Read on Twitter
My 2c on the "did the models/aggregates really miss" debate: I think we need some clarity on what these things really are. Are they meant to predict elections? Or do they just present a range of possible outcomes? 1/
It's not really that cut-and-dried, but there's sometimes a motte/bailey style argument made where people retreat to the variance when things aren't going well but embrace the point estimates when they are. 2/
In fact, we evaluate these goals quite differently; if the goal is to present a range of outcomes then we actually want the point estimates to be off. A model's 75% prediction should be wrong one-time-in-four. 3/
To his credit, @FiveThirtyEight has done those calibration checks here; this is more about the broader discussion and how people use them in the real world. 4/ https://projects.fivethirtyeight.com/checking-our-work/
Because it's really easy to take the second aspect (range of possibilities) and make the models basically unfalsifiable. 5/5
