Fogdog
Well-Known Member
voodoo? No, Nate sweats the details and he can give good reasons for the way his model works. It's not like he's predicting spring using a groundhog. If you can listen to Nate, he's obsessed with looking for reasons to prove the model is wrong.I go to their site and found the voodoo behind the models interesting, I thought there would be more statistics and less assumptions! Nate knows what he is talking about, he called it last time and seems to have some insights. The difference this time is Trump and covid, it is a novel situation and the models may break down, it's still art as much as science.
Covid is a novel situation? There isn't a Covid widget in his model, just polling data which accounts for public sentiment, not Covid. A surpise vaccine that Fauci endorses would be a positive factor for Trump but that's why there are fudge factors in his model to based upon time away from the election.
538's model is completely reasonable to inflate Trump's chances of winning due to unknown factors that can come up between now and Nov 3.
For example, Comey's midnight memo just a few days before the 2016 election that heaped doubt upon his earlier findings in Clinton's e-mail scandal. No one outside of the FBI could have predicted that. That right there is the time factor playing out in real life.
I think a problem with the perception of polling and forecasting is the way people take continuous data and turn it into a binomial outcome. For example, 538 gave Trump a 30% chance of winning the election in 2016. It was exactly correct because Trump just barely won. However, what is it we hear today? "Trump was projected to lose but he won". People don't understand what "30% chance" means. They just turn it into a binomial "win" or "lose" in their heads.