|
Read the rules in the OP before posting, please.In order to ensure that this thread continues to meet TL standards and follows the proper guidelines, we will be enforcing the rules in the OP more strictly. Be sure to give them a re-read to refresh your memory! The vast majority of you are contributing in a healthy way, keep it up! NOTE: When providing a source, explain why you feel it is relevant and what purpose it adds to the discussion if it's not obvious. Also take note that unsubstantiated tweets/posts meant only to rekindle old arguments can result in a mod action. |
On August 24 2017 05:51 Nebuchad wrote:Show nested quote +On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect.
http://www.npr.org/2017/05/05/526936636/pollsters-find-at-best-mixed-evidence-comey-letter-swayed-election
State polls were "historically bad"
By far, more Americans believed Clinton would win than Donald Trump. Ahead of the election, half of Americans thought she would, per one Economist/YouGov poll, compared with only 27 percent who believed it would be Trump. Forecasting models doubtless contributed to that belief for at least some voters. Predictions from some of the most popular models (FiveThirtyEight and the New York Times' Upshot, for example) ranged from giving Clinton a 71 percent to 99 percent chance of winning.
So when she lost, everyone (including NPR) tried to answer the question: Why did polls — and, therefore, forecasting models — so often point to a Clinton win?
First of all, only some polls were off, and it wasn't the national polls. Clinton won the popular vote by 2.1 percentage points, and polls had her winning the popular vote by an average of 3 points. That's not much of a gap at all, compared with past presidential polling.
But state polls were off by an average of 5 points, the largest average since 2000. This is where the researchers drilled down into the whys of what went wrong.
National polls were fine, but the state to state polling was off by quite a lot.
|
United States43219 Posts
On August 24 2017 05:53 Nyxisto wrote:Show nested quote +On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate. Of course it's easy now to say that it wasn't, but that is only a fair criticism if you could have known beforehand.
That's kind of idiotic though because it assumes they're unrelated events. Fine with dice rolls using a fair die, you can absolutely calculate the odds of 50 6s in a row. But for elections you should assume that the outcomes are related and that if an unlikely outcome happens in one state then it is likely that similar outcomes will happen in similar states.
He ought to have known that beforehand. Anyone could tell you that beforehand.
|
On August 24 2017 05:50 KwarK wrote:Show nested quote +On August 24 2017 05:49 GreenHorizons wrote:
Plenty of us on the left were warning for months that Trump could beat Hillary and the "it's all over but the crying" crowd were overestimating Hillary's support and underestimating Trump's. I prefer to think that they were overestimating the American public. Victims of their own belief in America being a better place than it turned out to be.
Turns out the "vote for the one who will only maim you" slogan was not as effective as many had hoped.
|
On August 24 2017 05:57 KwarK wrote:Show nested quote +On August 24 2017 05:53 Nyxisto wrote:On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate. Of course it's easy now to say that it wasn't, but that is only a fair criticism if you could have known beforehand. That's kind of idiotic though because it assumes they're unrelated events. Fine with dice rolls using a fair die, you can absolutely calculate the odds of 50 6s in a row. But for elections you should assume that the outcomes are related and that if an unlikely outcome happens in one state then it is likely that similar outcomes will happen in similar states. He ought to have known that beforehand. Anyone could tell you that beforehand.
Don't underestimate the number of idiots who aren't discovered because their dumb thoughts and assumptions are hidden in lots of math. If more people understood that, the world would be a significantly better place. This applies to academics, forecasters, hedge funds, and everyone else who results don't replicate, predictions don't accurately predict, or quant work that doesn't consistently beat the market (i.e. most).
Sam Wang is far from the only perpetrator of that crime.
Btw, I'm not directing this at you but just at potentially unaware readers of this thread in general.
|
On August 24 2017 05:53 Nyxisto wrote:Show nested quote +On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate.
This can already be criticized though. If something causes you to lose support in Pennsylvania it's likely to also go wrong in Michigan and Wisconsin cause those are (broadly) similar states. Very schematically you could lose three states as a result of a single thing going wrong, and just looking at aggregates won't give you that.
(sniped I guess :/)
|
|
|
People have been saying that for quite a while???
|
Saying it had nothing to do with it is just as stupid as saying it was the only thing.
|
On August 24 2017 05:53 mozoku wrote:Show nested quote +On August 24 2017 05:44 zlefin wrote:On August 24 2017 05:40 mozoku wrote:On August 24 2017 05:35 zlefin wrote:On August 24 2017 05:10 mozoku wrote:
I'm pretty surprised that so many people are defending the journalism status quo. I think the current echo chamber flavor of journalism is quite self-evidently destructive for society. That isn't really a partisan issue at all.
Nor is it necessarily journalists' fault, but the system is clearly broken imo. And a lot of what oBlade said is fairly true (journalists wield a massive amount of power in society by nature of their position, there's no real implicit or explicit checks on them to ensure they use it vaguely in society's interest, and journalism is arguably the largest contributor to the current federal government's dysfunction).
I'd love to see some public debate on how to fix journalism. journalism doesn't seem remotely like the largest contributor to the current federal gov't dysfunction. I see no reason to claim that; far simpler to say that the politicians themselves are responsible for that, throguh their own choices and actions. Isn't that putting the cart before the horse? Politicians act the way they do because they are (rationally) obsessed with their public image (i.e. what gets through the news media filter) due to election survivorship bias. If you want to change politicians' behavior, you either have to change elections or change the filter. I'm not comfortable switching to a CCP-style authoritarian government, so changing the way the news media operates seems like the better plan. that's not putting the cart before the horse. the politicians were responsible for SETTING UP the system; for setting up the elections, and for setting up rules in whcih the meida operates. If the media is only rationally acting in their own interest, why hate on them rathre than the politicians? politicians also have many ways to get throug hthe news media filter. and how do you change the filter without violating the first amendment? what you're proposing is FAR more CCP style than changes to the election system. and there's plenty of ways to alter elections, or the system in general, that could potentially work. there's of course also alot of flaws in the fundamental design of democracy itself. PS I'd added some responses in my previous post while you were typing up your reply. I never said it was the news media organizations' fault. I expect them to be profit-driven. All I meant was that I'd love to see public discussion on changes to media, which would then hopefully produce ideas to fix the clearly broken news media status quo, and in turn pressure politicians to change said status quo. Fwiw, I think a lot of old school journalists hate the news media status quo as much as I do. There's been a lot of complaints from them looking to strengthen their negotiating leverage with Facebook/Google and reduce their sensitivity to click-driven traffic. I'm fairly sympathetic to those complaints, though not informed enough to make strong judgments on the matter.
your quote chain is pretty clearly putting a lot of blame on the news media orgs, at least as compared to politicians. maybe that was not your intent, but that's very much how it reads. you might wanna look back through it and reread, I just did.
public discussion isn't likely to yield intelligent fixes, because the public is dumb, and moreover, the public doesn't have a good understanding of the topic, and will never be able to due to the inherent limitations on information and expertise (too many thing sto know). they've also proven they don't have interest in actual policy debate (proven by the last election).
from waht I've read about the topic, it's a result of structural changes in the economy for which there is no truly good solution. that and of course it's always had ups and downs in quality anyways, we've just been coming out of one of the higher points so the loss is more apparent.
|
|
|
On August 24 2017 06:07 Mohdoo wrote:Saying it had nothing to do with it is just as stupid as saying it was the only thing.
Of course it had something to do with it, I don't think anyone denies that. Just people like to focus on the writing of the stories and not the "extremely careless" way she handled them or the incessant lying about it, or they conflate her emails with the DNC and Podesta and dismiss them all and focus not on what they revealed, but how they were obtained.
Hillary, the DNC, and their supporters were just wrong.
|
On August 24 2017 05:57 KwarK wrote:Show nested quote +On August 24 2017 05:53 Nyxisto wrote:On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate. Of course it's easy now to say that it wasn't, but that is only a fair criticism if you could have known beforehand. That's kind of idiotic though because it assumes they're unrelated events. Fine with dice rolls using a fair die, you can absolutely calculate the odds of 50 6s in a row. But for elections you should assume that the outcomes are related and that if an unlikely outcome happens in one state then it is likely that similar outcomes will happen in similar states. He ought to have known that beforehand. Anyone could tell you that beforehand.
The problem is that the interconnection between states is something that is very hard to quantify, and putting subjective assumptions into your model makes you look smart if they work but it makes you look stupid if they don't. Given that historically relying on a pure polling based model was accurate, Wang continued to use it. (LL points to this in his post, the 'frequentist' approach)
It's always easy to justify your scepticism about the data if an unlikely event does indeed happen. So now he should definitely rely less on data alone. But it's a much harder call to make before you can update your beliefs about the accuracy of your model.
|
On August 24 2017 06:15 Nyxisto wrote:Show nested quote +On August 24 2017 05:57 KwarK wrote:On August 24 2017 05:53 Nyxisto wrote:On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate. Of course it's easy now to say that it wasn't, but that is only a fair criticism if you could have known beforehand. That's kind of idiotic though because it assumes they're unrelated events. Fine with dice rolls using a fair die, you can absolutely calculate the odds of 50 6s in a row. But for elections you should assume that the outcomes are related and that if an unlikely outcome happens in one state then it is likely that similar outcomes will happen in similar states. He ought to have known that beforehand. Anyone could tell you that beforehand. The problem is that the interconnection between states is something that is very hard to quantify, and putting subjective assumptions into your model makes you look smart if they work but it makes you look stupid if they don't. Given that historically relying on a pure polling based model was accurate, Wang continued to use it. (LL points to this in his post, the 'frequentist' approach) It's always easy to justify your scepticism about the data if an unlikely event does indeed happen. So now he should definitely rely less on data alone. But it's a much harder call to make before you can update your beliefs about the accuracy of your model.
I'm pretty sure that polling errors being correlated across different states was not a new thing in the 2016 election. At all. To assume that it was was not "being objective", it was just being stupid.
|
United States43219 Posts
On August 24 2017 06:19 mozoku wrote:Show nested quote +On August 24 2017 06:15 Nyxisto wrote:On August 24 2017 05:57 KwarK wrote:On August 24 2017 05:53 Nyxisto wrote:On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate. Of course it's easy now to say that it wasn't, but that is only a fair criticism if you could have known beforehand. That's kind of idiotic though because it assumes they're unrelated events. Fine with dice rolls using a fair die, you can absolutely calculate the odds of 50 6s in a row. But for elections you should assume that the outcomes are related and that if an unlikely outcome happens in one state then it is likely that similar outcomes will happen in similar states. He ought to have known that beforehand. Anyone could tell you that beforehand. The problem is that the interconnection between states is something that is very hard to quantify, and putting subjective assumptions into your model makes you look smart if they work but it makes you look stupid if they don't. Given that historically relying on a pure polling based model was accurate, Wang continued to use it. (LL points to this in his post, the 'frequentist' approach) It's always easy to justify your scepticism about the data if an unlikely event does indeed happen. So now he should definitely rely less on data alone. But it's a much harder call to make before you can update your beliefs about the accuracy of your model. I'm pretty sure that polling errors being correlated across different states was not a new thing in the 2016 election. At all. To assume that it was was not "being objective", it was just being stupid.
This. It being 50 unrelated races with no correlation is a far greater stretch than there being some, even if only weak, correlation between people on either side of a state line. The independent races model asks us to believe that if you poll two extremely similar towns on either side of a state line, with the same industries, demographics, politics and so forth, a polling error would not be replicated.
|
On August 24 2017 06:19 mozoku wrote:Show nested quote +On August 24 2017 06:15 Nyxisto wrote:On August 24 2017 05:57 KwarK wrote:On August 24 2017 05:53 Nyxisto wrote:On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate. Of course it's easy now to say that it wasn't, but that is only a fair criticism if you could have known beforehand. That's kind of idiotic though because it assumes they're unrelated events. Fine with dice rolls using a fair die, you can absolutely calculate the odds of 50 6s in a row. But for elections you should assume that the outcomes are related and that if an unlikely outcome happens in one state then it is likely that similar outcomes will happen in similar states. He ought to have known that beforehand. Anyone could tell you that beforehand. The problem is that the interconnection between states is something that is very hard to quantify, and putting subjective assumptions into your model makes you look smart if they work but it makes you look stupid if they don't. Given that historically relying on a pure polling based model was accurate, Wang continued to use it. (LL points to this in his post, the 'frequentist' approach) It's always easy to justify your scepticism about the data if an unlikely event does indeed happen. So now he should definitely rely less on data alone. But it's a much harder call to make before you can update your beliefs about the accuracy of your model. I'm pretty sure that polling errors being correlated across different states was not a new thing in the 2016 election. At all. this was pretty much the meat of the contention between Nate Silver and Sam Wang in 2016.
Sam Wang is a professor of neuroscience, making him not really an authority on statistics.
This makes for some interesting reading in retrospect :
http://www.politico.com/story/2016/11/nate-silver-huffington-post-polls-twitter-230815
It looks like Silver included national polls in his model while most other models used State polls and ignored the national ones, helping to explain the massive overestimates.
|
On August 24 2017 06:19 mozoku wrote:Show nested quote +On August 24 2017 06:15 Nyxisto wrote:On August 24 2017 05:57 KwarK wrote:On August 24 2017 05:53 Nyxisto wrote:On August 24 2017 05:51 Nebuchad wrote:On August 24 2017 05:49 Nyxisto wrote:On August 24 2017 05:45 Nevuk wrote:
Sam Wang's 99.99% model was pretty bad. The NYT and Huffpost models were also very far from the mark. The polls were bad, the model was fine. The relevant question is whether anybody could have predicted that the polls were bad before we had the election results, which probably isn't the case. What makes you say the polls were bad and the model was fine? My impression is that it's incorrect. Sam wangs model was literally just aggregating the state polls and running a monte carlo simulation on them which gives you 99%+ chance of winning when both candidates are 3-4% percent apart in the polls. There literally wasn't more than this to it, it's the most minimal model that you can go with and it relies entirely on the fact that your polling data is accurate. Of course it's easy now to say that it wasn't, but that is only a fair criticism if you could have known beforehand. That's kind of idiotic though because it assumes they're unrelated events. Fine with dice rolls using a fair die, you can absolutely calculate the odds of 50 6s in a row. But for elections you should assume that the outcomes are related and that if an unlikely outcome happens in one state then it is likely that similar outcomes will happen in similar states. He ought to have known that beforehand. Anyone could tell you that beforehand. The problem is that the interconnection between states is something that is very hard to quantify, and putting subjective assumptions into your model makes you look smart if they work but it makes you look stupid if they don't. Given that historically relying on a pure polling based model was accurate, Wang continued to use it. (LL points to this in his post, the 'frequentist' approach) It's always easy to justify your scepticism about the data if an unlikely event does indeed happen. So now he should definitely rely less on data alone. But it's a much harder call to make before you can update your beliefs about the accuracy of your model. I'm pretty sure that polling errors being correlated across different states was not a new thing in the 2016 election. At all. To assume that it was was not "being objective", it was just being stupid.
The errors were significantly worse than in previous election, where the model performed well, Wang wasn't any less accurate than Silver in the last two elections. The problem with tuning your prediction too heavily to fit specific data is that your model will generalise badly. It might just be the case that Nate Silver overshoots with his assumptions and his predictions in four years is less accurate. People who like the 'frequentist' approach like Wang generally aim to make accurate predictions from lots of data over the long term.
|
On August 24 2017 06:41 Nevuk wrote:
Sam Wang is a professor of neuroscience, making him not really an authority on statistics.
New rule: let's not trust neuroscientists named Sam for anything other than neuroscience.
|
On August 24 2017 05:50 KwarK wrote:Show nested quote +On August 24 2017 05:49 GreenHorizons wrote:
Plenty of us on the left were warning for months that Trump could beat Hillary and the "it's all over but the crying" crowd were overestimating Hillary's support and underestimating Trump's. I prefer to think that they were overestimating the American public. Victims of their own belief in America being a better place than it turned out to be.
Pretty much this. Unfortunately there were enough tribalists and enough people willing to believe a huckster like Trump living in just the right places.
It's very easy to point at Clinton/ DNC and say, "behold, all the things wrong with the not-right!" But that conveniently makes it their fault, and not the fault of a much larger portion of America which contributed to the state we're in.
|
Donald Trump supporters have been sharing fake photos of huge crowds at the US President's Phoenix rally on social media, following claims the event failed to draw big numbers.
Images showing thousands of people lining the streets in Arizona were widely shared by Republicans and the leader's core base.
But the photos were quickly debunked as fake. They actually showed an aerial view of crowds that had turned out for a championship parade for the Cleveland Cavaliers basketball team.
Official turnout figures from the Phoenix convention centre rally have not been released, but many have speculated they could be lower than expected.
Images posted to social media appeared to show Trump supporters filling just half of the room.
Mr Trump tweeted following the campaign event: "Thank you Arizona. Beautiful turnout of 15,000 in Phoenix tonight!" http://www.independent.co.uk/news/world/americas/us-politics/trump-supporter-phoenix-rally-crowd-size-photos-fake-half-empty-room-us-president-a7907801.html?cmpid=facebook-post
|
On August 24 2017 07:09 ticklishmusic wrote:Show nested quote +On August 24 2017 05:50 KwarK wrote:On August 24 2017 05:49 GreenHorizons wrote:
Plenty of us on the left were warning for months that Trump could beat Hillary and the "it's all over but the crying" crowd were overestimating Hillary's support and underestimating Trump's. I prefer to think that they were overestimating the American public. Victims of their own belief in America being a better place than it turned out to be. Pretty much this. Unfortunately there were enough tribalists and enough people willing to believe a huckster like Trump living in just the right places. It's very easy to point at Clinton/ DNC and say, "behold, all the things wrong with the not-right!" But that conveniently makes it their fault, and not the fault of a much larger portion of America which contributed to the state we're in.
Fault? You mean, as in, there is a certain group of people with values that aren't just different, but damaging? Whoa there, buddy. As a member of the right, I am a FIRM believer in moral relativism and do not think you can say a set of beliefs are objectively better than another.
|
|
|
|
|
|
EPT
