My Bets — An Updating List
If you have a belief, you should be willing to bet. It’s useful as mechanism to ensure people’s claim aren’t empty signalling. It’s also good as a way to check that your claims are concrete enough to be resolvable, and that you don’t change the way you score your predictions post-hoc. So I offer to bet.
Sometimes, the other side agrees. Here’s a list of who I’ve bet with:
I will pay 100 Euro to Grzegorz Chrupała, Associate Professor of Computational Linguistics at University of Tilburg, if there is no pair of languages between which an AI can translate a random popular press article as well as a professional translator, as of September 20, 2022, as judged by a blind panel. Otherwise, he will pay me 1000 Euro.
Won, Resolved — If the 7-day average value for R as reported by the Israeli Ministry of Health from July 12 — July 18, 2021 is above 1, Sam Elder will pay me $120. If it is below 1, I will pay him $80.
Relative 2020 Prediction Accuracy Bet w/ Peter Hurford
Lost — Friendly $10 bet on whose predictions fared better, based on VOX’s judgement about resolutions.
COVID infects 25% of Population
Lost, paid— I bet $200 at 5:1 odds (his $1000 to my $200) that more than 25% of global population will have been infected as of Jan 1, 2021, per estimates published in peer-reviewed journals as of June 1, 2021.
Lost, somehow — I will pay $60 to Giego Caleiro to his $300 if, as of 1/1/2021, the consensus of doctors is that using hydrocloroquine in treating COVID19 is/was effective, and that in RCTs, the impact of use was statistically distinguishable from no treatment. If we do not agree, if Fauci writes a paper or posts about it, his determination will control. Otherwise, Joao Fabiano will decide by request of arbitration. (Link goes to a private Facebook group.)
Note in August 2020: I don’t know if I would have taken this bet at even odds, and we have 4 months left, but I’m pretty happy about how this one turned out. And in November, further evidence.
Note in February 2021: Giego says he’d like to ask Joao to arbitrate.
Note end of March, 2021: Dispute resolution criteria matter. Oh well.
Fivethirtyeight Presidential Model Movement (Short term.)
WRONG. Paid. I bet $100 at even odds that Trump’s probability of winning as estimated by 538 would go up in two weeks post-debate. I think that the model is relatively well calibrated, but does not account for short term shifts as well as it needs to given the news about Trump and COVID, and think that his recovery is very likely, at which point his poll numbers will revert to nearer what they have been. This makes a Biden win very likely, but despite that, I think 16% Trump is too low an estimate at a week and a half before the election, and the model will reflect the continued possibility of a surprise.
Trump winning conditional on models predicting a Biden win (Short term.)
Correct. (Agreed, Paid.) There was a claim that “Polls will show Biden leads. Trump will win, and it won’t be close.” We agreed to bet $100 at even odds. (The conditional has been fulfilled.) (It has been paid.)