Part of the story could be median is not a linear function, and there is no guarantee that the median of the difference between two distributions equals the difference of their medians.
I don’t mean to say that, but maybe it’s entailed. (I don’t know.)
I’m saying even if it were the same group of forecasters who forecast all three questions, and even if every one of those forecasters is internally consistent with themselves, it will not necessarily be the case that the median for the “how many months” question will be equal to the difference between the medians of the other two questions.
Part of the story could be median is not a linear function, and there is no guarantee that the median of the difference between two distributions equals the difference of their medians.
So you’re saying that Metaculus incentivises betting based on the median, not the average?
I don’t mean to say that, but maybe it’s entailed. (I don’t know.)
I’m saying even if it were the same group of forecasters who forecast all three questions, and even if every one of those forecasters is internally consistent with themselves, it will not necessarily be the case that the median for the “how many months” question will be equal to the difference between the medians of the other two questions.
Sure. But my question was whether we take the Metaculus timelines as indicative of the forecaster’s averages or medians.