Jsevillamol suggests using mean log odds as an improvement over mean forecast.
Metaculus currently uses the median forecast (which will be the same whether or not we transform forecasts into log-odds or not). Based on the first ~850 questions to resolve, the average log-odds outperformed the median.
This question asks whether this will continue to be the case over the next 1000 questions.
Will mean log-odds outperform median log-odds for the next 1000 questions?
Using the public Metaculus api (assuming it still exists) we will take the median forecast from the weighted community forecast and the average log-odds forecast from the weighted community and use a logarithmic scoring rule to determine which method has the better performance.
The set of questions used will be the next 1000 binary questions to resolve, which close after this question opens.
If the metaculus api no longer exists, this question resolves ambiguously.
Related questions