Reverse-engineering the problematic tail behavior of the Fivethirtyeight presidential election forecast
This is a very polite post that focuses on technical details but it's frankly devastating to the modeling approach taken by and this cycle
538 has strong negative correlations between state simulations. Huh? Some possible answers
In Short: Nate was worried to release a model that had Biden at 85% or so, so he reverse engineering the model to give him 71%. In doing so he broke a lot of things on the state-level, that should give everyone pause about the model works in practice
I wish Mississippi wasn't the example here. Historically, wild outcomes in MS really have been negatively correlated with the northern-tier! IDK if that's actually relevant in the 538 model design, but it was hard for me to shake
Interesting dig into the 538 simulations
Andrew's the best "...Yes, I know my R code is ugly. But ugly code is still much more understandable than no code..."