|
Please clap...
|
#
?
Nov 7, 2016 01:04
|
|
|
|
| # ? Jun 1, 2024 20:48 |
|
|
Dead Cosmonaut posted:And the fact that Trump is basically a walking antithesis to any polling that was against him in the primary For example?
|
|
#
?
Nov 7, 2016 01:34
|
|
|
How about Nate shakes and the words hold still? They're not shook, he is.
|
|
#
?
Nov 7, 2016 01:36
|
|
|
https://twitter.com/NateSilver538/status/795446394169475072 Nate remains shook.
|
|
#
?
Nov 7, 2016 03:22
|
|
|
So let's just recap: Sam Wang: Clinton wins 323 EVs DeSart and Holbrook: Clinton wins 347 EVs Drew Linzer: Clinton wins 312 EVs RealClearPolitics, a right wing site that has consistently underestimated democratic support in every election: Clinton wins 297 EVs 538: Clinton wins 273 EVs It's really fun looking at the NC and FL tabs on 538 because the vast majority of polls for NC and FL for the past two weeks show Clinton winning those states. Silver only arrives at Trump winning NC, NV, and FL through a massive amount of 'unskewing'.
|
|
#
?
Nov 7, 2016 04:45
|
|
|
The Whole Internet posted:So let's just recap: 538 is projecting about 310 as the most likely value. Look at the histogram.
|
|
#
?
Nov 7, 2016 04:49
|
|
|
fermun posted:https://twitter.com/NateSilver538/status/795446394169475072 https://twitter.com/MattBors/status/795447974646063104
|
|
#
?
Nov 7, 2016 05:12
|
|
|
The Whole Internet posted:So let's just recap: Nate's model is currently estimating 291.6 EVs for Clinton.
|
|
#
?
Nov 7, 2016 05:41
|
|
|
nate's model is estimating a boot in his rear end in about 48 hours
|
|
#
?
Nov 7, 2016 06:32
|
|
|
C. Everett Koop posted:nate's model is estimating a boot in his rear end in about 48 hours If Nate walks into a store and wants to buy six feet of rope probably don't let him check out.
|
|
#
?
Nov 7, 2016 06:34
|
|
|
Vox Nihili posted:Nate's model is currently estimating 291.6 EVs for Clinton. He's predicting a Clinton loss in NV, NC, and FL. His electoral map adds up to 273.
|
|
#
?
Nov 7, 2016 06:35
|
|
|
Bip Roberts posted:538 is projecting about 310 as the most likely value. Look at the histogram. The Whole Internet posted:He's predicting a Clinton loss in NV, NC, and FL. His electoral map adds up to 273. Vox Nihili posted:Nate's model is currently estimating 291.6 EVs for Clinton. itt we learn about : -probability -the difference between the mode, the mean, and the median
|
|
#
?
Nov 7, 2016 06:37
|
|
|
FYI Enten and Cohn were on CNN today around 1pm and both predicted 325+ EVs for Clinton
|
|
#
?
Nov 7, 2016 06:39
|
|
|
so just tbcHarlock posted:FYI Enten and Cohn were on CNN today around 1pm and both predicted 325+ EVs for Clinton this is the median: "she's as likely to get 325+ as 325-" Vox Nihili posted:Nate's model is currently estimating 291.6 EVs for Clinton. this is the mean: "on average, she'll get 291.6 EVs" Bip Roberts posted:538 is projecting about 310 as the most likely value. Look at the histogram. this is the mode: "the single most likely number of EVs she gets is 310" The Whole Internet posted:He's predicting a Clinton loss in NV, NC, and FL. His electoral map adds up to 273. that's not how probability works - saying that something has a 50.5% shot of happening is not the same as predicting it'll happen.
|
|
#
?
Nov 7, 2016 06:43
|
|
|
Pinterest Mom posted:so just tbc .Yeah, I feel like the big difference in these is degree of uncertainty. Everyone has a very similar median result, but Nate, Linzer and Wang have high, medium and low levels of uncertainty. Wang's big article about the difference between him and 538 is basically 'if I add a bunch of uncertainity, I get to [Where Linzer's Model is]' Upshot's handy tabulation is: Upshot: 84% 538: 65% Huffpost: 98% PW: 89% PEC: 99% DK (Linzer): 87% Given that these models all have very very similar median results, I feel like these forecasts are being driven entirely by uncertainty, and PEC has none and 538 has a lot. I don't think either is right, but it's hard to tell.
|
|
#
?
Nov 7, 2016 06:59
|
|
|
Worth noting is that a certain EV total may become the mode simply because several different plausible combinations of states can add up to it. This is why the EV total that shows up in the greatest number of simulations isn't always the same as the EV total you get by calling each state in whatever direction it's currently tipping. According to 538, the single most likely specific combination would apparently give Clinton 272 EVs, but there are also enough plausible combinations adding up to ~325 for that total to be similarly-likely. In fact, in the latest pass, it looks like the mode happened to fall at ~360 for Clinton. 
Supercar Gautier has issued a correction as of 07:09 on Nov 7, 2016 |
|
#
?
Nov 7, 2016 07:06
|
|
|
Pinterest Mom posted:itt we learn about : Perhaps we should back up and learn addition first.  273 The nowcast is not an apples to apples comparison to the numbers from PEC/RCP/KOS that I gave. I'm comparing the forecast of likely winners in each state from these sites. Some people here are pointing at the nowcast's 291 as if that means anything. The nowcast throws in national polls, unicorn blood, economic tea-leaf reading, and feelings. I don't give a gently caress about it. Yeah I get that the splits given to NC/NV/FL are all so narrow that the probability of Clinton taking one is larger than taking none in 538's forecast, even if individually each is disfavored on its own. No need to act like that's the secret to the pyramids. (And there is a deep flaw with those huge 538 error bars. It also means the probability of Trump winning one of the 20 safe blue states is greater than 50% when you add up all the percents for each... when we can say for certain it isn't). You're all missing the point here, 538 differs significantly from all the other vetted polling aggregators in terms of the swing states it thinks Trump is favored in.... majorly in Trump's favor. The Whole Internet has issued a correction as of 07:19 on Nov 7, 2016 |
|
#
?
Nov 7, 2016 07:16
|
|
|
The Whole Internet posted:Perhaps we should back up and learn addition first. But it doesn't differ that much in where it centers the distribution. It differs a ton in the width.
|
|
#
?
Nov 7, 2016 07:18
|
|
|
Bottom line is that distribution is majorly off the eyeball test with regard to some of the states they would have to win at the edges. Edit: it's also notable that it's spiky on the Clinton side but normal on the trump side. I assume that means that the Clinton side is resolving state by state trends but the trump side requires major nationwide polling malfunctions.
|
|
#
?
Nov 7, 2016 07:25
|
|
|
Bip Roberts posted:But it doesn't differ that much in where it centers the distribution. It differs a ton in the width. Nate's weighing of state polls this season is just majorly hosed up. His aggregate of state polls is clearly well off the mark from other aggregates and even his own nowcast.
|
|
#
?
Nov 7, 2016 07:26
|
|
|
The Whole Internet posted:Perhaps we should back up and learn addition first. idt he's (intentionally) putting his thumb on the scales for trump. I just think all the dogshit (esp national) polls are pushing the uncertainty to 50/50
|
|
#
?
Nov 7, 2016 07:31
|
|
|
theflyingexecutive posted:idt he's (intentionally) putting his thumb on the scales for trump. I just think all the dogshit (esp national) polls are pushing the uncertainty to 50/50 He's basically in a pickle because he made this model basically let it go and now it's sorta awry. Also he can only be vindicated by a major polling malfunction but no one will give him credit unless the states he says are 51% go blue and those at 49% go red.
|
|
#
?
Nov 7, 2016 07:33
|
|
|
But if Hillary walks away with 200 EV Nate will truly be the only one who considered it and he's getting zero credit for getting every state wrong.
|
|
#
?
Nov 7, 2016 07:39
|
|
|
Bip Roberts posted:Edit: it's also notable that it's spiky on the Clinton side but normal on the trump side. I assume that means that the Clinton side is resolving state by state trends but the trump side requires major nationwide polling malfunctions. Yeah, the spikiness happens because certain particular numbers can be reached in lots of plausible ways, and certain numbers just can't (eg, lots of likely ways to make 322, but 321 might need a really odd mix). I think you might be right; the fact that the graph smooths out on the left end seems to suggest that's kind of a "massive breakdown" range where those correlations and probable state combinations get way less relevant.
|
|
#
?
Nov 7, 2016 07:44
|
|
|
Forecasting is hard and people should really learn to stop making them.
|
|
#
?
Nov 7, 2016 07:46
|
|
|
Bip Roberts posted:He's basically in a pickle because he made this model basically let it go and now it's sorta awry. Also he can only be vindicated by a major polling malfunction but no one will give him credit unless the states he says are 51% go blue and those at 49% go red. I love when he and his lackeys hide behind the "systemic polling error" but there are basically zero polls estimating D and LPV turnout like we've seen in EV so far, so any error would be in hrc's favor. absolutely everybody knows polls oversample old white people and undersample every minority group and under 35s, so to lean on "maybe polls didn't oversample old white people and undersample every minority group and under 35s" is weak as hell. I can't believe some polls have the audacity to weight underrepresented minorities. Sample more people you cheap fucks
|
|
#
?
Nov 7, 2016 07:51
|
|
|
Nate silver is actively reinforcing the rigged narrative and the further away his model is from hrc's actual numbers (double digits, btw) the more likely there will be violence post-election
|
|
#
?
Nov 7, 2016 07:53
|
|
|
are violent shithead trump supporters really reading fivethirtyeight.com though
|
|
#
?
Nov 7, 2016 08:02
|
|
|
theflyingexecutive posted:Nate silver is actively reinforcing the rigged narrative and the further away his model is from hrc's actual numbers (double digits, btw) the more likely there will be violence post-election youve gone off the rails dude
|
|
#
?
Nov 7, 2016 08:12
|
|
|
theflyingexecutive posted:Nate silver is actively reinforcing the rigged narrative and the further away his model is from hrc's actual numbers (double digits, btw) the more likely there will be violence post-election HRC doesn't have actual numbers and won't until Tuesday.
|
|
#
?
Nov 7, 2016 08:22
|
|
|
Once again, the issue with Nate's model isn't the uncertainty, it is the volatility. There are certainly some unjustifiable things that Nate does (e.g., weighing polls by sample size, therefore giving google consumer surveys 3rd highest weight in this model), but those things would only affect things at the margins. In a scenario of stability, Nate's model converges with the others. The problem with Nate's model is that stuff like trend line adjustments have no justification, no explanation, and lead to wild swings in his predictions. A more cynical person may think that that is in the model to get page views.
|
|
#
?
Nov 7, 2016 08:33
|
|
|
theflyingexecutive posted:I love when he and his lackeys hide behind the "systemic polling error" but there are basically zero polls estimating D and LPV turnout like we've seen in EV so far, so any error would be in hrc's favor. That's why his "lackeys" in 538's editorial content say that Clinton is more likely to win than what the model projects. The number of dog poo poo polls being thrown into the model and new Hispanic voters being tossed out by most LV screens is something they constantly talk about. The problem with Nate's model isn't anything to do with whether or not he uses a trend line adjustment or anything else in that stupid HuffPo article. The key mistake Silver made that everything else stems from is that he assumes more data is better than less data even if the data is of dubious quality. He makes the assumption that polls with poor methodology can somehow be accounted for and averaged out. In the past when there were more polls and overall quality was better he was correct. The nature of the polls this cycle is very different, though. He also should understand that more can be wrong with a poll than just lean. He assumes that data can be derived from changes in polls even if those polls have poor methodology as long as the poor methodology is consistent. He assumes you can derive good conclusions from poor measurements. He's wrong and I think next cycle he becomes more choosy about which polls he chooses to include in the model. It seems like he probably makes this assumption because he comes from the sports world. In sports you don't usually have problems with measurement like this. The stats are the stats and the leagues keep track of them. You can trust the measurements. There are very minor issues with stuff like balls versus strikes, but those can in fact be averaged out since umpires are highly trained and most often not acting in bad faith. Pollsters are not like umpires or referees. ErIog has issued a correction as of 09:09 on Nov 7, 2016 |
|
#
?
Nov 7, 2016 08:58
|
|
|
ErIog posted:That's why his "lackeys" in 538's editorial content say that Clinton is more likely to win than what the model projects. The number of dog poo poo polls being thrown into the model and new Hispanic voters being tossed out by most LV screens is something they constantly talk about. I agree with everything you've posted and I agree that his model's #1 source of error is input sanitization. I listen to every 538 podcast and read every article; every writer is couching their Clinton prediction with a "trump could win with significant and systemic polling error". This is technically a true statement but the chances for such an error benefitting trump over Clinton are minuscule The double digits thing is my personal estimate based on applying my estimates of demographic turnout to RCP's model. Sub 50% Latino turnout is not happening, sub 80% black turnout is not happening and I also believe their estimates for Latino black and Asian differentials skew way more towards trump than believable. I base this on Spanish-language-inclusive polling as well as a gut feeling as well as my personal estimate of gotv effectiveness (using 2012 as a baseline) and cash on hand shortfalls for last-minute ad pushes
|
|
#
?
Nov 7, 2016 09:21
|
|
|
Dogshit polls are definitely a problem on RCP, which just throws everything into the mix and takes the average. I'm looking at you "Remington Research". 538's problem seems less that it has dogshit polls and more that it weighs them above good polls sometimes. In fact an analysis of which pollsters came closest in 2012 seems to have a tenuous correlation at best to which pollsters are getting A-ratings on 538, which boggles my mind.
|
|
#
?
Nov 7, 2016 09:20
|
|
|
I can't help but wonder if 538's prescience relied upon the very stable electoral maps in 2008/2012 and a greater abundance of polling. It was very easy to draw correlative relationships between the red/blue lean of each state because they remained almost identical between election years, by the same margins with respect to the national vote. Now that we're in a year without Obama as a presidential candidate and there's some voter realignment going on, the leans of certain states have jumped out of order such that we may get as many as 10-12 out of position with respect to 2012's vote. So working off these prior assumptions of how states will act leads to phenomena like a Clinton +14 in Virginia lowering her win probabilities by 1%. I think the fact that people can see this happening in real time ironically contributes to why they're questioning it and getting Nate to melt down on twitter. At this point he's too committed to the model to adjust, so this is what we're going to get.
|
|
#
?
Nov 7, 2016 09:27
|
|
|
theflyingexecutive posted:I agree with everything you've posted and I agree that his model's #1 source of error is input sanitization. I listen to every 538 podcast and read every article; every writer is couching their Clinton prediction with a "trump could win with significant and systemic polling error". This is technically a true statement but the chances for such an error benefitting trump over Clinton are minuscule I agree with you and I think Nate's lackeys would agree with you too. I just don't see how to code "Yo there's going to be historic Latino turnout" into the model he made months ago. Like you could assume that might happen based on Trump's rhetoric, but to put any kind of numbers on it for each state would be complete out-of-thin-air guesses. Turnout, in general, seems like it's incredibly hard to predict. Trump is also putting a big strain on it since Nate's model is tacitly assuming Trump has, on average, a similar GOTV operation to past presidential candidates even though it's clear he doesn't. That could also have been predicted, but again we have no way to know to what degree that affects different states. I feel like Nate's model has some problems, but it's also running into the buzzsaw of this election being an outlier. I think we will probably end up seeing Clinton outperform her polls in all the key states while Trump broadly underperforms his polls in every state. It's just tough to know what that's going to be like. So Nate's model is assuming that polling error is equally likely in either direction which I think from an objective numerical standpoint is the best it can do (though, admittedly, the model's fundamentals could be stronger). I wish people would use the model as it was intended to be used, as part of the basis for conclusions about the outcome of the election along with other information. Instead people really want the model to be an all-encompassing oracle rather than a tool for making sense of public polling. I blame Silver for that a bit because his PollsPlus forecast expands the scope of what the forecast is beyond what it should be so people start assuming things like turnout are taken into account in a detailed way when it's really just Polls+DubiousEconomicCorrelations ErIog has issued a correction as of 09:35 on Nov 7, 2016 |
|
#
?
Nov 7, 2016 09:31
|
|
|
I think "this election is an outlier" is a dumb defense and I wish Nate would stop using it. We have so few elections that all of them will seem to be outliers, statistically speaking. And for all the huffing and puffing over the Trump candidacy, we're pretty much on par for 2012 redux, give or take a couple states. If your model worked off of good assumptions, then it would have survived the Trump phenomenon, as every other aggregator has.
|
|
#
?
Nov 7, 2016 09:36
|
|
|
When the models converge, does that mean they all get as volatile as the now-cast, or does it mean the now-cast becomes as stable as the others? Because some New Hampshire polls just single-handedly bumped up Clinton's national odds by 2%. Could be interesting to see what the final massive poll dump does to Nate's model.
|
|
#
?
Nov 7, 2016 09:37
|
|
|
exquisite tea posted:I think "this election is an outlier" is a dumb defense and I wish Nate would stop using it. We have so few elections that all of them will seem to be outliers, statistically speaking. I feel like candidates with record-setting unfavorables and one candidate having no GOTV to speak of makes the "this election is an outlier" defense a pretty good one. Supercar Gautier posted:When the models converge, does that mean they all get as volatile as the now-cast, or does it mean the now-cast becomes as stable as the others? Because some New Hampshire polls just single-handedly bumped up Clinton's national odds by 2%. Could be interesting to see what the final massive poll dump does to Nate's model. NowCast is just, "if the election were held today with all the current polls." That's why it's so volatile. On election day the NowCast and default view will be 100% the same. ErIog has issued a correction as of 09:43 on Nov 7, 2016 |
|
#
?
Nov 7, 2016 09:38
|
|
|
|
| # ? Jun 1, 2024 20:48 |
|
|
ErIog posted:NowCast is just, "if the election were held today with all the current polls." That's why it's so volatile. I know they're going to be the same. What I mean is, what will the nature of their convergence be? The now-cast has been capable of ridiculous single-day jumps, and the others have not been. Basically I'm wondering what severity of change is theoretically possible for tomorrow's polls to bring.
|
|
#
?
Nov 7, 2016 09:42
|
|