It may be interesting to see the distribution of predictions made by different people who participated in the contest. Are you planning on publishing this data?
Yeah that's a good idea, maybe when I have some time this weekend I'll anonymize it and put it up on Github.
There were some interesting patterns actually. Several people did a 1s and 0s strategy where they basically treated it like a long-shot lottery ticket and predicted each race with complete certainty (none of the winners did that though).
I might also write more on this later, but it's interesting to think about how different incentives change the optimal forecasting strategy. Like the strategy for trying to win a contest probably is different than the strategy for just maximizing calibration. A bit like how tournament-style poker is different from cash games.
It may be interesting to see the distribution of predictions made by different people who participated in the contest. Are you planning on publishing this data?
Ok here are the anonymized contest forecasts:
https://github.com/mikesaint-antoine/blog_projects/blob/main/election_forecasts_2024/forecasts/anonymized_contest_entries.csv
Yeah that's a good idea, maybe when I have some time this weekend I'll anonymize it and put it up on Github.
There were some interesting patterns actually. Several people did a 1s and 0s strategy where they basically treated it like a long-shot lottery ticket and predicted each race with complete certainty (none of the winners did that though).
I might also write more on this later, but it's interesting to think about how different incentives change the optimal forecasting strategy. Like the strategy for trying to win a contest probably is different than the strategy for just maximizing calibration. A bit like how tournament-style poker is different from cash games.