|
Harrow posted:He explains what that is here: http://fivethirtyeight.com/features/election-update-why-our-model-is-more-bullish-than-others-on-trump/ Keep in mind this was when 538 was at like 85% chance of Clinton victory and other aggregators at 90-100%, so the raw discrepancy Nate was explaining was much smaller than it is now. Joepinetree posted another in depth critique of 538's methodology that was useful. I haven't seen a completely clear explanation for how 538 adjusts the raw poll numbers (is this the proprietary part?), but keep in mind Nate has talked about the GOTV disparity being probably good for a 1-2% difference between votes and polls and I'm almost certain the model (rightly) includes no adjustment for that. I just mention that to remind that adjusting poll numbers for some factors as they do and not making any guess for an adjustment related to GOTV or any difference between the campaign machines themselves will almost certainly push the model towards Trump relative to what we expect the final votes to be, from everything we know about the relative strength of their campaigns. I suspect this is something Nate will specifically talk about when Clinton wins.
|
# ¿ Nov 3, 2016 15:29 |
|
|
# ¿ May 10, 2024 01:48 |
|
Chokes McGee posted:The problem with this is that other aggregators aren't adjusting for that either, but my sample size is Princeton, so uh probably a little bias there Oh yeah, bringing up GOTV wasn't meant an explanation for why 538 differs from the others, more a reason to stop freaking out so much. Everyone expects Clinton to do a better job at GOTV (even Nate), everyone has either an overwhelming chance of Clinton victory or a good chance (even Nate), and everyone expects the GOTV difference to make a measurable difference in votes compared to polls that favors Clinton and are not including it in their models (even Nate).
|
# ¿ Nov 3, 2016 15:38 |
|
Blue Footed Booby posted:My favorite game in the series by far is 12, pre zodiac job system. A voice of sanity, thank god. 12 gets way more poo poo than it deserves. Definitely not my favorite, but up there in the mix of good to great, way ahead of 13, the NES ones, and 8. Like this dumb fucker: Agrajag posted:12?! the literal death knell of, and the harbringer of, all the poo poo that has come since? the originator of the japanese hair band protag? get out.
|
# ¿ Nov 3, 2016 15:53 |
|
Cingulate posted:That's a really good summary, particularly if you're not a stats guy yourself. Those are Nate's own explanations. Here is a critique from someone else, which raises a few more problems or potential problems.
|
# ¿ Nov 3, 2016 16:04 |
|
Cingulate posted:I know. That's what I said: it's a good summary. A lot of people here do not know or understand it enough and still post about it. Sorry, my point was there are probably other factors that are important but either not easy to explain in a short article or potentially unflattering to 538's method that Nate did not bring up. I do agree that the correlation between states is probably the biggest deal, but the weighting and adjustment of polls by 538 is another big deal that Nate didn't touch on.
|
# ¿ Nov 3, 2016 16:11 |
|
Why not both?
|
# ¿ Nov 3, 2016 16:32 |
|
CascadeBeta posted:Does RealClear have a pollster rating on 538? 538 isn't a pollster so without looking, I'm going to guess no
|
# ¿ Nov 3, 2016 20:26 |
|
mcmagic posted:The standard we're talking about is electability. And by that standard she's NOT more qualified or accomplished and is dropping hourly. Despite having been through several rounds of this in just the last week I've been lurking... Sanders is probably less appealing to moderate Republicans than Clinton, due to being quite a bit further left. If there's any significant Republican vote for Hillary, like even just a few percentage points siphoned off more than usual, that's a big advantage of hers compared to him. He also has easily attackable policies that non crazy people can believe: "you thought Obamacare prices were bad, Bernie wants to double your taxes to give everyone free college!" *strategic shot of a diverse group of young people slacking off* I don't think it's crazy to think old and wealthy white people would be even more against Sanders than Clinton. No idea on the various groups of minority voters, but they and women would also probably be turning out less, if the primary is anything to go by. OK, that's all I'm going to say on it, because I know this can go on for pages and pages.
|
# ¿ Nov 3, 2016 20:33 |
|
Pull up pull uppp!
|
# ¿ Nov 3, 2016 20:41 |
|
This is really interesting, thanks for posting. Any more interesting info about breaking down supporters by primary candidate there (and if so can I get a link)?
|
# ¿ Nov 3, 2016 20:56 |
|
Soothing Vapors posted:A reply to that guy from a Trump supporter: It is a joke (I haven't checked the Twitter but it must be a joke lol)
|
# ¿ Nov 3, 2016 21:04 |
|
The Insect Court posted:http://blogs.reuters.com/talesfromthetrail/2016/07/01/belatedly-what-sanders-supporters-say-about-race/ Thanks for the link anyway. Yeah, I think the interesting part for me would be seeing how saner Republicans (non Trump supporters during the primary) stack up to Clinton supporters in various ways. Any indication I can find that a good chunk of Trump support is plugging their nose and really not being happy about it will make me feel better about people, even if they end up voting for him in the general in the end. Also makes me hopeful that Hillary might really be pulling a meaningful number of R voters (and independents).
|
# ¿ Nov 3, 2016 21:12 |
|
BarbarianElephant posted:Absolutely. However there is a real danger of the premiums increasing to levels where ordinary families will not be able to afford them. This happens particularly with freelancers and very small businesses, where the person in question earns too much for subsidies, but does not get any work health insurance. I remember a more in depth article by someone in the administration, but this is all I could find right this second that debunks the fact that many people will see increases. Subsidies will rise, and prices are where they were expected to be anyway because insurance companies under priced stuff initially : http://www.npr.org/sections/health-shots/2016/10/24/499190020/rates-rise-again-for-obamacare-health-plans-but-so-do-subsidies Also remembered reading that the biggest increases were in states with the lowest current premiums. Am I missing anything big or is all that relatively true? Like you say, some people will have prices go up somewhat, but most will be relatively unaffected?
|
# ¿ Nov 3, 2016 22:10 |
|
Geostomp posted:This election has very conclusively proven that morality, qualifications, and enen basic levels of connection to reality pale in comparison to having an R next to your name and "telling it like it is" (read: legitimizing the hatreds held by white men). I'm sure this is not a revelation to anyone else, but I have just one person on Facebook that constantly spouts garbage (probably legit dealing with mental illness plus social isolation from living in China for 1+ years, as he was a very normal lefty person all growing up and at least through college). Dude repeated some EMAILS, "why did Huma marry Weiner in the first place?!", and ended with "if you're voting for Hillary, do you identify with a set of principles, or just the letter D?" and I literally just can't even. I wanted to post the expanding but it wouldn't have meant anything to anyone besides me and it wouldn't have made me feel better
|
# ¿ Nov 4, 2016 15:30 |
|
PerniciousKnid posted:Why assume a regression to the mean instead of a continuing tend? If there's any mean regression, I would think it's toward 50/50, as people's memories fade after whatever news events previously boosted Clinton's numbers. Why would you assume 50/50 is the true mean that it would regress back to? I also haven't carefully checked, but whoever was saying it's oscillations around a 74% Clinton win is I think seeing coincidence. The up trends were all around obvious events (convention, debates) and then as time passed those bumps faded, right? As enthusiasm waxes and wanes, different proportions of voters pass the likely voter threshold, so it isn't like the model is saying every debate converted Trump voters to Hillary and then they changed their minds back. Not sure we should read the median chances as the true probability either, though it is more reasonable than assuming 50% if those are the only options. Finally, I like that Cingulate is sticking up for the model, because I do think many are not being fair to Nate or it. Talk of him changing things in the model day by day is not at all fair. But (1) that doesn't mean the model doesn't have problems or (2) that Nate's/538's punditry isn't questionable.
|
# ¿ Nov 4, 2016 17:33 |
|
Mel Mudkiper posted:Not exactly, bad months for Hillary like June/September took the chances down to the low 60s. I see what you mean, and me calling it coincidence was not quite correct (as in not quite what I meant to say). I meant more that assuming the true probability is around the median is not quite right, because there have been meaningful events that have pushed the probability in each direction. Since we can't assume each piece of news on either side is equivalent, we can't say something like "the first debate was great for Hillary so she reached 85% and the first FBI email non-indictment press conference was great for Trump and Hillary's chances went down to 65%, so the true probability is 75%". If there were no relevant events happening ever, I think we could take the mean and be fairly happy with that as an estimate of the true probability. But the odds have been pushed in both directions numerous times and by unknown deviations from the true probability, so we can't easily infer the "baseline"/true probability just by taking the mean. It took way more words to write out, but that's the sort of thing I was trying to bring up originally. As for 538 vs other aggregators, absolutely. I buy into the reasons why it's too uncertain and too bullish on Trump relative to others, and think those are legit. But I agree with Cingulate that it isn't shook Nate pushing the levers each day differently, or making some huge change to the model because he was embarrassed by the primaries. It's assumptions or methods that worked well in 2008 and 12, but that don't deal well with high undecided voters, or something like that.
|
# ¿ Nov 4, 2016 17:52 |
|
PerniciousKnid posted:Because there are two candidates, so absent any strong feelings about the candidates there's a 50% chance of choosing either. If you did a bunch of analysis of voter demographics and such you might determine a more refined number, but it would still be much closer to 50% than 74%. This isn't a coin flip with two equal outcomes, though. There are all kinds of predictors (economic, incumbent party, etc) that make one or the other generic D or R candidate more favored. Beyond that, these are two actual, non-hypothetical candidates. Even in the absence of any campaigning or outside events, I highly doubt voters would split 50/50 between the two. Keep in mind any deviation from 50/50 voting will make for a much larger change in probability of winning. A true voting probability of 50.1 to 49.9 is a 100% chance the 50.1 wins. If the preference for Clinton is 52 to Trump's 48, and we're pretty confident in that, a very close to even split in voter preference is a very high probability of Clinton win. I think you might be partially conflating those two.
|
# ¿ Nov 4, 2016 17:59 |
|
Cingulate posted:They were behind for the vast majority. Was that a 4 point error on the day of the vote, day before, week before? And is that an aggregate number or a particular poll? I am honestly asking because I don't know.
|
# ¿ Nov 4, 2016 18:02 |
|
PerniciousKnid posted:Sure, if you have some reason to be confident in that. You could also point out that Trump has been making bad press constantly to boost Clinton's numbers, and this might be only the first or second week since the primaries that Trump has stayed out of the spotlight. I wasn't trying to argue actual numbers. At a bare minimum, assuming the candidates really don't matter, we should have some estimate for the probability the incumbent party picks up a third consecutive term. It won't be a coin flip. We should have estimates for how economic factors affect the incumbent party (and also D and R) win probabilities. We have party registrations and demographics to estimate turnout. No election is going to be a literal coin flip, even if you ignore the candidates. If being the incumbent party seeking its third term consistently hurts you by 1 point, we now have an expectation of a 50.5 to 49.5 split, leading to some expected probability that the 50.5 side would win, given the uncertainty in our estimate of the "incumbent third term penalty." Then we see Dems have a few point registration advantage but a few point turnout disadvantage, so we can bake in those estimates. And on and on. A true coin flip win probability is just highly unlikely, so I don't think it makes any principled sense to assume it.
|
# ¿ Nov 4, 2016 18:16 |
|
Cingulate posted:IIRC, on the day or two before, the mean of polls was 50:50, and Leave won by 4. Thanks, good to know.
|
# ¿ Nov 4, 2016 18:18 |
|
Tiny Brontosaurus posted:Empirically I can hear every historian in the world grinding their teeth right now. Haha yeah
|
# ¿ Nov 4, 2016 19:09 |
|
Harrow posted:Someone post another picture of Michelle Obama doing healthy living things and then it can be politics for another page or so
|
# ¿ Nov 4, 2016 19:12 |
|
Cingulate posted:Hm. By whom? I assume you're not thinking about books by Thomas Kuhn, Karl Popper, Peter Singer, Steven Pinker, Karl Marx or Francis Fukuyama? (Not necessarily citing any of them approvingly.) You were talking about something like social/societal progress over time, right? Ignoring differences between regions in the same time, taking any particular issue, I think it is difficult to see a consistent upward trend from pre-history to now. Persecution of religious or ethnic or sexual or political minorities has risen and fallen more than simply improved linearly. Ancient Greece, some eras of Rome, the Enlightenment, French Revolution and Napoleonic era, and yeah during the latter half of the 20th century have been good in some respects (and bad in others). Caveat that I am not a historian and not a lay expert by any means, just a history nerd. But reading about the Roman republic and early French revolution is kinda sobering in how they had a ton right hundreds and thousands of years ago.
|
# ¿ Nov 4, 2016 19:28 |
|
Combed Thunderclap posted:But this is the thing, I'm thinking of things in terms of economic and social development whereas if you want to wade into whether or not a society is better re: rights then things get very subjective very quickly. But I honestly can't think of times when more people were more well-off and less dead/literally starving than when they are now, or literally any time period in the past where I'd want to be other than the present day. The Roman Republic and early French Revolution are cool time periods but they were still pretty lovely for lots of people. Yeah, I also think this is fair in general. It still is frustrating that we had good ideas about how to be less lovely human beings 2500 years ago and have not been able to put it into practice consistently.
|
# ¿ Nov 4, 2016 19:40 |
|
Cingulate posted:Ancient Greece and Rome were good with regards to nice marble statues and architecture, but they were also slave societies. Look at how Sparta treated the helots - hunting human beings like animals in an annual ritual. With regards to Human rights, most of the most abhorrent civilizations of the 19th and 20th century were better than Rome. As I said, they were good in some ways and bad in others. I don't even mean that flippantly. No one gave a poo poo about homosexuality. For large stretches of their history, Romans officially recognized and protected religious minorities. The Roman republic had peaceful political disagreements and power transfers for most of its history. The dissociation between political and religious rule has taken hold and evaporated at multiple points in history. Tolerance of religious minorities and homosexuals, as two specific examples, have not been some uninterrupted progression, and we shouldn't take it for granted that they will be. As for material well being, of course that is important, as is the declining amount of personal violence. I specifically brought up counterexamples.
|
# ¿ Nov 4, 2016 19:54 |
|
Cingulate posted:The difference is that while there are still factual slaves in the US and Dubai, Rome was a whole empire built entirely on slavery, and on constantly invading its neighbors. Rome is worse than the US in the same way that the antebellum south was worse than Sweden. I don't know about others, but I'm trying to suggest that these aren't monolithic "better than" and "worse than" judgments. You would rather be a homosexual in ancient Greece than in multiple countries in Africa and the Middle East today. You would rather be Jewish in ancient Rome than in Europe almost anytime in the past 1000 years. You would rather be an atheist in France in the 1790s than in the deep south in America today. For a less socially charged area, I suspect it won't be a long term phenomenon (please please), but declining basic scientific literacy and trust in science is another example where we are at risk of regressing. Etc etc. Having a steadily improving life expectancy does not negate other ways in which we have or have not made progress. Edit: If the counterpoint is that 2016 is better than 16, then that isn't the same thing as steady progress, and we could pick other pairs of dates to suggest steady decline or stasis. sourdough fucked around with this message at 20:19 on Nov 4, 2016 |
# ¿ Nov 4, 2016 20:16 |
|
Ice Phisherman posted:Actually being a Jew in the Roman empire was pretty bad too. See several suppressed revolts and the enormous body counts attached. In the BCE years, it was an officially legal religion and afaik there was not any systematic persecution. I believe the Jewish revolts were more about political power, but am not an expert. That seems qualitatively different than later persecutory anti-Semitism. Edit: That said, we're straying pretty far from USPol haha. Happy to hear if I have any of that wrong because I am absolutely not an expert, but probably otherwise fits in a history thread so I'll shut up about it sourdough fucked around with this message at 20:31 on Nov 4, 2016 |
# ¿ Nov 4, 2016 20:28 |
|
WampaLord posted:Here's a good answer And other ideas about deviations between 538 and other models from someone not affiliated with 538: http://predictwise.com/blog/2016/09/poll-aggregation-fight/
|
# ¿ Nov 4, 2016 20:32 |
|
weekly font posted:I understand I just think it's flawed as hell I really don't think this is a key problem, though perhaps 538 assumes too strong a correlation. Polls shouldn't be moving randomly, and if you see 5 states move 2 points towards Clinton after the first debate, you should assume a sixth state with no polling done yet also likely moved that way. Whether the model weights that too heavily or doesn't properly throw out those correlations when actual data comes in, I'm not sure.
|
# ¿ Nov 4, 2016 20:35 |
|
ImpAtom posted:It's largely this. Even if logically Nate Silver's prediction means Clinton will probably still win nobody wants to acknowledge how frigging terrifyingly close it is. Most other models/aggregators don't have it as terrifyingly close at 538 though. Wondering why doesn't mean you're denying reality or making GBS threads on 538 for dumb reasons (though you also might be!).
|
# ¿ Nov 4, 2016 20:57 |
|
Spacebump posted:Don't more people vote on election day and isn't Trump expected to attract a lot of voters that don't normally vote? People that don't normally vote but decide to randomly are more likely to vote on election day. Nevada and Florida, at least, have huge fractions of their vote happening early. I think most states with lots of early voting in 2012 had >1/3 of their votes cast early. I think Nevada was 2/3? And no, as has been mentioned, the only evidence we have is that a lot of Hispanic voters and others that are either first time voters or only voted in one of the last three presidential elections (which would filter them out of most polls of likely voters) are turning out for Hillary.
|
# ¿ Nov 4, 2016 21:00 |
|
AriadneThread posted:this is my chance to promote the ancient world podcast, the current segment bloodlines, starting around episode 13 we end up in judea and the next couple episodes focuses on the history of the area under the roman empire Awesome, I will check it out. I have hit most of the usual recommendations for history podcasts, but haven't found a lot I love.
|
# ¿ Nov 4, 2016 21:02 |
|
PerniciousKnid posted:Claiming a candidate has a 80-90% chance to win with a lead of 1.5% is absurdly overconfident in polling accuracy. Prove it.
|
# ¿ Nov 4, 2016 21:21 |
|
PerniciousKnid posted:Well, RCP final average was O+7.6 in 2008, final result was O+7.2. Poll average in 2012 was O+0.7, actual was O+3.9. In 2004 the average poll was B+1.5, actual was B+2.4. Those are differences of 0.4, 3.2, and 0.9. So aggregate polling error of a couple points is entirely plausible. Not likely, mind you, but significantly possible, especially when you consider the unique unlikeability of both candidates this year, etc. You are conflating polling errors and win probability errors. They are not the same thing, as I tried to explain to you earlier. Edit: This is actually several points rolled together, though. As a very simple exercise, assume deep south goes +5 for Trump beyond polling. Does Hillary's win probability drop? Beyond that dumb toy example that shows how national margin is not the same as win probability, if Hillary is up 3 points in national polls and polls are off by 2 points, she still wins. Polling errors are also not likely to be truly randomly distributed; in none of your examples did polling errors flip the winner. sourdough fucked around with this message at 22:00 on Nov 4, 2016 |
# ¿ Nov 4, 2016 21:56 |
|
PerniciousKnid posted:Win likelihood is a function of what you think the polling error is. And goons make way too much of the electoral college vs popular vote thing. When they're generally highly correlated. One or both of us is confused about what the other is arguing. Hypothetical here, but a polling error of 2 points that reduces Hillary's lead from 3 points to 1 point is a massive polling error. It cuts her advantage by two thirds. A 1 point lead in the popular vote gives you an overwhelming probability of winning the presidency. That is the exact opposite of thinking the electoral college and popular vote are not highly correlated. If you can show me somehow that a confident, aggregated estimate of a polling lead of 1-3 points empirically does not have a high win probability, I would like to see it. That's the kind of thing that could convince me 538 is in the right realm of uncertainty. What you've shown so far is that polling errors of 0.5-3 points have still let the polls get the winner right. Somewhat of a tangent, but I also suspect polling error magnitude is not independent of the difference between the candidates. Has anyone seen data specifically on that? I feel like 538 blog or somewhere looking at polling errors over the last x elections might have given some hint about this. Let me see if I can find it...
|
# ¿ Nov 4, 2016 22:49 |
|
Cingulate posted:I'm not so much wondering what might be the facts at hand, but why it is controversial. Maybe it's wrong, let us leave that aside - why is it offensive for me to claim it? I did not find it offensive, fwiw. I just think there are many measures by which we have not had consistent progress over time and that they are still the kinds of things we are trying to work on today as societies, so they are relevant.
|
# ¿ Nov 4, 2016 22:51 |
|
Charlz Guybon posted:And it's why Florida is a must win state for Hillary, it innoculates against some northern defections. That is not what "must win" means. Hillary is more likely to win each of Ohio, Iowa, Utah, Arizona, or Georgia than Trump is to win Michigan. If you want to take some of those aggregator probabilities literally, using the first column (NYTimes, which is less worried about Clinton than 538 but more worried than the other aggregators), Hillary is more likely to win both Ohio and Iowa than Trump is to win Michigan. Edit: Also, if you want a laugh, guess which column is 538 sourdough fucked around with this message at 00:47 on Nov 5, 2016 |
# ¿ Nov 5, 2016 00:45 |
|
Wizard Master posted:Can someone tell me if Trump is going to lose or not? This thread gets like 100 new pages per hour and it's mostly nerd poo poo
|
# ¿ Nov 5, 2016 00:51 |
|
PopZeus posted:What website is this chart from? NYTimes Upshot
|
# ¿ Nov 5, 2016 02:14 |
|
|
# ¿ May 10, 2024 01:48 |
|
pgroce posted:Full disclosure: I pulled AZ (and CO, to a lesser extent) out of my rear end. The general problem of accounting for locked-in votes remains a problem in 538's model, though, AFAICT. None of them do, as far as I know. Early voters are only reflected in anything polls-based to the extent that they should automatically pass any likely voter screen, regardless of their prior voting history or other things.
|
# ¿ Nov 5, 2016 02:16 |