[letitleak]

There are research projects/papers on the topic with claims that are all based on Brier scores. I'd say Brier scores are inherently biased by choices in framing/asking questions, but at least measure everyone uniformly/fairly who is given the same questions:

https://en.wikipedia.org/wiki/The_Good_Judgment_Project

For full datasets, I'm not sure if they provide it since not all researchers are good about open data. You could look at open prediction markets which are directly observable but they wont give you any insight into how professional analysts compare with the crowd.