Metaculus should adjust election forecasting questions such that forecasters are forced to make their forecasts add up to 100% over all options (with an additional option “noone of the above”).
The goal would be that forecasters would be forced to make internally consistent forecasts. That should reduce noise, firstly by reducing unintentional errors, secondly by cleaning up probabilities (by quasi-automatically adjusting the percentages of candidates who may previously have been considered to be low-but-relevant-probability candidates), and thirdly by crowding out forecasters who do not want to give consistent forecasts (which I assume correlates with low-quality forecasts). It should also make forecasts more legible and thus increase the demand for metaculus.
Metaculus currently lists 20 people who could be elected US President (“This question will resolve as Yes for the person who wins the 2024 US presidential election, and No for all other options.”, “Closes Nov 7, 2024″), and the sum of their probabilities is greater than 104%. Either this is not consistent, or I don’t understand it and with all due modesty, if that is the reason for my confusion, then I think many people in the target audience will also be confused.
Metaculus should adjust election forecasting questions such that forecasters are forced to make their forecasts add up to 100% over all options (with an additional option “noone of the above”).
What goals (for Metaculus, for observers, or for predictors) does that serve?
The goal would be that forecasters would be forced to make internally consistent forecasts. That should reduce noise, firstly by reducing unintentional errors, secondly by cleaning up probabilities (by quasi-automatically adjusting the percentages of candidates who may previously have been considered to be low-but-relevant-probability candidates), and thirdly by crowding out forecasters who do not want to give consistent forecasts (which I assume correlates with low-quality forecasts). It should also make forecasts more legible and thus increase the demand for metaculus.
Metaculus currently lists 20 people who could be elected US President (“This question will resolve as Yes for the person who wins the 2024 US presidential election, and No for all other options.”, “Closes Nov 7, 2024″), and the sum of their probabilities is greater than 104%. Either this is not consistent, or I don’t understand it and with all due modesty, if that is the reason for my confusion, then I think many people in the target audience will also be confused.