Il Douché and His Discontents: The 2016 Primary Voting Thread, Part 4

i'm saying his priors about pollster reliability were demonstrably wrong.?

In what sense? At this point he's been through hundreds and hundreds of elections. If he didn't get several of the 99% predictions wrong, it would mean he was miscalibrating things and expressing too little confidence in his predictions.

"if ppl look at his figures and assume they are 100% trustworthy that is a little bit on them"

if you're saying caveat lector that's fine. but people concluding this from a prediction that claims there is a >99% probability of a hilary win, which is what he said about MI iirc, are interpreting him completely correctly.

that is what the prediction he made _means_. it's the fundamental premise of doing things probabilistically is to do a better job of encoding your own uncertainty.

two polls out in the last day or two, first since super tuesday:

NBC/WSJ

Trump: 30
Cruz: 27
Kasich: 22
Rubio: 20

Clinton: 53
Sanders: 44

ABC/WashPo

Trump: 34
Cruz: 25
Rubio: 18
Kasich: 13

Clinton: 49
Sanders:42

there is a >99% probability of a hilary win, which is what he said about MI iirc, are interpreting him completely correctly.

I mean, based on what we knew Sunday, what do you think is a reasonable estimate for the probability of a Sanders win in Michigan? Anything over 10% seems very, very hard for me to justify.

Pierce:

But, as I talked to more and more people around Flint, I got the sense that the resonance of the exchange was not what HRC and her campaign thought it would be. The UAW members I talked to clearly considered HRC's use of the auto bailout against Sanders to be at best a half-truth, and a cynical attempt to win their support, and they were offended by what they saw as a glib attempt to turn the state's economic devastation into a campaign weapon. These were people who watched the auto industry flee this city and this state, and they knew full well how close the country's remaining auto industry came to falling apart completely in 2008 and 2009. They knew this issue because they'd lived it, and they saw through what the HRC campaign was trying to do with the issue. I have no data to support how decisive this feeling was in Tuesday night's returns, but it seems to me to be one of the more interesting examples of unintended consequences that I'd heard in a while.

she definitely could've made that critique better. the way she made it was sleazy but she could've said something like "the first bailout bill was opposed by senator sanders despite containing respite for the auto industry because he didn't like the rest of it - we need someone willing to compromise" - i don't know if it would've been effective but it a. would've been honest and b. would've hit sanders in a somewhat weak spot - the idea that he isn't a pragmatist who can get important things done bc of his ideology.

the fact that Silver blew Michigan left baseball analysis means he's worthless.

i enjoy gawker's various descriptions of trump, i'd poll them except they're too numerous to track down

― nomar, Tuesday, March 1, 2016 2:01 PM (1 week ago) Bookmark Flag Post Permalink

half-empty bag of rancid tapioca

A half-truth, and a cynical attempt: Hillary 2016

"his forecast updates its prior by itself though"

actually i have no idea what this means

― 𝔠𝔞𝔢𝔨 (caek), Wednesday, March 9, 2016 2:01 PM (14 minutes ago) Bookmark Flag Post Permalink

some ppl are mad that his model still has HRC with 99% chance of winning FL, the implication being that it should have become more uncertain and more in favour of Bernie after last night's results. if the forecast i work on severely over- or under-forecasts something, once we feed in the new data and update it, it updates its belief about subsequent outcomes downwards or upwards accordingly (holding all else constant). for Silver in this case it would update the prob of bernie winning upwards and discount poll data relative to other inputs. (i actually haven't looked at his methodology since last year, but i'm assuming he doesn't forecast each primary independently and estimates like the joint distribution of primary votes across states?) but the updating is done completely internally within model, you just feed in the data and it calculates the new prior distribution according to bayes' rule. aside from changing the specification of his model there isn't some crank he can reach in and pull to give it a desired result, and the whole point of his enterprise is to have a hands-off forecast unmuddied by the dirty misled intuition of pundits.

Oh hai guys Marquito's on campus today. What should I ask?

are you a wide receiver?

http://www.tampabay.com/resources/images/dti/rendered/2015/09/0610102014__1__15902438_8col.jpg

also tbc i don't think there's anything wrong with giving NS a good ribbing or saying he has a bad forecast (although idk if there are others that did better, is that princeton guy running his again?), just that reading political bias into it because we don't like the outcome is silly because that's exactly what we mocked the right for doing 4 years ago. saying well it's different because he made some bloopers this time isn't convincing to me. i get that the narrative has become much more critical of self-consciously data-driven analysis since 2012, but idk it still just seems like a dumb reaction to have. one of the best NS moments last election was after the first Romney Obama debate, when the poll swung in Romney's favour and all the right pundits were pissed at Silver because his forecast barely budged. like, are we going to do that every time now?

Come back Fred all's forgiven

Ask Rubio about climate change. Floridians love sea level rising.

got u flopson. no disagreements there.

i guess my point is "the polls were bad" does not at all absolve his or any other extremely confident prediction. it is a strength of his predictions that one of the outputs is a confidence. that confidence was very high. when that happens that's due no only to the polls having a 20 percent gap, but also him having high confidence in the polls. he was wrong in this case.

Let us take a break from our quibbling to focus on the real threat:

http://www.sexycongress.net/
"Who would you rather have sex with?"

some of these are very very difficult to choose between

they're just all so sexy!!!

nsfw i'm sure

btw the daily CNN Trump speech semed to be preempted today by Nancy's burial

Bobby Jindal coulda carried that coffin by himself

It isn't just "he feeds in the data" though, he ranks which polls to consider and that's his magic touch - if he thought that Bernie would beat the 20% polls then there's ways to express that - it's the 99% confidence that's the problem not that he called it for Clinton. I don't know what I would have done differently, but I'm not paid to be Nate Silver!

my mistake, she's lying in state at the liberry, plenty of time for the candidates to show up and genuflect if they can't make the Friday funeral

xp

https://pbs.twimg.com/media/CdIhsJxW8AQ13Vz.jpg

https://pbs.twimg.com/media/CdIhsJ_XEAEB4u8.jpg

In April 2015 Rubio had good national name recognition and positive press coverage. It was fair enough to call him a "real contender".

Marco Rubio IS base.

and most of us hadn't yet seen he can only muster the gravitas of a seventh grader

he sure could drink water

Nate Silver can be forgiven for that inasmuch as any pundit could, because it's an entire industry built on acting like you know shit you don't or can't know. There probably were people with more direct experience with Rubio who were more skeptical of him, although there were others who probably weren't. I guess the only thing with Nate/538 is that the stats-geek aspect of it gives stuff on the entire website an air of objectivity and science that it doesn't warrant.

rubio had a ton of endorsements and the support of the party which before this election was a strong indicator

Asked by Bloomberg why they aren’t backing Cruz, senators responded with awkward chuckles, long silences, and evading the reporter by ducking into a restricted elevator. Sen. Richard Shelby (R-AL) responded by saying he was headed to lunch.

sick burn shelby

Asked by Bloomberg why they aren’t backing Cruz, senators responded with awkward chuckles, long silences, and evading the reporter by ducking into a restricted elevator. Sen. Richard Shelby (R-AL) responded by saying "Bazinga".

johnny damon has endorsed donald trump

angry blue collar + dumb millionaire coalition

Marco Rubio, all your possible voting base are belong to us

It isn't just "he feeds in the data" though, he ranks which polls to consider and that's his magic touch - if he thought that Bernie would beat the 20% polls then there's ways to express that - it's the 99% confidence that's the problem not that he called it for Clinton. I don't know what I would have done differently, but I'm not paid to be Nate Silver!

― Andrew Farrell, Wednesday, March 9, 2016 3:26 PM (37 minutes ago) Bookmark Flag Post Permalink

this is what his website says about poll weighting/ranking:

The weights reflect the quality of each survey as determined by FiveThirtyEight’s pollster ratings, which grades polls based on their past accuracy and methodological standards. The poll weights also adjust for a poll’s sample size and how recently it was conducted. All polls are included in the weighted average unless they were internal polls released by a candidate or a candidate’s super PAC or if we have good reason to suspect that the poll faked its data or committed other gross ethical violations.

i guess he could go in and fudge the weights to make the ones that tend pro-bernie (or in the case of MI, relatively less pro-Clinton) have more weight... but i don't think this is a parameter that he has discretion over and can change at a whim

i also don't like the implication that NS the human = his forecast, and the suggestion that he should tinker with his forecast every time it makes a bad call or runs counter to intuition is exactly wrong and misunderstands why it's interesting. saying 99% confidence doesn't mean NS the human was literally feeling the human emotion of confidence, it just means the stand error was small enough, which is exactly what we would expect to happen when all the polls show hillary in the lead. people literally going "that smug neoliberal fuck is 99% confident Bernie's gonna lose! and he hasn't even updates his confidence on Florida" seem to not understand the degree to which it's largely out of his hands at this point--and that that's what makes it interesting

Pierce on the Trump Steak speech:

You will never see another performance like the one he put on last night. It was more than a demonstration of the degradation of democracy. It was also a demonstration of the degradation of capitalism. I mean, Jesus, have some pride, rich folks. At least Andrew Carnegie built libraries and Jay Gould wore stylish diamond stickpins. None of them tossed slabs of dead cow to their supporters from the stage. I'm not entirely sure, but I think that, halfway through the speech, I accidentally may have bought a time-share in Florida.

iirc he also handed out a copy of some magazine (trump magazine?) to someone in the crowd, halfway through.

Rubio, packing in the hometown crowds.

https://twitter.com/ByronYork/status/707684148316610560

yeah but it was a more intimate show

I just got home from the work, in the same building as the town hall. Apart from the MSNBC and NBC vans and a Jose Diaz-Balart spotting, no traffic problems today.

By the way: I'm sorry Miami inflicted Chuck Todd on the rest of you.

pretty sure i have never seen or heard him, but i now think of him as 'my man chuck todd'

Chuck T

Some Clinton supporters chose to vote in the Republican primary. We know 7 percent of voters in the Republican primary identified themselves as Democrats to exit pollsters, compared with just 4 percent of voters in the Democratic primary who said they were Republicans. “Those 7 percent of Dems were likely mostly Hillary voters who thought she had an easy win and they could do their part trying to stop (Donald) Trump,” said ***Bernie Porn*** of pollster EPIC-MRA. The exit-poll samples are too small, though, to check that.

http://fivethirtyeight.com/features/why-the-polls-missed-bernie-sanders-michigan-upset/ 8th paragraph down.

Bernie Porn?

There are fifty states in the US (I should know, I've lived there for a year!) which means 100 primaries every cycle, which means that at least once per cycle a 99% certain outcome could be wrong. This was the biggest upset since 1984 at least. It could be completely correct to say that the probability of this happening was >99%. Silver doesn't think so, though, he said on twitter that they've been tweaking the model. Michigan only went down to 98% Clinton, though.

#actually there are 46 states and four commonwealths

actually there is Texas and 49 lesser states

45 states, 4 commonwealths, and a principality.

Democratic debate tonight. Three days after the last one.

I'm burnt out- no more debates for me.

i also don't like the implication that NS the human = his forecast, and the suggestion that he should tinker with his forecast every time it makes a bad call or runs counter to intuition is exactly wrong and misunderstands why it's interesting. saying 99% confidence doesn't mean NS the human was literally feeling the human emotion of confidence, it just means the stand error was small enough, which is exactly what we would expect to happen when all the polls show hillary in the lead.

ok you can replace every time i said "nate silver was wrong" with "nate silver's model was wrong" if it makes you feel better about the implications. or passive voice?

anyway, his priors about pollsters are of course data-driven but the decision to apply priors at all is subjective (even if they are uninformative or flat priors), and the priors and the other uncertainties he folds into the model were obviously too confident here, to the extent that they lead him to rule out the final result with extreme confidence.

why am i saying "obviously" too confident? why is this not just "if you make 100 predictions with 99% confidence, you're going to be wrong about once"? the answer is the same reason that far more than 1 in 20 psychology papers that claim p<0.05 are wrong. the defense is true as far as it goes, but that's how you explain away random noise, i.e. statistical fluctuations. but the problems are systematic, not random. i don't think anyone is arguing unlucky random samples is what went wrong with the polls. i mean it could be what went wrong (there is a non-zero probability that 5 polls with 3 percent uncertainties could all be wrong in the same direction by 20% or so), but i think everyone is right in assuming there were _systematic_ problems. and it's concern about those systematic errors that should be in silver's priors and in the resulting confidence of the model.

people literally going "that smug neoliberal fuck is 99% confident Bernie's gonna lose! and he hasn't even updates his confidence on Florida" seem to not understand the degree to which it's largely out of his hands at this point--and that that's what makes it interesting

not trying to pull rank here, but i do know what i'm talking about re: statistics, and i can't vote in this election, so i'm not coming at this from a "neoliberal fuck" angle.