“Gut check, how sure do you feel?” vs. “When considering past data, what is the forecasted likelihood of this event?”
This sounds like a strawman. Bayesians do consider past data as well. If you read about Superforcasting, finding the right reference class and reasoning based on it is a key feature.
I think we need to split % into two symbols, Bayesian and non-Bayesian. One representing qualitative reasoning, reasoning from subjective observation, and one representing quantitative reasoning, reasoning from objective results.
The bayesian rule allows for engaging in quantitative reasoning just fine. While you can argue that choosing priors is a subjective act, so is choosing the distribution in frequentist statistics. Usually, frequentists model things that are not normally distributed with the normal distribution and just hope the modeling error is tolerable so they are not matching any objective truth either. Even the archetypical example of the normal distribution, human height is not normally distributed as there are a lot more dwarfs than the distribution would suggest.
This sounds like a strawman. Bayesians do consider past data as well. If you read about Superforcasting, finding the right reference class and reasoning based on it is a key feature.
The bayesian rule allows for engaging in quantitative reasoning just fine. While you can argue that choosing priors is a subjective act, so is choosing the distribution in frequentist statistics. Usually, frequentists model things that are not normally distributed with the normal distribution and just hope the modeling error is tolerable so they are not matching any objective truth either. Even the archetypical example of the normal distribution, human height is not normally distributed as there are a lot more dwarfs than the distribution would suggest.