Main differences between Silver and 538 models:

• Silver has PA extremely important >30% of being tipping point vs 538 has PA <20% to be tipping point.

• Silver: Trump 91% given he wins PA vs 538: Trump 77% given he wins PA

• 538 views FL + TX as legitimate swing states that could be the tipping point (17% chance) while Silver views that as a liberal fantasy (<5%)

• This generally means Nate predicts the election is like the snake chart (high correlation between states) while Elliot predicts it's more random (moderate correlation between states). The markets are more confident PA is the tipping point and TX/FL aren't than Silver

• 538 agrees with the market on Harris' chance to win popular vote (~77%)

• Silver agrees with market on probability of Trump EC win while losing NPV (~1 in 4) vs 538 (~1 in 8)

• Silvers model tends to favor Trump (liberals accuse him of including junk polls that skew the average), not in the scope of this poll which is about the methodology of uncertainty

• Looks like 538 is 90% polls, 10% fundamentals while Silver is something like 60/40

Personally, I don't really like either model, they both have their pros and cons, but there's no real insight gained from the models like there was in prior election. They all basically say it's a toss-up and the 7 swing states will decide it with extra emphasis on PA since it has 19 EVs and could push Harris to 270.

My mental model has roughly the same topline as 538 (Harris just under 2 in 3 chance of winning) but I agree more with Silver on there being less uncertainty about state correlations.

It roughly breaks down to 3 scenarios:

1: Trump wins most swing states (unlikely: 15%), this is what Silver had a couple days ago with the modal outcome being 312-226 but also possible Harris wins say WI and MI or Trump flips VA, but Trump is almost guaranteed to win PA. This is the only scenario where I could see Trump winning the popular vote but I'd have it around 50/50

2: It's close (say 250-290 range): possible (50%), in these scenarios PA essentially decides the election and its a complete toss-up (50/50) all 7 swing states are unpredictable, the other 43 not competitive. Harris most likely wins the popular vote in around 90% of these scenarios.

3: Harris comfortable win: possible (35%) this is how liberal media views the election with Trump seen as too radical, Harris wins popular vote by ~6% and comfortably wins the election with FL/TX in play and the blue wall solid.

Overall: I think Harris has ~60% chance of winning the election (agree with non-Silver models, but with less uncertainty) and a much higher ~87.5% chance of winning the popular vote.

I wish there was an empirical model that matched my intuition.

Will be interesting to look at this after the fact, especially if a state like FL/TX/VA/NH flips. Would be even more nice if we could see the probability density the forecast gave the actual outcome to find the ratio. Not sure if that's possible for the simulation based models.

Or if anyone has good data about how to model US elections. It baffles me that there's less than 10 well known US election modelers for an election that affects 100s of millions and they are taken to be gospel and each is somewhat unique but all are way too uncertain due to heavy reliance on polls, compare that to sports where the market has a strong consensus and analysts have varied approaches.

FYI: I don't subscribe to Silver Bulletin.