Some of this discussion reminded me of "Uncertainty in the Hot Hand Fallacy: Detecting Streaky Alternatives to Random Bernoulli Sequences" which tests the iid assumption on sequence data. The interesting aspect of their paper is the "streaky alternative" model they consider.
Yes! I love this paper. It's obvious once they say it, but the right prior for bit sequences is different than the right prior for individual bit rates, and there are major inferential consequences if you treat them as the same.
Some of this discussion reminded me of "Uncertainty in the Hot Hand Fallacy: Detecting Streaky Alternatives to Random Bernoulli Sequences" which tests the iid assumption on sequence data. The interesting aspect of their paper is the "streaky alternative" model they consider.
Yes! I love this paper. It's obvious once they say it, but the right prior for bit sequences is different than the right prior for individual bit rates, and there are major inferential consequences if you treat them as the same.