2016 Postmortem
Related: About this forumIt's a bit too early to rely on Nate Silver
Nate introduced his model in March 2008. A lot had transpired between 2007 and March. His model would have predicted Hillary to win, then flipped to Obama after early results. In other words, his model never captured the groundswell that propelled Obama to victory. What's 100% certain, Nate would have picked Hillary for a long time, then switched to Obama. Since we never see the flip, we believe his model is infallible.
I said this a couple of times, Bernie's greatest obstacle is that he's unknown. He really needs to spend a month in SC. Interestingly, I just read that that's what inspired Silver to create his model; the polls are off because nobody knows who Obama is.
What you heard on television was, Hillary was inevitable, shes up 20 points, he said. Shes up 20 points because people had heard of her. They hadnt heard of Obama....
In March, he introduced FiveThirtyEight.com, and it quickly became a go-to site for readers whose interest in raw numbers...
http://mobile.nytimes.com/2008/11/10/business/media/10silver.html?referer=

gcomeau
(5,764 posts)While he may write opinion on who he thinks will win future contests his numbers are always about the situation right *now*. Who would win if any given vote was held *now* based on existing polling and known polling biases and such in any given contest.
They are always subject to change up until the vote actually happens.
WhaTHellsgoingonhere
(5,252 posts)I just wanted to explain why he could be wrong 3 months ahead of a primary but predict it correctly later.
Dawgs
(14,755 posts)Anyone can go to realclearpolitics and get the same information. The polls are actually pretty accurate if you look at the average.
WhaTHellsgoingonhere
(5,252 posts)But he didn't make that conclusion until after some early results. I specifically quoted the salient information from the article for a reason. It answers your questions. If things go according to last 2008, Nate will dump Hillary by March 1. Then he'll have to explain that his model failed because no one knew who Bernie Sanders was, "It's identical to 2008, but my model doesn't capture this phenomenon. I didn't know that, because it didn't go live until March 2008."
It's more likely that he knows this but has no reason to divulge this information until his model fails.
Dawgs
(14,755 posts)I'm amazed that so many have bought what he's selling.
gcomeau
(5,764 posts)Dawgs
(14,755 posts)Copy and paste from the RCP site would have been enough.
Maybe the buyers are just simple.
WhaTHellsgoingonhere
(5,252 posts)I had to prove Nate isn't exactly infallible as he's being portrayed.
That's all.
mythology
(9,527 posts)But sure it's just luck that he got it so much closer than others in multiple elections than other pollsters.
Dawgs
(14,755 posts)His numbers are no better than the average of polls.
And, he didn't predict Florida in the 2012 election because it was so close. Others pollsters did, which absolutely made them more accurate.
gcomeau
(5,764 posts)WhaTHellsgoingonhere
(5,252 posts)is like jumping on the bandwagon. His model wouldn't have spit out Obama on, whatever, December 1, 2007.
It's really, really easy to understand.
Dawgs
(14,755 posts)The best I can say about Silver is that he's good at creating a website.
gcomeau
(5,764 posts)He's been *very* good. Occasionally others beat him out, but not by a lot.
The realclearpolitics average of state polls were got the exact same result in 2012. Coincidentally they were 50/51 in 2008; exactly the same as Silver.
The point is that you don't have to predict anything on election day...just look at the average of polls.
gcomeau
(5,764 posts)X/Y contests correct.
Dawgs
(14,755 posts)Silver adjusts his analysis daily based on polls. In the end his prediction will be as close as anyone that looks at the poll averages.
It's really not that impressive.
gcomeau
(5,764 posts)No, he does not just average the polls. And no, as the analysis rather clearly showed, he was not "as close as everyone". He was closer than most.
Dawgs
(14,755 posts)The point is that ALL he needs to is average the polls. He would come up with the exact same result as his super special analysis.
gcomeau
(5,764 posts)No. He. Would. Not. Come. Up. With. The. Exact. Same. Results.
He does not just average the polls. They are individually weighted according to accuracy, reliability, and demographic influences. And that has, as the figures clearly show, resulted in a more accurate analysis than just simply throwing all the poll numbers together and averaging them.
Another analysis... this one *specifically* comparing 538 to RCP... same conclusion:
https://www.informs.org/ORMS-Today/Public-Articles/February-Volume-40-Number-1/Did-Nate-Silver-beat-the-tortoise
[img][/img]
Dawgs
(14,755 posts)Because I can predict the exact same thing with just the average of polls. Analysis and statistics would still produce the same result as RCP.
Difference is that RCP and common sense only takes about five minutes.
gcomeau
(5,764 posts)I JUST SHOWED YOU that using the average of the polls gave worse results.
Here let's walk you through it.
This number: 1.45% ..... Is smaller than this number: 2.62%
And this number: 2.71% ..... Is smaller than this number: 3.01%
Those numbers are the average prediction error.
The smaller number is better.
538 had the smaller number. In both sets.
RCP average of polls had the larger number. In both sets.
Now, take it home with me! That means..... 538 did better and the RCP average of polls did worse. Not the same. Worse.
Dawgs
(14,755 posts)Not with ONLY the RCP average, but also a little common sense.
See. I know numbers and how to read. You?
gcomeau
(5,764 posts)Or, just keep shooting your mouth off on subjects you clearly do not understand very well. You know, either way.
Dawgs
(14,755 posts)Yes or no?
gcomeau
(5,764 posts)It is about significantly more than the binary win/loss prediction.
Dawgs
(14,755 posts)If RCP the average is just as accurate at picking the winner then why should anyone care.
And, why should anyone care that Silver is a little bit more accurate?
gcomeau
(5,764 posts)It is brain meltingly obvious why people care about modelling accuracy. In one given contest an inferior model may manage to pick as many winners and losers as a superior model but over the long term the more accurate model will outperform.
If you're picking team members to go to the Olympics to compete in archery do you just say anyone who hits the target is equally good.... or do you pay attention to how close to the fucking bullseye they are?
ram2008
(1,238 posts)He needs to stop pretending to be a good pundit because he is not. He said Bernie and Trump had zero chance even when it was clear they had a path and now they're undeniably building steam. No more punditry for Nate.
What he is good at doing is breaking down numbers in polls and measuring their accuracy, the "now" as you said. He needs to go back to his roots.
Response to ram2008 (Reply #9)
CobaltBlue This message was self-deleted by its author.
brooklynite
(96,882 posts)He's been campaigning for 9 months.
He's got millions of campaign supporters.
He's got money for campaign ads.
He's been in five debates.
He's given two election night speeches.
How long will it take him to be "known"?
Dawgs
(14,755 posts)WhaTHellsgoingonhere
(5,252 posts)I was just writing this for those who think of his polling as the gold standard.
Clearly, no one here does.
Dawgs
(14,755 posts)May be Silver himself. Who knows.
Response to WhaTHellsgoingonhere (Original post)
CobaltBlue This message was self-deleted by its author.