The Economist: Forecasting the US elections

[The Mikado]

I really don't think a 99% popular vote winning chance for Biden is crazy. When's the last time Trump led a reputable national poll? Even before covid hit he was consistently losing by mid single digits to every democratic candidate.

Decided to find out the serious answer to that question.

Let's see...

I see a tied 43-43 result in late April from IBD-TIPP. That was three months ago. But a tie isn't a win.

Another tie from Fox News in early April, 42-42.

Emerson had Trump up 52-48 in mid-February, right in the middle of the "Dead" part of Biden's Lazarus routine. Emerson is...not a great pollster, but that's literally the only lead for Trump over Biden of any poll in calendar year 2020.

[The Mikado]

Looks like the 538 model, if it's ever published, is going to be very pro-Trump based on that exchange. Really disgusting stuff from Nate Silver, but what do you expect at this point?

It's telling that Cohn and Silver went after Morris' Popular Vote % of 99% rather than his EV % of 93%. I think it's literally just an allergy to 99% as a number.

If the data sleuths in the other thread are right and the 538 not-yet-public model has Biden at an 86% chance of winning, compared to Economist's 93% chance of winning, I'd say that they're basically in total agreement. Watch 538's Popular Vote for Biden certainty be at, like, 97% just to avoid that 99%.

There's no way you can feed in "9 point average lead for challenger three months out in the middle of a recession with the highest unemployment since The Great Depression" into an election model and not have the model spit out "LOL incumbent's f**ked." There's just no way around it, that's how models are.

[The Mikado]

Feels rather nitpicky to me.

I dont think anyone wants to repeat the stupidity of HuffPost or others saying Clinton was a 90+ percent shoein but even with uncertainty the sort of polling error and change of fortune it would take for Trump to win the PV would be simply extraordinary.

^^^

There's no way you can feed in "9 point average lead for challenger three months out in the middle of a recession with the highest unemployment since The Great Depression" into an election model and not have the model spit out "LOL incumbent's f**ked." There's just no way around it, that's how models are.

Like, the only way you can talk about "Maybe Trump can come back" is gut/feeling stuff at this point and that's not what models do. If Trump does come back, it means everything people thought they knew about electoral politics is wrong, not that the models are wrong.