At the risk of stating the obvious, an incumbent president contracting a deadly virus is not something that’s in the training set for our election model. My prior is that this has the _potential_ to cause a 99th-percentile shift in the race, while still not being all that likely.
Our model explores a range of outcomes for potential poll change between now and Election Day. It’s constrained by the historical distribution of poll swings in the last month of last campaigns. 99th-%ile means a change that comes outside the margin of error of that distribution.
I need to think about this more. On the one hand, our model says that a poll swing from a president getting a deadly virus is a 99th-percentile (IE very rare) event. Surely that’s the case with Trump’s diagnoses? A 1-in-100 events? And, it might not even change anything.
Like, we still WANT 1-in-100 events to be rare, or else we’re overstating their likelihood.
An open question is whether we handled that right re: covid, which we only have inflated uncertainty for via a proxy in economic volatility. So our model does have an uncertainty boost.
I agree with Joel here. But I’d urge people not to think too hard on what chance modelers assigned to Trump getting covid in the grand scheme of our model, 1948-2016. Surely it was unlikely?
This is one of those times where it’s best to just wait and see how the public reacts.
I guess one of my points is that there’s still a distinct possibility that this doesn’t actually change the posterior distribution of vote intentions much (or at all?). Our model doesn’t forecast Trump/Biden’s % of getting covid, just vote share on Nov 3.
It’s worth noting that Boris Johnson didn’t get a bounce in support after being hospitalized for covid.
Share this Scrolly Tale with your friends.
A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.