-article-the-chambers-no-3-democrat-said) giving up on getting a conviction: > Representative James E. Clyburn of South Carolina, the Democratic whip, said that the vast majority of House Democrats believed the president must be impeached for his conduct but that top leaders were still trying to determine how to punish Mr. Trump without hamstringing the first days of Joseph R. Biden Jr.’s presidency with an all-cons...

@(johnnycaffeine) It's stunning how quickly the government collapsed. I am surprised that they did not really put up any fight. Even though we were seeing government soldiers surrender/flee in rural areas, I expected the government and commando units to fight for the cities, particularly the capital. I updated as major cities fell, though not as quickly as the community (seems like a flaw in weighting the final prediction so heavily that I'm getting 100+ points on this question), but I was initially skeptical when ~2 weeks ago the government still contro...

@(Linch) they did not suspend habeas corpus. They denied a petition that presumably would have required them to just immediately release anyone who had been in custody for more than 24 hours without an arraignment or bail hearing due to delays from covid. All they ruled was that they will get people in front of a judge as fast as possible, but that under the current circumstances that might exceed 24 hours. This is not a suspension of habeas corpus by the executive branch, this is a judge saying we will get people before a judge as fast as is humanly pos...

Wouldn't being booked into jail upon being charged qualify for a positive resolution as "This question will resolve positively if Donald Trump is incarcerated for any period of time, no matter how brief, before June 30, 2028." even though that contradicts the text "5) A judge would have to decide on a sentence that included time in jail." If booking for a mug shot is not considered to count, being denied bail surely would even though there would be no finding of guilt.