Nate Silver has long defended keeping them in. It’s not that the absolute number is any good, but a change in the number can be good. If Rasmussen shows a 3 point shift between two polls, that’s probably real and can be applied to the model.
It’s not a bad take - if it shows a consistent bias, it’s still consistent data. It’s translating the bias from a descriptive to a predictive model that’s the hard part. Maybe they found that the swings in correlation were too wide.
Nate Silver has long defended keeping them in. It’s not that the absolute number is any good, but a change in the number can be good. If Rasmussen shows a 3 point shift between two polls, that’s probably real and can be applied to the model.
It’s not a bad take - if it shows a consistent bias, it’s still consistent data. It’s translating the bias from a descriptive to a predictive model that’s the hard part. Maybe they found that the swings in correlation were too wide.