What does a BIDEN Presidency look like?

[Valmy]

I hope Trump's 2016 victory inoculated just enough people against a repeat.  Hopefully at least some liberals learned to take the threat seriously, because clearly most didn't, and at least some conservatives would recoil from seeing what came.

The funny part was at the time the weird confidence of some of other leftwing voters was all I had to feel better about the situation. I figured...well maybe they know something I don't? Well they didn't.

[crazy canuck]

I think most did - including the pollsters.  His victory was a big surprise shock.

I recall the polls being very close throughout with Hillary threatening to pull away a few times only for it to tighten up again.

From an article that analyzes why all the polls predicting a Clinton victory got it wrong.

In the weeks leading up to the November 2016 election, polls across the country predicted an easy sweep for Democratic nominee Hillary Clinton.

https://www.scientificamerican.com/article/why-polls-were-mostly-wrong/

All of this accords with my recollection that all the pollsters were predicting a Clinton win.  I made a bet at the time with one of my partners that Trump would win.  He gave me good odds based on his confidence in what the polls were saying. 

[OttoVonBismarck]

FWIW while I largely think polling data is much less accurate now because of the low phone response rate (this forces pollsters to use ever more arcane models to adjust for their sample size being of the increasingly narrow subset of people who are willing to answer a phone from an unknown number and take a phone poll); I think the projections for 2016 weren't off that much, what was off was people's understanding of projections.

For example Nate Silver (who I largely ignore now, and think isn't a great analyst these days), had Hillary at I think a 70% probability of winning. There was a crack w/the Princeton group who had Hillary at 99%, which I think showed how bad that model was. But 538's projection of 70% was pretty reasonable--and widely ill-understood, a lot of people were treating "70% chance to win" the same way you'd treat "70% polling" i.e.--70% polling you are winning in a hilarious landslide. In the 538 model, 70% to win literally means in all their simulation runs, Trump was winning in almost a third of them. Treating that like a guaranteed HRC victory was really dumb, but was largely the fault of people who didn't understand the basic concept behind the various models.

[crazy canuck]

I think there is a lot to that.   I took the bet based on the fact Silver had not ruled him out, so I had a long shot with a shot.  But, we are talking about what people were thinking the result would be, not what they should have been thinking/concerned about if they had understood the probabilities correctly.  And that is why people were shocked.  All the polls predicted a Clinton victory - including Silver.

What all the pollsters missed, including Silver, is that their data sets were not complete because of the phenomenon you have described and so they missed the support Trump actually had in key states.

[alfred russel]

What all the pollsters missed, including Silver, is that their data sets were not complete because of the phenomenon you have described and so they missed the support Trump actually had in key states.

Silver isn't a pollster and I don't think he missed anything.

If Silver projects 100 random races each with a confidence of 70%, if his expected candidate wins all 100 he has a deeply flawed model. His expected candidate should lose 30% of the time.

Polls have errors, and the errors can be linked in a cycle. For example, if the gold standard of polling in a given cycle misses a chunk of voters, and those voters are predisposed to be Trump voters, all the high quality polls may not only be wrong but also all independently understate Trump support. Silver's models do take this potential into account. That is a major source of uncertainty.

[alfred russel]

To get super nerdy, the reason some prognosticators gave Hillary a 99% chance to win in 2016 was that they did not consider that polling errors would be linked across outlets and states.

For example, we all know that if a single poll says Trump is down 4% that is prone to error and with a 4% margin of error he may be anywhere from even to down 8%. If all the polls say he is down 4%, that polling error is reduced--the stats get trickier but some folks would say a reasonable probability would be Trump is probably down 2% to 6%.

But then what really threw people off is all the state polling in critical states showed Trump was down.

Silver's model holds open that error in a single poll in a single state is replicated generally in polling across the country. Some stupid people didn't hold that possibility, and thought that all the polling in one state showing Trump down ~4% meant there was a minimal chance he would win that state, and when he needed to win say Florida, Pennsylvania and Wisconsin and the odds were 10%, 5% and 3%, the odds of all three happening were way under 1%. Silver was like, dude, the chance he wins Florida is probably more like 40%, and if he wins Florida, the odds he wins Pennsylvania is like 70%, and if he wins Florida and Pennsylvania, the odds he wins Wisconsin is like 90%.

[crazy canuck]

That is wading too far into the weeds for the simple point I was making to Valmy, that the polls were all predicting a Clinton win.

[alfred russel]

Nate Silver is my god and i can't tolerate him being blasphemed.

[Sheilbh]

That is wading too far into the weeds for the simple point I was making to Valmy, that the polls were all predicting a Clinton win.

And more crucially I think 2016 was proabbly the highpoint for "decent Republicans will flee and he'll drive people away". All the Never Trumpers (of whom some have gone back), National Review disowning etc.

In fact I think decent Republicans did flee more in 2018 and 2020 - but I think the talk about it was far more prominent in 2016 while 2018/20 were more around the die-hards/shoot someone on 5th Avenue levels of support. Not the first time discourse has been wildly disconnected from what actually happens

[Admiral Yi]

Nate Silver is my god and i can't tolerate him being blasphemed.

You actuaries are a protective group.

[crazy canuck]

That is wading too far into the weeds for the simple point I was making to Valmy, that the polls were all predicting a Clinton win.

And more crucially I think 2016 was proabbly the highpoint for "decent Republicans will flee and he'll drive people away". All the Never Trumpers (of whom some have gone back), National Review disowning etc.

In fact I think decent Republicans did flee more in 2018 and 2020 - but I think the talk about it was far more prominent in 2016 while 2018/20 were more around the die-hards/shoot someone on 5th Avenue levels of support. Not the first time discourse has been wildly disconnected from what actually happens

Yeah, in 2016 there was talk of the party turning on him, even after he won the election.  Nobody saw him turning the party into what it is today.

[Sheilbh]

Nate Silver is my god and i can't tolerate him being blasphemed.

You actuaries are a protective group.

Yeah I've no idea what's the appeal of Nate Silver - never quite understood it

Although I did love it when he tried to do British politics got it catastrophically wrong and never tried again

[DGuller]

The appeal of Nate Silver is that he upped the sophistication of election predictions by a huge margin.  Maybe he didn't have much success in Britain, but in the US he brought good science to a field filled with quackery.  His models later on also allowed him to have a real-time progress bar of the upcoming elections, which tends to feed the addiction of those who are already obsessed with elections.

[Valmy]

That is wading too far into the weeds for the simple point I was making to Valmy, that the polls were all predicting a Clinton win.

I was there. I was reading 538 obsessively like everybody else and he was saying it was really close in several key states and only have Hillary a 70% chance which in his metrics is "slightly favored". She was favored but even in a Clinton victory it was going to be close. That in itself was very worrying.

I have this recollection she got a little separation after the Democratic convention, then it tightened up the rest of the way. And then that FBI stuff right at the end. It was nerve wracking as fuck.

Both of what we are saying is true CC. Sure the polls predicted a Clinton win but the margins in many key states were razor thin, which is what gave Trump his punchers chance. And the way he did win was predicted by the polls so far as I saw it, it was just not the most likely outcome.

[The Minsky Moment]

The appeal of Nate Silver is that he upped the sophistication of election predictions by a huge margin.  Maybe he didn't have much success in Britain, but in the US he brought good science to a field filled with quackery.  His models later on also allowed him to have a real-time progress bar of the upcoming elections, which tends to feed the addiction of those who are already obsessed with elections.

I agree with this.  It's not that the 538 model was so extraordinarily good, it was that it was rationally thought out as compared to the competition.  The fact that many prominent election models were just assuming that polling errors across states were entirely independent and uncorrelated was just nuts; that's the kind of mistake that undergrads are taught to catch.

Silver started out as a baseball analyst which I think was an advantage. The immediate consumer of election models was (and is) political journalists - and political journalists as a class tend to be the kind of people that you can just wave some credentials at ("Princeton model"), and throw out a bunch of math and jargon and they will just soak it up uncritically.  Especially if the result accords with their own preferences or prejudices.

Sports analysts don't have it that easy.  If you are going to question Derek Jeter's fielding or claim that Craig Biggio was better than Ken Griffey Jr., talking up Ivy League credentials is just going to piss people off.  You have to spell out your models and assumptions very carefully. In a political model, accounting for possible error correlation across states adds work burden and forces you into the difficult an easily contestable task of specifying the likely degree to that correlation.  But for a sports analyst used to that kind of heat, it's just part of the job.