@Celene not significant, especially since it's play money. The standard argument is: you could easily short a company's stock and then assassinate all the executives to make it go down, but nobody does this because we have laws and also you'd have to be a complete psychopath.

@ErickBall exactly. Who exactly is going to kill 20 people to make a few thousand Mana?

@Celene We should make some big markets involving easier actions, just in case.

@MarcusAbramovitch you could probably daisy-chain this prediction with others, like daily active users on Manifold, and a bunch of others. At the end of it, you could be talking hundreds of thousands of fake money!

@CromlynGames I don't know...how do you know that such a trend isn't counterbalanced by a decline in some other potential cause of running amok? There are a million things that could make someone go on a killing spree. I don't think picking out one that's trending up is a very rigorous method

@Duncan A jump as a result of a very public shooting, but there was not actually 20 people killed. It is probably a lot higher.

@Duncan It's a bet against impulsive buyers, as well as an attempt to profit off of market swings. Presumably some of these people have automation to cancel their orders if the market is swinging too quickly.

@LucaMasters 6 events in 11 years (2012 through 2022), but since 2 events were in a single year, this would have been YES in 5 of the last 11 years = 45%. But we are 7 weeks into the year, so that leaves ~39% of YES using your approach.

@AlQuinn Good thought and observation, however there is not a lot of data to really support a measure of final precision...are you sure an instrumenalist approach really works on this problem set? Need to back up into philosophy and make sure we are not being so, "smart," about it (in terms of being good at pulling out fancy impressive math) that we actually are being stupid (in terms of a philosophy of science perspective - can you really build a measurement of final precision? What right do we have that future A will be similar to future B?)

@PatrickDelaney Base rate is really all we have here, though maybe there are some copy-cat things going on at the margins that create temporal correlations would be apparent if the sample size was larger. But looking at the array of methods for the data we do have, a credible range for this is ~30-40%. Might be easier to predict earthquakes since there we at least have Omori's law!

@PatrickDelaney There's nothing particularly fancy about 1) assuming the future will be like the past (5 of the last 11 years having such an event) and noting that a sixth of the year is already over (adjust down several percentage points from 5/11). That basically takes you to exactly where this market is at the moment.

Also, for what it's worth, Wikipedia has only one shooting with over 20 deaths occurring between 2000 and 2011, bringing the total for the last 22 years to 6 out of 22 (27%). It definitely makes sense to upweight the more recent years at least somewhat, but the relative lack of such events in the decade before 2012 should definitely still have some downward influence.

@Harrison As a Christian, I strongly disagree, especially if you wind up spending your mana on donations to the orphaned or widowed

@Harrison nah, you'll be fine. It's better to discuss it than sweep it under the rug.

@ShakedKoplewitz Where did you get that data?

@brp I found data which gave me a 36% probability if modeling as a Poisson process, but that assumed there was only one qualifying mass shooting in 2017, when in fact there were two. Doubling the number of relevant mass shootings in 2017 increased the probability to 45%.

@brp Does it really fit a poisson process well? Seems unlikely each person killed would be independent.

@JemBishop size of each event is independent

@JemBishop I think the Poisson model mentioned by @ShakedKoplewitz was probably a model quantized using mass shootings with over 20 deaths as independent events. Doing that with data from 2010 to 2021 got me a 36% probability. There are not many events, so this model is very quick, dirty, and error-prone.

@brp right ok that seems more likely. I think that extreme value analysis would be more robust though, as you say, you throw away a lot of information in only considering the >20 events.

@JemBishop 36% is not too far off from Jem's finding of 30% using the Extreme Value Distribution. While I have not done a full treatment of the two methods, I find that when you reduce frequentist analysis down to its geometrical components, and assume around the same p-value, and sidedness, there is a reason for it, e.g. a mathematical reason, somehow underneath the two analyses are very similar to one another in how the underlying math is transforming the set data. Again, not sure about this...just a hunch - I have seen it before.

If my hunch is correct, that says nothing about the measure of final precision, it just shows that you each used methods that agree with one another.

I am really not sure that an instrumentalist approach to this problem will be really defensible. Why? Much of it has to do with legislation, costs, things that have a huge amount of noise and confounding factors. When the cost or accessibility of guns goes way down, you tend to see more gun deaths (including suicides), so you might see more mass shootings, which confounds attempts to model what really doesn't want to be precisely modeled. TD;LR humans be unpredictable.

@JemBishop I wonder if the state of the economy, including layoffs, inflation, etc influences this chance. Also, given the state of the world in 2020, I wonder if it should be excluded (not sure that would make a difference, though).

@PatrickAupperle there is a small upward trend in the data which might suggest a higher probability, but also we are also ~10% through the year which lowers the probability

@JemBishop Why Gumbel specifically?

@ShakedKoplewitz If you try and model extreme values, eg the maximum per year, then the theoretical distribution of the max (regardless of the actual underlying probability distribution) is part of a family of the so called extreme value functions of which gumbel is a commonly chosen special case. This is the approach when for example predicting flood discharges in hydrology. An alternative is to use the POT (peak over threshold) method.

@PatrickAupperle Inflation has been low last 6 months and unemployment is at a >50 year low. But crazy can crazy whenever it wants.

@JemBishop First off, very impressive yet again that you actually attempted to fit a mathematical function on this problem set. This is in no way disparaging your approach as that is extremely laudable and beyond what I attempted. I think you should put your findings together in a blog post, honestly. I am not familiar with the Gumbel distribution and am left reading wikipedia and trying to decipher something I am not an expert in. Meanwhile, I see this study where they used Poisson: https://www.imse.iastate.edu/files/2022/05/LeiXue-dissertation.pdf

That being said, it is super intriguing what you would have to say and I would 100% read it and post it on Twitter or whatever.

@PatrickDelaney Thanks! I elaborate on the methodology I used here. https://medium.com/@jembishop/predicting-the-probability-of-mass-shootings-with-extreme-value-theory-2be88e829da2

@JemBishop I'd tip you for the write up if I could

@JemBishop https://twitter.com/iotmpls/status/1628222070495608834

@Luka Or if a person dies days later in the hospital would that count?

@Luka I mean, if the person dies in the hospital from bullet wounds I would say that would count. If they died from falling down the stairs I would say no.

@ForrestTaylor Agreed.

@StrayClimb I don't want to feed into the idea that these markets are bad/cause people to murder. Awkwardly changing the wording doesn't do anything

@MarcusAbramovitch my claim in that wording does matter because people don't naturally understand short selling or view markets as two sided.

there'es something incredibly confused about this claim. Do the nuclear risk markets phrased as 'will a nuclear weapon be detonated' have a similar issue?

@jacksonpolack where possible I think it's beneficial to us to phrase things so that "yes* resolutions are good because the average person/journalist thinks that way. I agree it's not always possible and wouldn't suggest it in those cases. This particular claim seems like if a bad event did happen could easily be taken very out of context.

Imagine you're the head of school safety after such an event. How would it feel to confirm "yes I bet that this would happen" vs "I raised the alarm that I believed we were at ruseh of failure because I wanted to send a message that we need to change".

My argument is that the former bet can more easily be misconstrued as evil/uncaring than the later even though they're logically identical.

even if that is a real effect, the size of the effect on 'people disliking manifold' would be 10,000 times lower than stuff like https://manifold.markets/IsaacKing/did-jews-make-up-the-holocaust-and or https://manifold.markets/levifinkelstein/are-black-people-genetically-dispos so I don't think it's worth worrying about

@StrayClimb I wish I could downvote your comment. Too much political discussion is dominated by the need to say things that have "positive emotional valency" -- i.e. that sound nice. When vague associations take precedence over analytical reasoning, the results are not good...

@jacksonpolack If the holocaust-and-racism markets embolden others to openly discuss issues like how frequent mass shootings are in the US, that seems like a pretty substantial good...

@jack 2017 had 2, 2019 had 1, and 2022 had 1. 2018 had one with 17, so pretty close.

Seems like about a 50/50 chance each year.

@PatrickAupperle but it should go down with each month right?

@jknowak Correct.

@jknowak Looks like January and March are less likely to have mass shootings, though.

@brp I was using data from a newspaper article for this. Wikipedia has more comprehensive data. For 2022, 4.4% of mass shooting deaths and 5.8% of mass shootings occurred in January and 6.0% of deaths and 5.8% of mass shootings occurred in February.