urbanSTL

Ebsy wrote:Also, Nate Silver is a discredited fraud and his methodology is not really all that sound.

Well this is just not true.

MarkHaversham wrote:

Ebsy wrote:Also, Nate Silver is a discredited fraud and his methodology is not really all that sound.

Well this is just not true.

Well, he has certainly made an embarrassment of himself this cycle, with the primaries (arguing for months that Trump wouldn't win) and now the general (arguing for months that the race was close and that Trump could win). It's not really surprising though, since Silver isn't a politics guy and came into it through the sports side. He isn't even all that good at the sports stuff, but there is an insatiable need for "data journalism" in our culture. His big claim to fame is correctly predicting states in 2008 and 2012, but I know a lot of people that made similarly accurate predictions without fancy algorithms or demographic models, not necessarily knocking those.

chaifetz10 wrote:Are you reading the same articles from FiveThirtyEight that I am? I... I don't think you are. Nate doesn't predict based on his own feelings, but uses data to try to understand trends and reasons to make statistical forecasts. He never said Trump wouldn't win, but always provided the percentages that showed what would have to happen for him to win. As for 'embarrassing himself' with the primaries? He pretty much nailed them except for one or two upsets which even he admitted was based upon poor polling data and was within the realm of expectations.

Yeah, I am, and I can definitely link you the many articles Silver wrote saying Trump had no chance of winning, using statistical analysis to paint whatever picture he wanted. Do I need to bring up endorsement points? While Nate Silver was puttering around saying Trump couldn't win and ignoring the polls, there were other people doing much better analysis like Nate Cohn over at the Upshot or Sam Wang at the Princeton Election Consortium (a guy who from the beginning took Trump seriously and didn't ignore the polls). Cohn in particular has been doing groundbreaking work this election and has been fundamentally changing the understanding of demographics and demographic modeling. And in regards to predicting primary outcomes, if you just literally add the polls together and find the mean, you can pretty much predict the outcome of any race. He shouldn't get credit for just going with (a slightly modified) polling average any more he should be blamed for poor polling quality in Michigan and Indiana leading to missed predictions. Nate Silver might be a popular "celebrity statistician" or numbers guy, but that doesn't mean he is particularly good, only that he has done a decent job building a (false) reputation for being good. Of course, that reputation has taken a very serious hit this cycle, and with very good reason.

Ebsy wrote: Yeah, I am, and I can definitely link you the many articles Silver wrote saying Trump had no chance of winning, using statistical analysis to paint whatever picture he wanted. Do I need to bring up endorsement points? While Nate Silver was puttering around saying Trump couldn't win and ignoring the polls, there were other people doing much better analysis like Nate Cohn over at the Upshot or Sam Wang at the Princeton Election Consortium (a guy who from the beginning took Trump seriously and didn't ignore the polls). Cohn in particular has been doing groundbreaking work this election and has been fundamentally changing the understanding of demographics and demographic modeling. And in regards to predicting primary outcomes, if you just literally add the polls together and find the mean, you can pretty much predict the outcome of any race. He shouldn't get credit for just going with (a slightly modified) polling average any more he should be blamed for poor polling quality in Michigan and Indiana leading to missed predictions. Nate Silver might be a popular "celebrity statistician" or numbers guy, but that doesn't mean he is particularly good, only that he has done a decent job building a (false) reputation for being good. Of course, that reputation has taken a very serious hit this cycle, and with very good reason.

Yeah Cohn has does some good work on demographics. Though I don't really understand how you disparage Nate silver for "averaging polls together" and turn around and talk up Cohn, at the end of the day his model really simplified is an average of polls as well. No one knew who Nate Cohn was 2 years ago so you certainly have to give silver credit for building awareness of stat models to predict politics. I guess I'd like some details on why silvers model is so bad and the upshot is so good. I research and analyze statistics for work a fair amount and I don't find anything particularly agregious as to how silvers model is constructed.

I agree, The Upshot's election forecast is pretty similar to Silver's (and the Big three, Rothenburg, Cook and Sabato, along with PEC and DKOS Elections) and more than likely will end up being fairly accurate on election night, but this isn't really a tribute to either as forecasting elections in the age of massive numbers of pollsters and state and national polls is not that difficult. I'm more praising Cohn for his other work, such his analysis of the "missing white voters" and this fascinating piece on how different pollsters can take the same microdata and get varying results. His expose of the LA Times/USC Daybreak poll has also been a highlight of this year. The stuff coming out of 538 in general has been pretty bland and uninteresting and the stuff coming from Nate Silver himself has increasingly resembled Dick Morris or Bill Kristol levels of punditry. His big contribution in 2012 was casting doubt on the importance of Likely Voter screens (they ended up moderately skewing towards Romney) but I can't really say he has done anything on that level this year. Also, his Senate forecast model is pretty bad and any model that included Google Consumer Survey (they once put out a DC poll that had Clinton +6) polls is pretty bad.

Silver wasn't walking around saying Trump couldn't win the primary any more than Cohn was, as far as I could tell they basically had the same take on it: Trump is an unusual candidate with an uphill battle to win against the establishment, and his long-shot chances relied on an unusually large and uniformly ineffective field of candidates.

Plus, there's a wide gap between "Cohn is doing good work, Silver is boring" and "Silver is a discredited fraud".

Ebsy wrote:I agree, The Upshot's election forecast is pretty similar to Silver's (and the Big three, Rothenburg, Cook and Sabato, along with PEC and DKOS Elections) and more than likely will end up being fairly accurate on election night, but this isn't really a tribute to either as forecasting elections in the age of massive numbers of pollsters and state and national polls is not that difficult. I'm more praising Cohn for his other work, such his analysis of the "missing white voters" and this fascinating piece on how different pollsters can take the same microdata and get varying results. His expose of the LA Times/USC Daybreak poll has also been a highlight of this year. The stuff coming out of 538 in general has been pretty bland and uninteresting and the stuff coming from Nate Silver himself has increasingly resembled Dick Morris or Bill Kristol levels of punditry. His big contribution in 2012 was casting doubt on the importance of Likely Voter screens (they ended up moderately skewing towards Romney) but I can't really say he has done anything on that level this year. Also, his Senate forecast model is pretty bad and any model that included Google Consumer Survey (they once put out a DC poll that had Clinton +6) polls is pretty bad.

Alright I agree with that assessment, the upshot certainly does far more interesting pieces on voter data mining and the look into the USC poll was really interesting.
Just curious as to your concerns with 538's senate model? Not enough polls to have real robust polling data so it does have to be built off of a lot of assumptions.

So I was listening to St Louis PR many months back and I remember hearing something to the effect of the following:

Slay was not running for reelection because Clinton would instate Claire McCaskill as a member of her cabinet (They floated Secretary of State I believe?? Can't recall for sure however), and that Jay Nixon would then put Slay into McCaskill's seat to finish out her term.

Any thoughts on this??

Chalupas54 wrote:So I was listening to St Louis PR many months back and I remember hearing something to the effect of the following:

Slay was not running for reelection because Clinton would instate Claire McCaskill as a member of her cabinet (They floated Secretary of State I believe?? Can't recall for sure however), and that Jay Nixon would then put Slay into McCaskill's seat to finish out her term.

Any thoughts on this??

Makes about as much sense as anything I suppose. But that makes Lewis Reed Mayor for 4 months going into an election for the open mayoral race, which Reed is a candidate in. As there is little love lost between Reed and Slay, I have a hard time seeing Slay bite that bullet and giving Reed a chance to prove himself in the actual office.

timing doesn't work out... Nixon will be out of office by time the new president will be sworn in. May not be a big deal if Koster wins governor race but of course if Grietens wins that would be a different story. I've also heard hearsay rumors that Slay might be up for an Ambassador position or something like that in a Clinton Administration.

Ebsy wrote: Also, Nate Silver is a discredited fraud and his methodology is not really all that sound. Anyone with a calculator can average polls together, but what Silver is doing now is unskewing polls to match what he thinks the actual numbers are. Always a risky business, and one that usually ends up in disaster on election night. Now, Clinton will probably win by a wide enough margin on election night for it to not matter for Silver this cycle, but there are much better election statisticians then Silver.

Silver successfully called the outcomes in 49 of the 50 states in the 2008 U.S. Presidential election, He also correctly predicted the winners of every U.S. Senate race.
In the 2012 United States presidential election, Silver correctly predicted the winner of all 50 states and the District of Columbia.

98% is discredited fraud? lol...

Well, calling 50 states correctly is no big deal really, since there are only like two or three states in question. Basically everyone remotely knowledgeable will be 95% accurate using no math whatsoever. What would be more interesting is comparing errors in the state-by-state poll margins.

Nate Silver's big miss as far as I can tell seems to be dismissing Trump's chances in the primaries. And that was him going against what his models were telling him and personally dismissing Trump. His model has been extremely accurate. At the time he left Baseball Prospectus PECOTA was the best publicly available forecasting model out there. And quite a bit better than a bunch of the in-house stuff too.

I could see how his model being more accurate than human prognosticators would rankle you if you don't really understand how statistical analysis works or if you are one of those "NERDS ARE RUINING BASEBALL' type of old timers, but it's still a pretty big leap to "discredited fraud".

ebo wrote:Nate Silver's big miss as far as I can tell seems to be dismissing Trump's chances in the primaries. And that was him going against what his models were telling him and personally dismissing Trump. His model has been extremely accurate. At the time he left Baseball Prospectus PECOTA was the best publicly available forecasting model out there. And quite a bit better than a bunch of the in-house stuff too.

I could see how his model being more accurate than human prognosticators would rankle you if you don't really understand how statistical analysis works or if you are one of those "NERDS ARE RUINING BASEBALL' type of old timers, but it's still a pretty big leap to "discredited fraud".

his dismissing Trump last fall had nothing to do with polling...just what he thought based on past Trump- like candidates....i think more people than not had the same feeling about Trumps candidacy. His model gives trump a 13% chance right now..
others in the field that have Trump between 1 and 7%.

^That was kind of my point. He said his dismissal was him trying to play pundit instead just being an analyst. Trump was such an outlier that a lot of people, not just Silver, couldn't take him seriously.

Things look pretty grim right now for Trump in Silver's model as you indicated. The drama on election night seems like it will be figuring out how much damage Trump does to down ballot candidates.

dbInSouthCity wrote:

ebo wrote:Nate Silver's big miss as far as I can tell seems to be dismissing Trump's chances in the primaries. And that was him going against what his models were telling him and personally dismissing Trump. His model has been extremely accurate. At the time he left Baseball Prospectus PECOTA was the best publicly available forecasting model out there. And quite a bit better than a bunch of the in-house stuff too.

I could see how his model being more accurate than human prognosticators would rankle you if you don't really understand how statistical analysis works or if you are one of those "NERDS ARE RUINING BASEBALL' type of old timers, but it's still a pretty big leap to "discredited fraud".

his dismissing Trump last fall had nothing to do with polling...just what he thought based on past Trump- like candidates....i think more people than not had the same feeling about Trumps candidacy. His model gives trump a 13% chance right now..
others in the field that have Trump between 1 and 7%.

I don't remember Trump having a projected 60% chance of winning the Republican primary in any models as of Fall 2015.

No one had (or really could have) a 60% chance of winning the Republican primary in the fall of 2015. It wasn't even entirely certain who would be in the race. But beginning in February of 2016 I believe Fivethirtyeight had Trump as the favorite. (Cruz was the favorite when the model opened, but Trump overtook him fairly quickly.) There were a couple of upsets: Fivethirtyeight had Trump taking Iowa and Oklahoma, both of which went to Cruz. (Which is to say essentially that Trump actually underperformed his polls slightly in [portions of the primary season. We forget that because he won. But . . . go back and check.)

You can go through individual states here and see Fivethirtyeight's predictions . . .
http://projects.fivethirtyeight.com/ele ... epublican/
And compare them to the actual results here:
http://www.nytimes.com/elections/results

They're not perfect, but they're generally pretty close. In the end they're really just trying to untangle the sometimes contradictory and confusing world of polling. Polls are good, but they're obviously never quite perfect. Perhaps Fivethirtyeight has a relationship to the election somewhat similar to the relationship between an index future and the broader market at the date of the contract's maturity. (I was going to say an index fund, but given that they're trying to predict where something will be weeks or months down the line maybe futures are closer.)

The path to 1237 wasn't exactly their prediction, but it was one way of looking at the nominating process. (Which, to be fair, is a lot more complex than the general election, thanks to polling taking place on different days, in different ways, and with delegates awarded in differently.) The "path" gave a baseline against which you could measure candidates' performance. Candidates who were beating the baseline were generally doing better. Candidates failing to meet it were doing worse. After Iowa the only Republican who ever outperformed his baseline was Trump, and on that basis Fivethirtyeight fairly consistently predicted that was more likely to win the nomination than anyone else. I don't feel like investing the time to measure their performance perfectly right now, but as memory serves they did fine. And at a glance, the numbers seem to bear that out. Feel free to go through and compare state by state, but please realize that no odds maker was going to give anyone in that fifteen way cagematch monstrosity more than 50% in 2015 for precisely the same reason none will give a particular baseball team anything like a 50% chance of winning the World Series before opening day: there's were just too many players, too much season, too much that could happen, and too much blind stupid chance. Note: The dreadfuls didn't surpass 50% until they won the pennant. And their odds aren't a whole lot better than 50% now. (Lord I'd like to see an upset in that one. Let's have a nice baseball October surprise . . . and I'll try to leave the politics out of it. Though I will say the only T I'd allow on a red hat is the little one that goes with the big S and L.)

http://projects.fivethirtyeight.com/ele ... publicans/

My recollection is that 538 was initially skeptical, and compared him to every other blowhard with a hardcore fan base in every other primary that ever reached 30% in polling and flamed out when the voting started. They got on the Trump train at exactly the right time: when the voting made it clear that Trump was enduring and the establishment candidates were flailing.

I believe you are recalling Silver acting like a pundit. One thing 538 doesn't officially do is make predictions very far out. They'll give you odds, but they try not to make predictions. And their odds from the first available polls had Trump second, and very soon after that first. Which was initially just a reflection of the national polling, I think, but incorporated state data as soon as it became available. They try not to be pundits too often, but once in a while everyone slips. Trump is fundamentally a media figure and not a politician. That makes him a rather odd candidate. Pundits trip on that stuff.

Most primaries have oddballs like Trump running. Heck, the 2012 election had Actual Trump. The only unusual thing about Trump is that he won the primary instead of fading down the stretch. Saying "Trump is polling reasonably well but it probably won't last" isn't Silver being a pundit, it's what any person with an understanding of electoral history would say about a wacko like Trump polling well on the basis of a dedicated niche.

Trump left the 2012 election in May of 2011. He played no real part in the primary. I cannot think of a primary with a candidate quite like him. Perot might be the closest parallel I can think of, but he didn't run as a Republican . . . and he's not a terribly close. He wasn't a media figure beforehand. He was far more articulate and much less divisive; much more the Tea Party economic Republican of today. I can't find a major party primary candidate in the last twenty years who was such an outsider as Trump. There was Sharpton in 2005. He has never held elected office, so far as I can recall, but he's been so politically active for so long that I don't think he's a particularly good comparison. (And he never polled well.) I suppose Forbes might be a parallel, but he'd been an associate of Reagan's some ten years earlier, he'd long been politically active, and he was never the sort of media figure that Trump is. And he won precisely two states in 1996, not including his home state, and none in 2000 when he never crossed 9%. Sure, there's usually a contest. There are often several strong candidates in the out of power party, but they're generally established politicians: governors like Dean and Kasich or Senators like Cruz and Clinton. Heck, very often they come back and try more than once. (The strongest candidates often do in the modern era: McCain and Clinton being but the two who actually won the second time around. . .) At least during the period of my own political activity few people that hadn't held elected office have chosen to run for the presidency, and none before have come so close. The last political outsider I can think of elected to the Presidency is Dwight Eisenhower . . . if you can call him an outsider. Trump really is rather unprecedented, so far as I can tell. And I assure you, I'm looking diligently to try to find a precedent closer than Europe during the depression. There might be one, but even so, it seems quite clear that he's a fairly rare bird. (Thankfully.)

It's time for Roy Blunt to go away. He is the epitome of a crony career politician just like Hillary. Trump's promise of a term limits constitutional amendment is almost enough to get my vote. Both parties have wanted this for a long time (when they aren't campaigning).

Pat Robertson, Pat Buchanan, Ron Paul, Steve Forbes... there are factional candidates like Trump practically every year. How was anyone supposed to predict Trump would beat those guys' performances? Maybe Steve Forbes couldn't get away with blatant white supremacist pandering the way Trump did, maybe he could've if he tried. I don't think anyone was predicting in mid-2015 that Trump could claim Mexican-American judges weren't qualified to try him, or that he could kill someone in broad daylight and get away with it, and go on to nearly win not just a few states, not just the whole primary, but maybe the election. Trump's success has been a rarity, but I don't think it was predictable.