I liked reading 538.com during the election season as much as anyone, and still think it's an insightful site with a lot of interesting things to say. That said, I've been kind of surprised by the post-election surge of praise for Nate Silver. You would think, based on some of the commentary, that his innovative and complicated formula wound up giving us some incredible insight we couldn't have gotten from anywhere else.I call on Nate to do a YouTube and dis Matt. More blogger beef please.
The Beautiful Struggle: A Father, Two Sons, and an Unlikely Road to Manhood 
Copyright © 2009 by The Atlantic Monthly Group. All rights reserved.
This is much better than the Soulja Boy-Ice T beef. Seriously though I wonder what gives because I like reading Yglesias' blog but that came off as straight hater-ade from out of the blue. Did Nate take his girl?
BTW, didja hear about the NHL guy getting suspended for saying to a throng of reports that another NHL player fell in love with his "sloppy seconds"? Now thats BEEF!
Yglesias is mad that Nate Silver is more inflential than his book "Heads in the Sand."
Nate attracts so much blogger beef it's amazing. The most entertaining was between him and RealClearPolitics, but the Zogby stuff was pretty fun too.
Team Nate!!
I hope he slaps Yglesias back; there is so much incompetence, error and laziness in MY's work, he won't have to struggle to find points of attack.
Nerd fight! I can't wait to see one of them use a policy white paper as a folding chair.
Did Nate take his girl?
*ponders* You know, several blogs refer to Nate as "everyone's new secret boyfriend" and such; maybe he did. Or maybe Matt's jealous that he didn't make the "hottest 2008 bloggers" list.
In all seriousness, if the site wasn't offering something--predictions, outstanding analysis of different polls, the great series going to all the different campaign offices--that people couldn't find elsewhere at well-established polling sites, the site wouldn't have gone from 0 to top-rated in a few months.
I hope Nate rocks his AI jersey and tells Matt to "eat a di-k".
The point of 538 is polling data and something more.
The more is Nate Silver working his craft. He's building something worth having, calmly, carefully, and with skill. There's barely any sturm-and-drang, but considerable satisfaction as something sturdy comes into being. There's some anger, but it's turned into muscle rather than noise.
The shape of this year, for me, will be the sudden sense that I could feel that all around me. There's Nate, and there's Sean Quinn and Brett Marty on his site and there are the campaign workers Sean and Brett described. There's Rachel Maddow quoting Winthrop and saying "infrastructure" with delight. And, of course, there's Barack Obama taking the prize.
And then, there's the guy who wrote "Time spent worrying about some fools who you can't control, is time away from improving your chosen craft." (TNC, I keep going back to that line, because it carries such a tight statement of a good way to work and be.)
Silver isn't just about numbers. He's about the craft of sharing numbers.
The incredible value that 538 provided during the election (not just Nate's formula and commentary, but also Sean's reporting from the road) was a long-term perspective and a way to vicariously tap into the preternatural calm of the Obama campaign. And Nate gave the bad news for Ds (where it occured) with the good, so no inflation of hopes.
It was a stark contrast to the bulk of left/liberal/progressive/Democratic/whatever bloggers who got caught up in waves of "OMG! OMG! OMG! Obama is being too cocky! Obama is being too conciliatory! Obama is losing women! Obama is losing men! OMG! OMG! OMG!" sprinkled with occasional "Landslide! The Republican Party will cease to exist on November 5!" posting.
Eh, I'm kind of with Yglesias on this one. I mean, he had this crazy statistical model that he changed whenever he felt like it wasn't giving him the right answer. And at the end of the day his predictions were less accurate than just averaging all the polls together. I still don't get why everybody got so excited about Nate's ad hoc modeling.
And, to Deborah, the pictures of the polling sites were due to 538 co-blogger Sean Quinn and his friend Brett Marty, not Nate Silver.
I need to get out my webcam so I can call Nate out.
Nate and Matt should go on blogging heads and nerd da phuck out!
I would definitely watch that for free on youtube!
Nate and Matt should go on blogging heads and nerd da phuck out!
I would definitely watch that for free on youtube!
BWA HA HA HA HA HA HA
The visual of that is hilarious.
thanks sp. i appreciate that.
This is simple jealousy. Or what some might refer to as 'hating.'
MY's blog has turned into a boring echo chamber. He's as partisan as Hannity, and less entertaining. The fact that Sullivan has an award named after him is very, very puzzling. Maybe he used to be better. I've only read him from the beginning of '08.
MY does get pretty lazy sometimes, and just reaches for his gut reaction more often than he should. E.G. His posts on the auto bailout. If he read anything that has been put out their about what is happening in the auto industry, or had any understanding of its structure, he wouldn't keep wondering why GM going under would take out Ford or maybe even Toyota.
Silver's method provided the intellectual backbone - used by many bloggers like Yglesias - for rebuttal against flawed network polling and helped push incompetent douchebags like Michael Barone towards obscurity. Blumenthal deserves credit, too.
I think you all are being to hard on MY. Sure he's a bad speller, has bizarre love of congestion pricing, and is really too young to be much of an authority on anything, but he's always an interesting read and he is well-read and well-sourced.
As for new beefs, I am waiting for the inevitable beef between Sullivan and TNC over black homophobia, the Bell Curve, amongst other topics.
@br
"I mean, he had this crazy statistical model that he changed whenever he felt like it wasn't giving him the right answer"....
what? even if this was true...that he did infact "change" the model...i still can't understand how thats wrong in a statistical sense...any given statistical model, when dealing with a data-source that changes in time, has to take into account that change, if it's supposed to be viable...it's the basis of linear regression in time.
I dont of course know what u mean by "changing the model"...im only assuming you mean recalculating the weights of the model to fit the new data. even rasmussen gave him his props by starting to use his model in one of their polls....
and br....i haven't really seen any evidence that
"his predictions were less accurate than just averaging all the polls together"...pls give me a link for this...as i would like to see it for my self. The reason for the "crazy statistical model" is that it gives a better prediction over time and specific polls, because of the fact that different polls have different qualities...std's, sizes, quality in how they were conducted...it's not something he thought of himself...it's a something known in most statistical modeling books...atleast those i've read...
i don't know the reason for MY's hostility towards 538, but it seems to be unfounded and based on some kool-aid he drank back in the earlier days of the election....maybe he thought MS's model could project that he would be more popular than he is...i don't know...
There is a reason why Sullivan has an Ygelsias award.
Matt's points are debatable, but part of that means they are plausible, and I dont really think they should be dismissed as simple hating.
I mean, I loved Nate's blog (particularly the state-by-state on site analysis), but averaging the polls in the aggregate was also extraordinarily accurate (possibly more so)
But similarly, Nate's efforts, with his steady demeanor and calm clear-eyed analysis, should not be dismissed either. I dont think Matt meant to, but perhaps it came off that way.
"There is a reason why Sullivan has an Ygelsias award."
Well, I'm aware of what the award is. I just don't think that MY represents the spirit of the Yglesias award in any way. He's terribly, terribly partisan. Not saying I don't agree with him the majority of the time, but the idea of him having an honorary award for speaking out against his party doesn't seem right to me. Unless it stems from one issue in particular that I'm not aware of. Or maybe he was just worse during the election cycle. I don't really read him much anymore.
BTW, didja hear about the NHL guy getting suspended for saying to a throng of reports that another NHL player fell in love with his "sloppy seconds"? Now thats BEEF!
I heard that on the way in, it was Sean Avery talking about that hot actress in 24 and Old School (Elisha Cuthb????) While it was a bit crass, I would consider it a suspendable offense.
DougEMI
I wonder if they did it because Avery was going to play the other guy's team soon or something. Me being the azzhole I am, would have promoted it like a fight over a girl back in high school! By the way the NHL does seem to have a history of passing chicks around. Didn't Anna Kournikova (sp) put in work with like 3 different NHL players?
I thought what MY said was right, but mostly irrelevant to Nate Silver's popularity. I'm largely concerned, though, that this could be the start of lefty-coast blogger violence resulting in the loss of some of my favorites. Stay clear of this drama, Mr. Coates.
sgwhiteinfla,
I was going to bring up the puck bunnie thing myself. Kournikova, I know was with Federov and Bure, and I do think she was with at least one more.
I saw some hockey players at the bar last spring, they were ugly MFers, but had the women flocking all over them.
I know in the past the NHL has suspended guys for specific games against a given team after a cheap shot or ugly hit where there might be retaliation.
Link for bruce to the summary by the princeton election consortium. Also, dumb averaging by pollster.com miraculously got ev count correct.
Nate falls into the same trap that all of these statistical sports forecasters do... their model is the best thing since sliced bread until it tells them something wrong, and then they tweak it until it tells them something wrong again. And nobody gets the answer right forever.
The problem with his approach is not that he would change his outcome predictions based on new data, it was that he would change the internals of his prediction algorithm to overfit the new data. Complex models with lots of tunable parameters can be adjusted to fit any past behavior. It's predicting the future that's hard, and I'm saying that Nate didn't do a particularly good job.
Beef is not what Yglesias said to Silver;
beef is when working bloggers can't get jobs.
"It's predicting the future that's hard, and I'm saying that Nate didn't do a particularly good job."
Br,
Nate correctly predicted the presidential winner in 49 out of 50 states, as did the Princeton Election consortium. (The District of Columbia was a given for Obama, and Nate didn't need his model to predict that.) The state he got wrong was Indiana, the same state that the Princeton Election consortium picked incorrectly also. So your assertion that Nate didn't do a particularly good job of predicting the outcome of the presidential race is not supported by the facts.
BTW, Pollster.Com correctly 48 out of the 50 states. The 2 states they got wrong were Missouri and Indiana (Missouri for Obama, Indiana for McCain). However, since both states have 11 electoral votes, Pollster's electoral vote projection was only off by 1. (Pollster, like Nate and the Princeton Election consortium, failed to predict Obama's win in one of Nebraska's congressional districts.)
eltoro,
Fair enough. I still think it's fair to say that he didn't do better than vanilla averaging of the polls.
@br, 538 was also more accurate during the primaries. which in a sense is harder to predict.
i think matt y is getting way too much crap here. i read the whole post & it wasnt hating on nate. yeah he threw a wet blanket on nate's accuracy but he isn't wrong. he also gave nate props for being right during the primary.
Silver adjusts his model to fit the data as it comes in. Since the availability of data was poor and is improving, it makes sense to revise the model as conditions change.
Coming from a baseball fan's background, I'm familiar with Silver's work going back to PECOTA. Contrarians were out in full force discounting his work in that field, as well. Now, PECOTA projections are unilaterally regarded as the best in the field, although not perfect, as you rightly point out.
Also, for what its worth, pollster doesn't do "dumb averaging". They don't weight the polls against each other, but they *do* weight the polls by their age and sample size. The average is the result of a regression analysis, which is already halfway to what Nate Silver's model does.
What Silver's model does (that the others don't do) is account for national (in addition to state) polling and attribute weighting to account for pollsters that are more accurate than others (i.e. Rasmussen wipes his ass with Zogby's face).
Where is your dispute, then? Reason it out.
back from a sparring set! Joel, u just beat me to it...Silvers regression technique is not a major lift from the other regression models i've seen... it's more that his background creates this hype around his baseballesque polling-technique...
as for the PEC-polls...they are quite sound and pretty nifty...
no, they aren't based on a "sophisticated magical model" like NS's...but they are definetly not as simple and obtuse as BR would like you to think...i took a look at the 4 matlab-files he uses in the polls, and they are quite sophisticated...i could go over them for u, but it's a bit of a hassle...they are very eloquently written...u can read about the method he uses on
http://election.princeton.edu/methods/
here's a little paragraph for u:
The second step of the calculation is complex: it calculates the probability distribution of every possible electoral vote (EV) outcome.
wow, a simple "just averaging all the polls together" model??? i don't think so...it's as eloquent of an approach as u can find...really nicely done actually...
it's a good read if u're a math-geek like me. Anywhoo...the thing is, he uses a stronger model, but doesn't tweak and post-process as much as Silver does...it's really quite simple...same amount of work, just done in fewer steps...
What was true this year...again, 4 years from now, both Silver and PEC could be terribly wrong, if same polling is used on other sample, this is the problem with most major polling-companies...they do brute force polling, and hope it will fit, they tweak it around in the last minute, and hope that the shit will stick...
"What was true this year...again, 4 years from now, both Silver and PEC could be terribly wrong, if same polling is used on other sample, this is the problem with most major polling-companies...they do brute force polling, and hope it will fit, they tweak it around in the last minute, and hope that the shit will stick..."
Bruce,
The work done by Nate, PEC, and Pollster.com this year may lead to polling process improvements by Rasmussen, SurveyUSA, PPP, etc.
eltoro...good point...didn't think of that...i guess i underestimate the strengths of the polling-results...
My two cents on this are:
538 and the PEC were my two sanity checks during this entire election cycle. Their techniques differ and there is some esoteric statistical debate to have about which of them has the better method. PEC did extremely well in 2004 as well, and we can't know for Nate.
As for MY's post, I read it as saying that:
1) 538 is an interesting site, but not significantly better than just averaging polls.
2) Lots of people seem to be give a lot of post-election kudos to Nate that may or may not be justifiable, but which probably are just bandwagon jumping.
For my part, (1) is debatable, but there is no doubt that Nate performed yeoman's work in the primaries. He may have been overtuning his model in the GE, but again, that's esoteric statistician's debate stuff.
(2) is probably right on MY's part.
Again, just my $.02
super-geek smackdown.
nate got the attention not because he was right, but because he came out with a shiny new product that HE developed. Not the Princeton blah blah blah....HIM.
MY has wit and raw smarts in lorryloads.
This very post demonstrates the reason for Sullivan's award. He's calling it as he sees it. Other factors account for Silver's popularity than his uncanny polling genius.
Ygz wasn't talking shit about Silver. He was talking shit about Silver's fans who don't understand that Silver's early work may have been dope, but his latest (despite all the creative shit he tried in the studio) sounds pretty much the same as what everyone else is putting out - it's nice, but nothing special.
Yglesias is actually right in this particular case, but Sam Wang (at the Princeton site br linked to) is righter.
Just averaging national polls does fine in an election like this one, which ultimately wasn't balanced on a knife edge. If you want to do well in a squeaker like 2000 or even 2004, where individual state results matter a lot and there could easily be a popular/electoral vote split, you need better technology than that.
But Nate Silver's model was still far more complicated than it needed to be. You can aggregate state polls and still do it more simply and consistently. Wang's processing of state polls was simpler than Silver's, didn't involve constant algorithm changes every week or two, and came closer to the mark in the end--Wang's final prediction of Obama's most likely EV tally, which admittedly went a tiny bit beyond the confines of his public model, was off by only one vote! Silver was low by, I think, 17 votes because of an ad hoc correction he made for imaginary tightening of the race in the final hours. If he hadn't done that he would have been closer.
That said, I really enjoyed reading Silver's site over the course of the campaign. More than the model predictions, it was the reports from the road, with beautiful photos, that kept me coming back. They gave a great sense of what it was like to be out there working the campaigns.
Matt McIrwin,
The one shortcoming in both Silver's and Wang's calculation of the expected EV total was that it didn't tie to the EV total of the states that they called for Obama. As Wang admits on his website, both he and Nate called 49 out of 50 states correctly (and called D.C correctly as well), and they got the same state wrong (Indiana). If we calculate the electoral votes on the basis of the states that they actually called for Obama, the actual prediction for Obama's EV total given by Silver and Wang was 353. In other words, they were both off by 12 EVs.
Don't know why they reported the mean and not the median (better for the end user). Probably just a measure of safety (i.e. when the Marcel predictions say that so and so is going to hit 25.3 home runs next year).