538 model & poll tracker thread

[emailking]

538 is being super lazy this year. There's no excuse for the model not being out yet when the primaries have been done for months and even the 2016 version came out sooner.

The site is free. They're not under an obligation to make a model.

[emailking]

It's honestly a disgrace that their model isn't up yet. What is the hold up?!

Design and run your own model if it's that much of an issue for you. They have no obligation to even make a model at all much less on your preferred timeline.

[emailking]

Nate on his podcast released a few hours ago said he's 98.7% done, but there will be a bit of a delay of when it is released and that he is not rushing anymore to release it.

Nor should he rush it. He doesn't have an obligation to release one at all

I mean, he does though? He runs a politics website that relies on advertising dollars, especially after it was bought by ABC. They need clicks and views and the forecast is a huge part of the website every 4 years.

He doesn't have an obligation to us. We didn't pay him. Don't go to his site if you don't want to. What obligations he has to people he's actually signed contracts with, I don't know. But there's a ton of content on the site, and I doubt he ever guaranteed anyone that he would have a formal model of the 2020 election.

[emailking]

or the polls would have to be pretty systematically biased towards Biden. Both of which are possible, of course, but you can't really put a number on it.

You can though based on historical systematic bias. It's applied both to Trump and Biden in the model.

[emailking]

I imagine most of it is the model accounting for the possibility that the polls are all way off because of the same systematic errors. (But it applies this to both Biden and Trump which is why some huge Biden wins are possible.)

[emailking]

Biden up to 75, and Trump down to 25, in the model.

[emailking]

I don't agree with his assessment. We've had 58 Presidential elections which is nowhere near enough to know how weird things could get if you could run this election 40,000 different ways. Heck just look at 2016 where a single sate Utah is about 20% less Republican than it should be given how the other states went.

[emailking]

I don't agree with his assessment. We've had 58 Presidential elections which is nowhere near enough to know how weird things could get if you could run this election 40,000 different ways. Heck just look at 2016 where a single sate Utah is about 20% less Republican than it should be given how the other states went.

UT is the most cherrypicked example to underline this point. Third party candidates were uniquely strong in 2016 in comparison to other election cycles and that an independent conservative from UT whose whole campaign was basically about winning UT would perform quite well there should be factored in, even with no polling (and far as I remember, it was actually?). Such circumstances are simply non-existent to that degree this year.

Or what would be a logical explanation of OR going red with SD going blue, apart from that state correlations are way to low?

I have no idea. Yes there's a reason why Utah was not as red as expected, just like there'd be a reason why Biden wins everything but New Jersey, if that happens. It's probably a reason that wouldn't apply to the succeeding election either. Who knows?

[emailking]

I don't agree with his assessment. We've had 58 Presidential elections which is nowhere near enough to know how weird things could get if you could run this election 40,000 different ways. Heck just look at 2016 where a single sate Utah is about 20% less Republican than it should be given how the other states went.

These weird scenarios keep coming up in the 100 scenarios that actually show up on the website though.

He addressed that in this thread.

https://twitter.com/NateSilver538/status/1300825856072454145

Basically the model thinks there's a non-trivial chance something weird can happen. The specific maps that are being called out though individually may have a very low likelihood of occurring.

[emailking]

I don't agree with his assessment. We've had 58 Presidential elections which is nowhere near enough to know how weird things could get if you could run this election 40,000 different ways. Heck just look at 2016 where a single sate Utah is about 20% less Republican than it should be given how the other states went.

Utah didn't vote 20% less Republican than it should based on the polls.  At this point in the 2016, there were several showing Trump with a very narrow lead in Utah.  So if the model came up with an output in which Utah voted Democratic, there was at least evidence to support this.  Trump actually outperformed his polls by a significant degree in Utah.   

By contrast, I'm not seeing any evidence whatsoever suggesting any strongly Democratic states have the potential to go rogue in this election.  And I can't even remember any huge shock results like this in the past.  Probably the biggest shock relative to PVI was Obama winning Indiana in 2008, and this too was predictable from the the polling by September.

Sanders's win over Clinton in Michigan in 2008 was a 20 point swing from the polling that week.

Now I know you can give me reasons why that's different and why it won't happen this time (and yes I know it was the primary). The point is that shock results happen. And your guess is as good as mine at how likely a shock result in Oregon is if you give it a go 40,000 times.

[emailking]

I think there is less than a 1-in-40,000 chance that Biden wins in a landslide but loses any deep blue state

I don't agree with that, but obviously impossible to know either way. One way I can think that could happen is a last minute write-in campaign by someone really popular, focused in just one state, to make a point. I don't think Nate's model is predicting write-in campaigns, but I do think it sees Indiana and Utah (apparently) randomly swinging hard from one election to the next. That affects how it models the fat tail, and it should.

[emailking]

There's no obsession with Trump winning Oregon. All they're doing is plugging in poll numbers and re-running the model, which, by the way, thinks a Trump win in Oregon is pretty unlikely.

[emailking]

There seems to be a lot of cases where Trump wins OR in a Biden landslide though

A lot of example maps. He hasn't explained how those are selected but it appears to be forcing some kind of even distribution across the range of possibilities, so you get something from the tails every time. Each of these Oregon losses in a Biden landslide almost certainly has a very small chance of occurring.

[emailking]

There seems to be a lot of cases where Trump wins OR in a Biden landslide though

A lot of example maps. He hasn't explained how those are selected but it appears to be forcing some kind of even distribution across the range of possibilities, so you get something from the tails every time. Each of these Oregon losses in a Biden landslide almost certainly has a very small chance of occurring.

Small yeah, but more than 1 in 40000? Biden has a 3/100 chance of winning SD and Trump a 9/100 chance of winning OR. That's already a 21 out of 10000 chance of happening, assuming state results are independent events. Once you implement some sort of correlation metric (since states results are highly correlated across similar states), I can't see how Trump winning Oregon and Biden winning South Dakota are within a 1/40000 chance

yeah I don't know what the chances of those maps are, or should be. But to conclude it means they are obsessed with Trump winning Oregon is pretty hyperbolic.

[emailking]

https://twitter.com/AndrewSolender/status/1304979308822265857

So infuriating. Nate consistently had him at a 1 in 4 to 1 in 3 chance in the last month or so, and had him doing better than any other model.

[emailking]

I'm going to go ahead and guess that the Silver truthers in this thread didn't read his book.

[emailking]

How is this possible?  What is actually necessary to move the needle?

He's going to gain a point every 3 or 4 days from now until election night. Unless the bottom falls out on Trump. That's what will happen.

[emailking]

He said in the aritcle that went up last night that Biden's odds would be 91% on Election Day if things stay consistent.

Given that 91% was exactly the odds he gave Obama 2012 on Election Day when Obama only had a 1.5% national polling average lead over Romney, it says a lot about how much more cautious this model is that it'd give Biden's much larger lead the same chance.

I'm not so sure about that. The model also detected structural advantages for the democrats in the electoral college in 2008 and 2012. But now it thinks Trump has a structural advantage.

[emailking]

LOL! A single poll showing Biden +3 made Biden's lead in the poll aggregate crumble almost a point. And it's not like it is from an A+ pollster.

Can't simply re-running the model cause this?

[emailking]

Oh whoops I misunderstood too.

[emailking]

The Nebraska districts alone tell you everything you need to know. Supposedly rural Nebraska (NE3) will trend more Democratic than Omaha based NE2. It's absurd on it's face.

It's not. Here's a counter argument. It's much more Republican so it has more room to move Democratic in a polarized society. Here's another counter argument, you and others in this thread seem to base a large fraction of your criticism in how the tails are modeled, scenarios which are very unlikely to unfold and it agrees are very unlikely to unfold. I don't care if Oregon goes Republican before TX once in a blue moon. I care if Texas is going blue on Nov 3.

In other areas, you can tell how heavily the model relies on polls. Why the difference between Oregon and Washinigton despite being very similar states? - relatively good polls for Trump in Oregon, and very bad ones for him in Washington.

This is supposed to be a bad thing?

[emailking]

Well he's right.

[emailking]

Trump is down to a 15% chance of winning.
Biden has a 35% chance of winning in a landslide.
Georgia is on the cusp of flipping.
I am beginning to wonder where the model can bottom out.

Based on the "polls now" tidbits Nate has given us before, the model probably thinks Trump's chances are about 5% if the election were tomorrow. So unless he can improve in the polls, that's probably where we'll end up.

[emailking]

86-14 now.

[emailking]

https://twitter.com/realDonaldTrump/status/1315293995128573952