In the paper, we focus on percentage of question duration as our primary measure of time. This measure standardizes our data with respect to the wide range in question duration and calendar start and end dates. The figures below present confidence and accuracy with respect to some alternative measures of time.
This measure considers the temporal distance of the judgment in days from the eventual resolution of the question. Because of varying question length, we have many observations near 0 because all questions in the sample have closed, but fewer and fewer observations as we move to more distant judgments which were not possible on shorter questions.
This measure considers the temporal distance of the judgment in days the question was opened for forecasting. Because of varying question length, we have many observations near 0 because all questions in the sample had a first day, but fewer and fewer observations as we move to more distant judgments which were not possible on shorter questions.
We can also ask how confidence and accuracy have varied over the passage of time across three tournament years without specific reference to the individual question timelines.