This article is Nate Silver's take on the election modeling aspect of this election, including the reasons that made this a close election, and pretty solid arguments as to why Biden's win is far...
This article is Nate Silver's take on the election modeling aspect of this election, including the reasons that made this a close election, and pretty solid arguments as to why Biden's win is far more significant than it looks at first glance (especially during the initial counting).
Having seen my fair share of hatred, toxicity and bile spat out on twitter against NS and 538 especially the first day, I would say this is an important read.
I think the fairest criticism of 538 is that he models virtually every possible outcome, and thus is 'never' wrong with his forecasts. They'd carry a lot more weight if he said 'here's my...
I think the fairest criticism of 538 is that he models virtually every possible outcome, and thus is 'never' wrong with his forecasts. They'd carry a lot more weight if he said 'here's my predictions based on X Y Z'.
Biden was clearly the favorite to win. But what we've learned is that Trump's base is fervent and/or ignorant and that polling in the modern era is useless.
And Trump served his purpose: Packing the courts with conservative judges and the 'real bad' wishlist while giving the rest of the party plausible deniability.
What does that even mean? Yeah, the model models "virtually every outcome", and then you divide the number of outcomes where Biden wins by the number of total outcomes to get the probability. It's...
I think the fairest criticism of 538 is that he models virtually every possible outcome, and thus is 'never' wrong with his forecasts.
What does that even mean? Yeah, the model models "virtually every outcome", and then you divide the number of outcomes where Biden wins by the number of total outcomes to get the probability. It's monte carlo sampling.
It's like modeling the probability of a coin: 1 head, 1 tail. Probability of head = 1/2 = 50%. The basis of frequentist statistics.
If you have a normal distribution that you belief to be centered on 0, with a std dev of 1, and you run it, and get 0.7, just because it's not exactly 0 does not mean you're unconfident about the parameters.
For sure. I mean you could even see it play out the middle of this week. The paths for Trumps victory sure did involve a LOT of ANDs. He needed to win PA flat out, or instantly lost, AND he needed...
For sure. I mean you could even see it play out the middle of this week. The paths for Trumps victory sure did involve a LOT of ANDs. He needed to win PA flat out, or instantly lost, AND he needed to win like 3 of Georgia, Arizona, Nevada, North Carolina. He also had to win Ohio and Florida or he also instantly lost. Biden could win just PA, OR Arizona + Nevada OR Georgia + Arizona OR Georgia + Nevada by the end of like Wednesday.
It was a very hard election for Trump, hence the confidence.
It's likely a lot of people translate "90% chance to win" as: "Final vote share will be 90% Biden, 10% Trump". I think a lot of people don't even realize the difference between a model of "we...
It's likely a lot of people translate "90% chance to win" as: "Final vote share will be 90% Biden, 10% Trump".
I think a lot of people don't even realize the difference between a model of "we think Biden wins 9 out of 10 times" and "We interviewed people and 9 out of 10 told us they would vote Biden".
Yep. Everyone who takes predictions like 70% chance of Clinton win too seriously has never played xcom. A 95% chance to hit at point blank range still means you have a 5% chance of punishing the...
Yep. Everyone who takes predictions like 70% chance of Clinton win too seriously has never played xcom.
A 95% chance to hit at point blank range still means you have a 5% chance of punishing the pavement with great prejudice.
Call the surprising victories as they fell and look at the way the probabilities change. Now factor in the initial doubt because a lot of the votes hadn't been counted, and you can reconstruct the week.
But he does apply probabilities and expectation to the various outcomes, and can be called out on the degree to which his predictions did fail. Even on his own stream they were mentioning the...
But he does apply probabilities and expectation to the various outcomes, and can be called out on the degree to which his predictions did fail. Even on his own stream they were mentioning the places where their projections failed to match the outcomes. If a race was projected to be more than a tossup, there was a note each time the race went the other way.
How many outcomes where they actually wrong on? I only recall seeing 4-5 races across the Presidency and congress that were surprising based on their models.
How many outcomes where they actually wrong on? I only recall seeing 4-5 races across the Presidency and congress that were surprising based on their models.
IIRC from reading their live blog throughout the week, two of their Leaning D across Senate and Presidency were wrong, and all confident calls were correct. House fared worse though, but I didn't...
IIRC from reading their live blog throughout the week, two of their Leaning D across Senate and Presidency were wrong, and all confident calls were correct.
House fared worse though, but I didn't look at their numbers.
Florida. Though that was always going to be a close one. However: check out https://projects.fivethirtyeight.com/2020-election-forecast/ and you'll notice one major race that went against the...
Florida. Though that was always going to be a close one. However: check out https://projects.fivethirtyeight.com/2020-election-forecast/ and you'll notice one major race that went against the grain there and it's Florida. They gave Biden 2 in 3 odds there.
Yeah, they have mentioned that it seems like there may have been a larger than usual polling miss in Florida. I'm curious to see what they say about it once the dust has settled and they've really...
Yeah, they have mentioned that it seems like there may have been a larger than usual polling miss in Florida. I'm curious to see what they say about it once the dust has settled and they've really had time to dig in.
Clinton did very well with Hispanic voters in 2016, and nothing in polling or intuition expected that to be different. However, the reality was that Cubans and Venezuelans in Miami-Dade absolutely...
Clinton did very well with Hispanic voters in 2016, and nothing in polling or intuition expected that to be different.
However, the reality was that Cubans and Venezuelans in Miami-Dade absolutely tanked that district for Biden. Hispanic vote was like 70/30 for Trump in Miami-Dade. Turns out all you have to do is call someone a socialist and they get destroyed in Miami-Dade.
No this is not a convincing win. The article glosses over the fact that trump received more votes than Obama in ‘08. Yes I know there are more voters now than before but it is evidence for just...
No this is not a convincing win.
The article glosses over the fact that trump received more votes than Obama in ‘08. Yes I know there are more voters now than before but it is evidence for just how deeply flawed America’s democracy is.
How can someone that bungled coronavirus, was impeached, had an incredible level of staff turnover still only loose the popular vote by 4 points?
It shows that the electorate is susceptible to lies and propaganda. That media will gleefully repeat those lies to support their guy.
This article is Nate Silver's take on the election modeling aspect of this election, including the reasons that made this a close election, and pretty solid arguments as to why Biden's win is far more significant than it looks at first glance (especially during the initial counting).
Having seen my fair share of hatred, toxicity and bile spat out on twitter against NS and 538 especially the first day, I would say this is an important read.
I think the fairest criticism of 538 is that he models virtually every possible outcome, and thus is 'never' wrong with his forecasts. They'd carry a lot more weight if he said 'here's my predictions based on X Y Z'.
Biden was clearly the favorite to win. But what we've learned is that Trump's base is fervent and/or ignorant and that polling in the modern era is useless.
And Trump served his purpose: Packing the courts with conservative judges and the 'real bad' wishlist while giving the rest of the party plausible deniability.
What does that even mean? Yeah, the model models "virtually every outcome", and then you divide the number of outcomes where Biden wins by the number of total outcomes to get the probability. It's monte carlo sampling.
It's like modeling the probability of a coin: 1 head, 1 tail. Probability of head = 1/2 = 50%. The basis of frequentist statistics.
If you have a normal distribution that you belief to be centered on 0, with a std dev of 1, and you run it, and get 0.7, just because it's not exactly 0 does not mean you're unconfident about the parameters.
For sure. I mean you could even see it play out the middle of this week. The paths for Trumps victory sure did involve a LOT of ANDs. He needed to win PA flat out, or instantly lost, AND he needed to win like 3 of Georgia, Arizona, Nevada, North Carolina. He also had to win Ohio and Florida or he also instantly lost. Biden could win just PA, OR Arizona + Nevada OR Georgia + Arizona OR Georgia + Nevada by the end of like Wednesday.
It was a very hard election for Trump, hence the confidence.
It's likely a lot of people translate "90% chance to win" as: "Final vote share will be 90% Biden, 10% Trump".
I think a lot of people don't even realize the difference between a model of "we think Biden wins 9 out of 10 times" and "We interviewed people and 9 out of 10 told us they would vote Biden".
Yep. Everyone who takes predictions like 70% chance of Clinton win too seriously has never played xcom.
A 95% chance to hit at point blank range still means you have a 5% chance of punishing the pavement with great prejudice.
I can, by the way, recommend looking at 538's predictions and look at them again. Particularly their explore model: https://projects.fivethirtyeight.com/trump-biden-election-map/
Call the surprising victories as they fell and look at the way the probabilities change. Now factor in the initial doubt because a lot of the votes hadn't been counted, and you can reconstruct the week.
But he does apply probabilities and expectation to the various outcomes, and can be called out on the degree to which his predictions did fail. Even on his own stream they were mentioning the places where their projections failed to match the outcomes. If a race was projected to be more than a tossup, there was a note each time the race went the other way.
How many outcomes where they actually wrong on? I only recall seeing 4-5 races across the Presidency and congress that were surprising based on their models.
IIRC from reading their live blog throughout the week, two of their Leaning D across Senate and Presidency were wrong, and all confident calls were correct.
House fared worse though, but I didn't look at their numbers.
Florida. Though that was always going to be a close one. However: check out https://projects.fivethirtyeight.com/2020-election-forecast/ and you'll notice one major race that went against the grain there and it's Florida. They gave Biden 2 in 3 odds there.
Yeah, they have mentioned that it seems like there may have been a larger than usual polling miss in Florida. I'm curious to see what they say about it once the dust has settled and they've really had time to dig in.
Clinton did very well with Hispanic voters in 2016, and nothing in polling or intuition expected that to be different.
However, the reality was that Cubans and Venezuelans in Miami-Dade absolutely tanked that district for Biden. Hispanic vote was like 70/30 for Trump in Miami-Dade. Turns out all you have to do is call someone a socialist and they get destroyed in Miami-Dade.
Yeah, around that number, and I believe they were mostly ones that were only leaning one way or the other rather than the very confident races.
There is an explanation here of how they check their accuracy. It's not very intuitive, though.
No this is not a convincing win.
The article glosses over the fact that trump received more votes than Obama in ‘08. Yes I know there are more voters now than before but it is evidence for just how deeply flawed America’s democracy is.
How can someone that bungled coronavirus, was impeached, had an incredible level of staff turnover still only loose the popular vote by 4 points?
It shows that the electorate is susceptible to lies and propaganda. That media will gleefully repeat those lies to support their guy.