I lost 177 points on this question. My reasoning for assigning such a low probability to a positive resolution relied on two flawed assumptions: 1. I thought the clock wouldn't move to 1 minute to midnight, so I assumed it wouldn't move closer to midnight. It didn't occur to me that it could move in increments of less than a minute. This mistake was particularly inexcusable given that there is a precedent for a fractional move: in 2017 the clock moved from 3 to 2.5 minutes to midnight. 2. Like Jgalt, I thought it would be absurd to move the clock closer...

Intriguingly, @Rugashan has posted a total of four comments on Metaculus, three of which refer to mysterious "reports" that left users wondering.

@(nagolinc) If 538 was accurately reflecting these considerations, we should expect their model to be well-calibrated. But [Gelman et al](http://www.stat.columbia.edu/~gelman/research/unpublished/forecast_incentives3.pdf) write that "according to the calibration plot shown by Boice and Wezerek (2019) of Fivethirtyeight’s political forecasts, in this domain 20% for them really means 14%, and 80% really means 88%." Moreover, this isn't just what the data shows, but also what theory would predict. We should expect 538 to be underconfident, as a way of pr...
@(casens) As a side note, I was pleased to see him praise Metaculus: >If it's a boring enough news day that you want to cover me, consider instead covering the many other fascinating and under-covered people and institutions in and around the rationalist community, some of whom are probably women or minorities or whatever. The Qualia Research Institute is doing absolutely picture-perfect mad science. Metaculus is fast becoming what PredictIt should have been; I intend to shill it pretty hard but I can't do it all by myself. Catherine Olsson, Ibasho, and...


This seems incorrect. Here are the top five contenders according to electionbettingodds.com:

  1. Biden: 28.6%
  2. Warren: 15.5%
  3. Harris: 13.5%
  4. Buttigieg: 13.2%
  5. Sanders: 11.4%

As you can see, there is no discontinuity between the third and the fourth candidate.

— edited by PabloStafforini

What? I pointed to a betting market aggregator website as evidence against the claim that there are only three major Democratic candidates.

@Uncle Jeff

Via @Jgalt:

• @FiveThirtyEight: 77% Biden

• @TheEconomist: 86% Biden

• @NewStatesman: 80.3% Biden

• @NiskanenCenter: 99.5% Biden

• @Plural_Vote: 71.7% Biden

• @JHKersting: 78.2% Biden


  • Metaculus: 68% Biden

Is this consistent with your hypothesis?

As a measure of how ridiculous the current median value is (87%), consider that the probability assigned to SpaceX landing people on Mars prior to 2030 is higher than the probability assigned to a Mars landing prior to mid-2069 by SpaceX, NASA, or any other private firm or government agency!

@Uncle Jeff

Would you mind phrasing your comments in a less opinionated way? Imagine how much less pleasant Metaculus would be if everyone expressed their personal opinions on questions involving issues that they care strongly about.

FWIW, I am strongly in favor of this proposition. I just would like Metaculus to remain focused on forecasting.

On the seasonality of COVID-19, I found [this note](https://ccdd.hsph.harvard.edu/will-covid-19-go-away-on-its-own-in-warmer-weather/) by Marc Lipsitch, a professor of epidemiology at Harvard's School of Public Health, to be informative. I took some notes, reproduced below, in case they are of interest (though they don't add any information not already included in the paper). -- There are four factors that contribute to seasonality in the spread of viruses: 1. _The environment_. Both indoor and outdoor air is typically less humid in winter. Lab studie...

@Anthony I interpreted the question as asking about the party that would control the Senate, and my performance in this tournament will be adversely affected by a negative resolution (to be concrete, I predicted 75%). However, the question text specified the resolution criteria very clearly, and if you look at the comments below, many predictors were interpreting the question based on this text. While I agree that the title was confusing, people (such as myself) could have avoided this confusion by reading the text carefully.

@Jgalt We should create a question along the lines of "Will the cause of the mysterious bumps in so many recent questions be known by 2021?"

@nostream All the prediction markets give Trump much higher chances than the models do. I don't think the discrepancy can so easily be dismissed as resulting from manipulation, unless there's concrete evidence in favor of that hypothesis. Do you know of any such evidence?

I personally give Trump ~35% chances of winning the election, and in line with that belief I have made several bets for Biden totalling $7,000.

Sadly, Maria Giuseppa Robucci is dead. Kane Tanaka and Lucile Randon are the only two left.

It's not a "really, really obvious mistake to make". More importantly, your argument implies that pollsters would have been at least as prone to making these mistakes in all elections prior to 2016, so the historical polling error is the best estimate we have of the error we should expect in the coming election. But the FiveThirtyEight model adjusts its confidence intervals so that they reflect this historical error rate. (That's why, in 2016, it got closer to the actual result than a naive interpretation of the polls would have suggested, and in fact th...

This spreadsheet computes the chances of a positive resolution from the current eurovisionworld.com betting odds.

— edited by PabloStafforini

Via the Ought Slack workspace, I learn about @beala's excellent analysis of this question, which as far as I can tell has not been mentioned in this thread.

(This also provides a good example of how the current scoring system fails to reward users who put a lot of effort into a few questions, relative to users (like me) who put little effort into lots of questions.)

@kale I wouldn't read too much into it. The volume is extremely low (because the market is new), so the behavior of just a couple of traders might explain the discrepancy. I moved the probability of 'No' from 38% to 58% by buying $450 worth of 'No' shares.

— edited by PabloStafforini