Given that:
- The vast majority of participants make 1 or 2 predictions at the start of a question
- Multi-part questions can have open bins which are effectively closed based on the criteria of the question, but they are still forecast
I am going to test the following accuracy-improvement strategy:
- Stay at crowd level until material information becomes available (most questions are about things that don’t become clear until they become clear)
- If a bin is determined but not closed, predict the crowd level conditional on that bin being closed
- Once material information becomes available, incorporate it in the prediction, which will be more accurate than the crowd, because 98% of the crowd has gone home
Advertisements