Mozart Data reposted this
Teoscar Hernandez won the Major League Baseball (MLB) #HomerunDerby. Going into the final round Bobby Witt Jr. had hit 37 homeruns and Hernandez had hit 35. If that's the true base rate (with say, the same denominator of pitches), and they'd see about 33 more pitches (27 plus the bonus format), I was interested in the likelihood Witt would win. ChatGPT can simulate (assuming a binomial distribution) and arrive at 52.3% (so very close) for Witt Jr. This is an example of a problem where #AI excels (and is an amazing tool to use). I think it's also good to know the bias of the underlying assumptions. Witt and Hernandez were likely tired, which would probably drive the result closer to 50/50. As a casual observer, there seems to be serial correlation (being in the zone / momentum / hot hand), which would also drive the result closer to 50/50. There is probably some mean reversion, which would also drive the result closer to 50/50. There might be a different base rate than I instructed it suggested by batting history outside of the derby, initial (or update) betting odds, or other observables (like barrel rate), which could go in the other direction. Given how incredibly close the odds are (and most biases would suggest that a 52% favorite might be too high an estimate), it's not at all a surprise Hernandez won.