US Election

[Grumble]

Republicans like to block traffic, don't they. What's that all about? Particularly in New Jersey, New York bridges...

I thought blocking traffic was a terrible thing that only terrorists do, or is that only when pedestrians do it?

[geejaytee]

538 has some funny stuff going on at the extremes as well. For example, Biden gaining even more of an advantage in Washington results in his odds of winning Mississippi going down.

While that might be true if it was early in the campaign and Biden was changing policy, as a policy that was popular in Washington might be unpopular in Mississippi, I can't imagine a mechanism for that to happen late in the campaign, with state based advertising campaigns and fixed policies.

I wonder how much of the 538 correlations are measuring long term policy preferences, and how much they are measuring states with related media markets.

The Economist model, btw, fixes the correlations to be greater than zero.

Andrew Gelman had a post on his blog about this a few days ago: https://statmodeling.stat.columbia.edu/ ... -forecast/ how changing one state from possible to certain affects other states in non-intuitive ways. I don't know whether I should mention that Gelman is involved in the Economist model. (He's also had a post on his blog about how the popular vote prediction in the Economist appears to be underestimating Biden's support compared to polls: https://statmodeling.stat.columbia.edu/ ... our-graph/)

Unsurprisingly, his conclusions are: 538 are making a funky prediction that's not intuitive, and the Economist average is working perfectly as intended

[dyqik]

538 has some funny stuff going on at the extremes as well. For example, Biden gaining even more of an advantage in Washington results in his odds of winning Mississippi going down.

While that might be true if it was early in the campaign and Biden was changing policy, as a policy that was popular in Washington might be unpopular in Mississippi, I can't imagine a mechanism for that to happen late in the campaign, with state based advertising campaigns and fixed policies.

I wonder how much of the 538 correlations are measuring long term policy preferences, and how much they are measuring states with related media markets.

The Economist model, btw, fixes the correlations to be greater than zero.

Andrew Gelman had a post on his blog about this a few days ago: https://statmodeling.stat.columbia.edu/ ... -forecast/ how changing one state from possible to certain affects other states in non-intuitive ways. I don't know whether I should mention that Gelman is involved in the Economist model. (He's also had a post on his blog about how the popular vote prediction in the Economist appears to be underestimating Biden's support compared to polls: https://statmodeling.stat.columbia.edu/ ... our-graph/)

Unsurprisingly, his conclusions are: 538 are making a funky prediction that's not intuitive, and the Economist average is working perfectly as intended

I think that was where I saw the reverse engineered correlation matrices (or a tweet related to that, anyway).

What I'm wondering is how you can build a model that incorporates the observed correlations, but takes account of the timescale those correlations operate on.

[lpm]

Who wants a model that copes with extremes? What would be the point? If you cared about Mississippi and New Jersey, you'd be better off building a couple of separate models for extreme D and extreme R. Leave the main model for central ground.

[Bird on a Fire]

538 has some funny stuff going on at the extremes as well. For example, Biden gaining even more of an advantage in Washington results in his odds of winning Mississippi going down.

While that might be true if it was early in the campaign and Biden was changing policy, as a policy that was popular in Washington might be unpopular in Mississippi, I can't imagine a mechanism for that to happen late in the campaign, with state based advertising campaigns and fixed policies.

I wonder how much of the 538 correlations are measuring long term policy preferences, and how much they are measuring states with related media markets.

The Economist model, btw, fixes the correlations to be greater than zero.

Andrew Gelman had a post on his blog about this a few days ago: https://statmodeling.stat.columbia.edu/ ... -forecast/ how changing one state from possible to certain affects other states in non-intuitive ways. I don't know whether I should mention that Gelman is involved in the Economist model. (He's also had a post on his blog about how the popular vote prediction in the Economist appears to be underestimating Biden's support compared to polls: https://statmodeling.stat.columbia.edu/ ... our-graph/)

Unsurprisingly, his conclusions are: 538 are making a funky prediction that's not intuitive, and the Economist average is working perfectly as intended

I think that was where I saw the reverse engineered correlation matrices (or a tweet related to that, anyway).

What I'm wondering is how you can build a model that incorporates the observed correlations, but takes account of the timescale those correlations operate on.

I posted a link to Gelman's post on this thread a few pages back

Who wants a model that copes with extremes? What would be the point? If you cared about Mississippi and New Jersey, you'd be better off building a couple of separate models for extreme D and extreme R. Leave the main model for central ground.

The point is more that the extremes are (probably) just the tails of the distributions used by the rest of the model. If something weird is happening in the tails, that might (or might not) mean that the model is built on some weird assumptions, which might or might not matter.

But it's quite tricky discussing models whose inner workings are a commercial secret

[EACLucifer]

RE: the 538 model and Nevada; you can actually check the effect of assigning a state to one candidate or the other on the odds in other states. Assigning Nevada to Biden has very little effect on any of them. If one views the various possibilities the main model, a significant number of the few that favour Trump have him taking Nevada. It seems the effect of assigning Nevada isn't so much to do with correlations in other states, but that a number of Trump's paths to 270 in the model actually include Nevada, and unexpectedly good sun belt performance for Trump. The same effect can be seen if one assigns New Hampshire, but with the far north instead of the sun belt.

[EACLucifer]

Though I was being lighthearted by including a picture of the War Rig from the latest Mad Max film, for those not up to speed, a pack of Trump supporters in pick up trucks recently boxed in a Biden Campaign bus full of volunteers - but not containing either Biden or Harris - and tried to force it off the road, including ramming it in a scene reminiscent of the Mad Max films. The FBI are investigating.

Trump, meanwhile, despite yelling LAW AND ORDER at the slightest provocation, described the lawless, dangerous mob "patriots" and said they "did nothing wrong"

[Martin_B]

At this point I wouldn't be that surprised if sometime early on Wednesday morning Trump doesn't try and instigate the 2nd American Civil War, although in this case it'll be the American Incredibly Crass War.

[headshot]

And the Texas GOP issued this statement:

[discovolante]

At this point I wouldn't be that surprised if sometime early on Wednesday morning Trump doesn't try and instigate the 2nd American Civil War, although in this case it'll be the American Incredibly Crass War.

I actually woke up at 5am today and couldn't get back to sleep, and was almost there when my brain suddenly hit on the idea of Trump losing and deciding to use the nuclear codes in a fit of rage. Got to love 5am thoughts.

[Vertigowooyay]

And the Texas GOP issued this statement:

‘Hmmm. How can I sign off on this statement?... oh, I know. A big ol’ dollop of antisemitism.”

[JQH]

At this point I wouldn't be that surprised if sometime early on Wednesday morning Trump doesn't try and instigate the 2nd American Civil War, although in this case it'll be the American Incredibly Crass War.

I actually woke up at 5am today and couldn't get back to sleep, and was almost there when my brain suddenly hit on the idea of Trump losing and deciding to use the nuclear codes in a fit of rage. Got to love 5am thoughts.

That one has been on my mind quite a while. As has civil war.

[EACLucifer]

At this point I wouldn't be that surprised if sometime early on Wednesday morning Trump doesn't try and instigate the 2nd American Civil War, although in this case it'll be the American Incredibly Crass War.

There are people pushing for that, and they support Trump. One has been charged for shooting up and setting fire to a police station in Minneapolis as an agitator, something Barr was happy to blame on the left at the time. He was in contact with another who comitted murder against police in another state.

[El Pollo Diablo]

Biden is basically going to win Colorado, Michigan, Minnesota, NE-2, New Hampshire, Nevada and Wisconsin, pretty much without much doubt. I decided this by checking whether there had been any polls at all in recent times showing Trump in front (discounting any at C+ or worse in 538's rating). Nevada is the only one where there's really any doubt, simply because there haven't been that many polls there.

Those choices get Biden to a >99% chance of winning. With those and Democratic dead certs, Biden gets to 258 EC votes, 12 away from victory.

In that circumstance, Trump needs basically all of Arizona, Florida, Georgia, North Carolina, Ohio, Pennsylvania and Texas to win. Any one of those going to Biden (even Arizona, with 11 EC votes) will lose him the election. Being generous, let's give Trump all those except Pennsylvania. 538 still reckons Biden has an 85% chance of winning. Switch Nevada to Trump, and it's 77%, because Pennsylvania is still very likely to go blue.

And now averages, along with the number of polls in the last two weeks that have shown a Trump lead, and the number of results showing it's even:

Arizona (Biden +3.1, 4/16 polls for Trump, two ties)
Florida (Biden +2.2, 2/26, one tie)
Georgia (Biden +1.0, 5/17, four ties)
North Carolina (Biden +1.9, 2/23, three ties)
Ohio (Trump +0.2, 6/8, no ties)
Pennsylvania (Biden +5.1, 5/35, one tie)
Texas (Trump +1.1, 7/14, one tie)

[lpm]

The "time left" effect in the 538 has now gone to zero. The little adjustments for incumbency and economics have disappeared.

90% chance. Economist has 95%.

[jimbob]

At this point I wouldn't be that surprised if sometime early on Wednesday morning Trump doesn't try and instigate the 2nd American Civil War, although in this case it'll be the American Incredibly Crass War.

I actually woke up at 5am today and couldn't get back to sleep, and was almost there when my brain suddenly hit on the idea of Trump losing and deciding to use the nuclear codes in a fit of rage. Got to love 5am thoughts.

Sadly that was a thought that I first had in 2017

[El Pollo Diablo]

Beware of exit polls.

[El Pollo Diablo]

Does the USA have the same rules we do about not broadcasting political material on election day?

[jimbob]

At this point I wouldn't be that surprised if sometime early on Wednesday morning Trump doesn't try and instigate the 2nd American Civil War, although in this case it'll be the American Incredibly Crass War.

I actually woke up at 5am today and couldn't get back to sleep, and was almost there when my brain suddenly hit on the idea of Trump losing and deciding to use the nuclear codes in a fit of rage. Got to love 5am thoughts.

Sadly that was a thought that I first had in 2017

http://www.internationalskeptics.com/fo ... st11979437

[dyqik]

Does the USA have the same rules we do about not broadcasting political material on election day?

That would clearly be an infringement of the First Amendment.

[lpm]

Economist has 95%.

Economist now up to 96%!!!

Of course, that might be going from 95.49% to 95.51% but I prefer to see it as Trump's odds plummeting by 20% from 5% to 4%.

[Little waster]

Economist has 95%.

Economist now up to 96%!!!

Of course, that might be going from 95.49% to 95.51% but I prefer to see it as Trump's odds plummeting by 20% from 5% to 4%.

538 has just dropped Biden from "Clearly Favored" to merely "Favored".

I think it is time for our resident USians to dust off their English-to-Canadian phrasebooks and learn to embrace cheesy chips and gravy.

[lpm]

Per Nate "I have the best model" Silver:

Much to Fivey's chagrin—or perhaps delight—Biden keeps oscillating between like 89.9% ("favored") and 90.1% ("clearly favored") in our forecast with every random poll we add. (And boy have there been some random polls today.)

[Brightonian]

Somebody told me a long time ago that some voters in western states wait for results in eastern states, where polls close earlier, to vote for a candidate of the party that looks to be winning (to try to get more favourable treatment from the government). Is that of any significance these days? I suppose this time the greater polarisation re Trump, and more early voting, reduces the effect however.

[Little waster]

Per Nate "I have the best model" Silver:

And boy have there been some random polls today.)

No sh.t, over the weekends the polls have given everything from Biden up 18 points to Trump winning by 1.