Quote:
Originally Posted by Iowa_Flames_Fan
So Nate Silver has apparently started a twitter war with Bitecofer on this, calling it “bad math”. He came across a little nasty about it, but the issue is her model ignores the fact that polling errors are not independent events but correlated.
I’m not smart enough to weigh into the math debate. What I liked was her theory that vote preferences are not particularly elastic, and haven’t been for a few years. That “feels” right to me, based on my hunch and nothing else. If it’s true, movement in the polls is mostly statistical noise and this cake has been baked for a while.
|
That's interesting, because that non-correlation of polling errors is exactly why places like HuffPost Pollster were predicting a 99% chance of a Clinton win, with the head of HPP even writing a take-down of 538's model as a result, just days before their 99% prediction got destroyed.
It seems like following that, any predictors not accounting for that should have a very good reason why they aren't, rather than this simply being the standard way of doing things. That said, I agree that there's a lot about Bitecofer's argument that makes sense to me, and I'd be really interested to see what her model, adjusted for correlated polling errors, looks like.