@Joker If you want to maximize your points in expected value (and therefore maximize your points almost certainly over repeated trials, assuming statistical independence), you should predict your actual belief.

Several people have already brought up the idea of a new prediction parameter: called "certainty" or "confidence" etc. These ideas are clearly getting at a real issue: that the current scoring system favors many low-quality predictions over a few high-quality ones, for example. This incentive lowers the overall quality of predictions available to Metaculus, and points more broadly to how Metaculus is unable to distinguish between a prediction made off a first impression and one made with hours of research and estimation: both types of predictions are in...

@randomuser2323 I think “insurrection” is the more technically correct term, since this attempt to seize power was not backed by threat of military force or elite political power.

@Reprisal It counts, but the question also has the condition “as long as they do not explicitly rescind said concession within 24 hours.”

@rakyi it's included as a part of "> Jan 1, 2100"

@(johnnycaffeine) I don’t exactly trust Ann Coulter to do due diligence in reporting what can or can not happen as a result of strangulation. Here’s a source pointing out that “blood foam” is a possible result of strangulation: http://bmcsagar.edu.in/new_upload/STRANGULATION-.pdf Foaming at the mouth is a symptom of seizures (including when induced by drug use), which can be caused by nervous system damage via oxygen deprivation: https://www.medicalnewstoday.com/articles/321756 A web search turns up no shortage of news stories where foaming at the mou...

@EMP A 95% chance gets a brier score of 0.005. It would take 100 of those to make up for the loss incurred by a 50% prediction.

@Fruo Those predicting higher than 1% are not considering the types of events where human adaptability plays much of a role. The focus is more on the type of event that would destroy all macroscopic life on Earth.

@jabowery ironic imo to link to a YouTube podcast that titles its episodes in alarmist style with all-caps words like “ATTACKS”, “WARNING”, and “CRAZY” while warning of the dangers of groupthink.

ironic also to condemn derision with an another heaping of derision.

But let’s allow the log score to be the ultimate judge.

@JavierSouto It appears to me that you made no real effort to understand how 538's model works. Here's some reading to get you started: https://fivethirtyeight.com/features/how-five…

On the shy voter hypothesis: https://fivethirtyeight.com/features/shy-vote…

@yshemesh One solution would be to randomize the winners. For instance: each entrant gets draws in a lottery in proportion to 1 - their average Brier score. This would ensure that the strategy to maximize expected prizes is also the strategy to optimize Brier score.

Pfizer vaccine demonstrates over 90% efficacy in Phase 3 trial with n=43,538.

They expect to have 50 million doses produced this year, and to apply for emergency use authorization from the FDA in the third week of November.

— edited by JonathanShi

@(johnnycaffeine) Yes, yes, the medical examiner’s report has already been studied to death here. One thing you might not be understanding correctly (and which was reported incorrectly in the media at the time) is that “cardiopulmonary arrest” is a very non-specific term. It simply means that the heart and lungs stopped at some point. More precisely, it is considered a *mode of death* or *mechanism of death* not a *cause of death*. Asphyxiation would be an example of a cause of death whose mode/mechanism is cardiopulmonary arrest, which would also not ca...

I am not sure if the incentives are there for the prosecutors to drop the case, or if the judge will necessarily dismiss the case because of the pardon. My understanding is that he may still be found guilty; it is only his sentence that is waived by a pardon.

— edited by JonathanShi

@AvrahamEisenberg Well, with a continuous finite PDF, the probability will always be less than 1% in a small enough neighborhood of the lower bound if the range is compact. Artifact of using a continuous PDF to predict a discrete set of outcomes.

@DNorm800 I tried what happens if you set all of PredictIt's probabilities under 0.15 or over 0.85 to match FiveThirtyEights's, in this way effectively ignoring all states where PredictIt had extreme probabilities. In this case, 538 still wins, 0.07164562314 to 0.07212090157.

Same if the threshold is set to 0.1/0.9: 538 wins 0.07164562314 to 0.07410227137.

— edited by JonathanShi

I don't know why people continue to make models of presidential elections based on indirect indicators or influencers of public opinion, whether it be primary support or economic metrics or anything else along those lines. Why would you expect any of those to be more predictive than direct polling, when opinion formation is such a complex phenomenon? This feels like using diet history or weight measurements to predict if someone has diabetes, despite having access to direct measurements of blood sugar. Maybe you could make the case here that individual...