|
Patter Song posted:In fairness, Nassim Nicholas Taleb isn't exactly "some rando." Oh, man, would love to see these two go at each other. Educated stupid each for a different reason. 
|
#
?
Sep 11, 2016 02:50
|
|
|
|
| # ? Jun 8, 2024 18:25 |
|
|
Prof. Lurker posted:http://projects.fivethirtyeight.com/2016-election-forecast/ohio/ 1) The PPP poll is great for Clinton because it shows a Clinton lead and it's from a pollster with a good track record. 2) The Quinnipiac poll is great for Trump because it shows a Trump lead and it's from a pollster with a good track record. 3) The Emerson poll is okay for Trump, and while it shows him ahead, his lead at less than 1%. 4) The YouGov poll is good for Clinton because it shows her up big but is from a lower-tier pollster. 5) The Survey Monkey poll is good for Trump because it shows Trump with a lead and because it is with a large sample, but is with registered voters, not likely voter. 6) The Monmouth poll is good for Clinton because it shows her ahead and is from a highly respected pollster, but has a small sample size. 7) The Marist poll is okay for Clinton because it shows a slight trend in her favor, but has the race tied overall 8a) The Ipsos poll is good for Clinton because not only shows a lead for Clinton, it also shows a trend towards Clinton from the previous poll, but is hampered by the really small sample size. 8b) The Google consumer surveys poll is good for Trump and shows Trump's lead holding over two polls. 9) The Gravis poll is good for Clinton because it shows her up slightly but is with registered voters, not likely voters. Overall, it looks like it matches up with the odds they're showing. If we break the polls down into tiers, based on the weight they're assigned, I think it'll help show why their forecast says what it says. Top tier: Trump 1 (Quinnipiac) / Clinton 1 (PPP) Middle tier: Trump 1.5 (Emerson, Survey Monkey) / Clinton 1 (YouGov) Lower tier: Trump 1 (Google Consumer Surveys) / Clinton 3.5 (Monmouth, Marist, Ipsos, Gravis) I gave a half point for the Emerson and Marist polls because they're each essentially a tie with a minor edge to one side or the other. The two best polls say contradictory things, the middle tier polls are slightly in Trump's favor, with one poll on each side showing contradictory information, and one essentially in the middle with a slight Trump edge, and the lower tier polls being favorable to Cilnton. The misleading part is, when it first loads, the default view displays ten polls, but four have very little impact on the overall percentages because they're superseded by newer polls from the same pollster, and three of those four polls are favorable to Trump. The remaining polls do favor Trump, but as soon as it is expanded, a bunch of Clinton-friendly polls show up. Based on their model's assumptions of the ratings and weight of the polls, I think it's showing an accurate percentage. Whether their model is right is a whole separate discussion, all I'm saying is that the percentages they display are in line with the polls they display.
|
|
#
?
Sep 11, 2016 04:29
|
|
|
Sorry for the double-post, but I also want to point out just how many of those polls are on the absolute low end of the acceptable sample size. They're mostly less than 900 respondents, some quite a bit smaller than that, which means that they've all got a pretty big margin of error. It's not surprising, since most of them look to be part of a larger polling effort that is broken out by states. In Ohio, the Ipsos polls are a good example of how small sample sizes can produce some pretty big swings that probably aren't happening. For a really wacky example, look at New Hampshire, where that same Ipsos poll had around 150 respondents and showed a 15 point swing from Trump to Clinton from one survey to the next. Obviously, that's out of the ordinary even for a sample size that small, but I think it well illustrates the problems with the absence of large sample size polling.
|
|
#
?
Sep 11, 2016 04:39
|
|
|
from following taleb on twitter for like two weeks before it became too much id say hes weird as poo poo
|
|
#
?
Sep 11, 2016 05:00
|
|
|
today in truth bombs
|
|
#
?
Sep 15, 2016 18:20
|
|
|
Tayter Swift posted:today in truth bombs Of course he had to stick to a two way race to make that point, because the number of people who vote with the two major parties vs third party is extremely important this year.
|
|
#
?
Sep 15, 2016 19:01
|
|
|
When people hear "independent", they interpret it to mean "moderate" or someone who could be persuaded more easily to vote for one party or the other, but that's just not the case. It's like the idea that Ohio, with it's status as a perennial swing state, has a lot more of those moderate, persuadable voters than the electorate at large, and that's not the case either. Ohio is a swing state because it's got roughly equal population of voters on each side and that has stayed pretty consistent over time. Up until this election, I was one of those folks who liked to identify as independent, but who voted Democrat pretty reliably. Trump and the Republican reaction to him has pushed me to finally identify as Democrat, but that's essentially just a ratification of what I was already doing. A lot of the folks that I talk with who identify as independents do so because they assume that someone who identifies as Democrat or Republican must vote party line 100% of the time or they identify that way as a protest against a subset of the policies of one party or the other, but when election time comes, it's clear which party represents their interests better, so they swallow their misgivings and vote as they usually do. I think that Trump is capturing those independents at such a high rate simply because so many of them are independent because of their dissatisfaction with the two-party system in general and he's catching a lot of protest voters. It'll be interesting to see if, after the debates, how that independent voter mix shakes out. Evidence suggests that Clinton is losing voters to third parties at a greater rate than Trump, and it'll be interesting to see how many of them stay with third parties if the election continues to tighten. This assumes that Trump's won all the independents that he's likely to win, and that his challenge is to win back those that identify as Republican but who have gone Democrat or third-party.
|
|
#
?
Sep 15, 2016 19:11
|
|
|
Tayter Swift posted:today in truth bombs some people say independents are good like the keys in Super Mario World but actually they're useless like the 1-ups in New Super Mario Bros.
|
|
#
?
Sep 15, 2016 19:18
|
|
|
Tayter Swift posted:today in truth bombs i've never heard of that brand of alcohol before
|
|
#
?
Sep 15, 2016 19:25
|
|
|
Azathoth posted:When people hear "independent", they interpret it to mean "moderate" or someone who could be persuaded more easily to vote for one party or the other, but that's just not the case. The notion that there's this huge excluded middle of moderate independent swing voters (who hate partisanship and want both parties to stop bickering and come together and attend to serious issues like The Deficit and Entitlement Reform) out there is a myth promulgated by the centrist third-way types who dominate elite political discourse. You see a lot of this when events cause Party ID numbers to change sharply. Obama gets rolled in debt ceiling negotiations, the number of people identifying as Democrats goes down. Obama kills Bin Laden, the number of people saying they are Democrats skyrockets. It's just the same group of people giving different answers to the same question depending on how they feel about their party on any given day. The total number of Democratic voters remains the same, it's just a bunch of them slosh between identifying with the Dems and pretending to be above-it-all independents, depending on whether their team is doing well or not.
|
|
#
?
Sep 15, 2016 21:40
|
|
|
Tayter Swift posted:today in truth bombs can't believe this guy is single
|
|
#
?
Sep 15, 2016 21:44
|
|
|
FMguru posted:It's been pretty well established for a while that the vast majority of "independents" are actually reliable voters for one party or another, they just like to call themselves independent because they like imagining themselves as sophisticated free-thinkers and not mindless party drones. It's objectively true, but the public really doesn't want to admit it, and certainly not the third of them who self-identify as "independent" because the lie is just more noble to believe. Elections always tighten, as all those "independents" come back home and vote for who they usually vote for, and for all craziness of this election cycle, I think that's gonna happen again. All those "independents" get to feel like they made an informed decision, even if the outcome of the decision wasn't in doubt for most of them, media gets to stroke their egos, and everyone's happier at the end of the day. The media trying to convince "independents" that they're really partisan and have just been lying to themselves is about as likely as them running a story mid-December about how Santa Claus isn't real. It may be true, but no one really wants to hear it.
|
|
#
?
Sep 16, 2016 00:16
|
|
|
Azathoth posted:The media trying to convince "independents" that they're really partisan and have just been lying to themselves is about as likely as them running a story mid-December about how Santa Claus isn't real. It may be true, but no one really wants to hear it. Well, let's not forget that there are also independents who just hold really strong opinions that are contradictory and/or crazy. Like Alex Jones followers who hated the Iraq war (and Bush), think the government is coming for our guns, smoke weed every day, and talk about how black lives matter is a hate group. Or the Bernie Sanders supporters who were also on the side of the those militia idiots who took over the bird watching station.
|
|
#
?
Sep 16, 2016 00:33
|
|
|
e_angst posted:Well, let's not forget that there are also independents who just hold really strong opinions that are contradictory and/or crazy. Like Alex Jones followers who hated the Iraq war (and Bush), think the government is coming for our guns, smoke weed every day, and talk about how black lives matter is a hate group. Or the Bernie Sanders supporters who were also on the side of the those militia idiots who took over the bird watching station.
|
|
#
?
Sep 16, 2016 04:55
|
|
|
http://www.clickhole.com/article/playing-it-safe-nate-silver-will-spend-next-month--4742
|
|
#
?
Sep 17, 2016 00:15
|
|
|
Concerned Citizen posted:can't believe this guy is single  Immediately heard the song in my head when I read the name.
|
|
#
?
Sep 17, 2016 02:17
|
|
|
Helical Nightmares posted:
the harry enten song has been stuck in my head for days now
|
|
#
?
Sep 18, 2016 02:22
|
|
|
Concerned Citizen posted:the harry enten song has been stuck in my head for days now Have another https://m.youtube.com/watch?v=kGNYgEQvXdY He's incredibly annoying but drat if they arn't catchy. In the I want to trepan myself kind of way.
|
|
#
?
Sep 21, 2016 10:26
|
|
|
posted some stuff on another forum, thought I'd throw it in herequote:screw it, time to bitch about 538 again. yesterday's "Clinton is leading in EXACTLY 272 EV" article:
|
|
#
?
Sep 23, 2016 22:37
|
|
|
The model doesn't seem too broken imo.
|
|
#
?
Sep 24, 2016 02:15
|
|
|
Bip Roberts posted:The model doesn't seem too broken imo. hmm yes, but have you considered, that it is??
|
|
#
?
Sep 24, 2016 03:12
|
|
|
gonna update my "goon self-hating" hypothesis to include nate silver. much like garrison keillor, goons hate someone who is better at shitposting and goonery than they are, because 
|
|
#
?
Sep 24, 2016 03:39
|
|
|
Their Now-cast has Hillary winning New Hampshire by 4.3% as of today with the five highest weighted polls being adjusted to Trump +1, Hillary+8, Hillary+5, Hillary+4, Hillary+5 based on the pollster's partisan lean, with the raw results being Hillary+2, Hillary+9, Hillary+5, Hillary+5, and Hillary+6. Just taking those five polls, adding up their overall margins and dividing by five, it puts the adjusted number at Hillary +4.2% and the raw number at Hillary +5.4%. While I think that unskewing polls is fraught with danger, I've always found their attempts to correct for overall partisan lean to be worthwhile, as it seems to help prevent cherry-picking of polls. From watching their updates, it seems that their model thinks that the polls this cycle are slightly biased towards Hillary, and I think that 1-2% is pretty much on target with what they are saying about the national election, so I don't find it surprising that they're seeing a 1.2% correction towards Trump in New Hampshire. That said, whether Hillary is at +4.2% or +5.4% in New Hampshire, I think a lot of the angst towards 538's model comes from their use of percentages for state results in a way that a lot of people, myself included, find baffling. New Hampshire is a perfect example of this, as Hillary's currently got a 72% chance to win in the Now-cast but has led in every poll taken since July, aside from one that had a sample of 85 likely voters. The only way that Trump wins in New Hampshire, if the vote is held today, is if the polls are all completely skewed towards Hillary by a margin that seems almost impossible, yet they're saying that there's somewhere between a 1 in 4 and a 1 in 3 chance of that being the case, or if we had a portal into the infinite multiverse and could look in on 100 randomly selected worlds where voting was occurring today, but everything else is the same, Trump would win New Hampshire in 28 of those poor, besotted, Trump-crazed worlds. For a more extreme example, take Minnesota (since as a DFLer, I take exception to Trump having a shot here). Their Now-cast has Hillary at +6.5% and a chance of victory at 84.6%. That's about 1% lower than Obama's margin over Romney, which seems pretty accurate to me, given the number of homemade "Trump That Bitch" signs that I see dotting the countryside and the number of Bernie 2016 bumper stickers I see in town. Also, the best result that Trump has had in any poll here is +5 Clinton from back in January, or +6 Clinton from a recent poll. However, the percentage suggests that in that same infinite multiverse, 15 of the 100 elections would come out with a Trump win. Unless they put the odds of a small nuclear weapon detonating somewhere in Hennepin or Ramsey counties at 1 in 8, it's hard for me to see how Trump has any chance in an election held today. To flip it around, the model puts Trump's chance of winning Minnesota at about the same chance as Hillary winning Montana, in an election held today that is. In ever more flagrant stupidity, if the election were held today, Hillary has a 5% chance of winning in North Dakota. I would love to know how that number is not literally zero. Is it that there's a 5% chance that the entire state takes all that farm subsidy money and gets pee-your-pants drunk on Everclear and like 5 teetotaling Democrats show up to vote while only 4 very hung over Republicans remember it's election day? Their percentages are useful in evaluating states relative to each other, in that Colorado and New Hampshire are more likely to vote Trump than Michigan, Wisconsin, and Minnesota, but thinking about them as actual odds it's pants-on-head stupid. I've never seen a good explanation for how they're supposed to be used either as they seem to think of them like the odds of making or missing a field goal in football, but it's hard to reconcile their numbers with polling, history, and the overall partisanship of a state. At the very top, in their Clinton vs. Trump as the next president prediction odds, I think they come out pretty accurate, as they've got Trump as having a 35.5% chance in their Now-cast. This pairs decently with the 272 EV article and what they consider to be a 26.7% chance of Trump winning Colorado, where a couple polls have shown him ahead, but the aggregate shows him behind. That, plus a few percentage points for some crazier outcomes, such as winning Maine's 2nd District and New Hampshire or a shocker in Pennsylvania, seems like a good approximation of the status of the race, though I'd love for them to do some kind of an article explaining how their percentages make even a lick of sense outside of the heavily polled swing states.
|
|
#
?
Sep 24, 2016 04:29
|
|
|
Azathoth posted:Their Now-cast has Hillary winning New Hampshire by 4.3% as of today with the five highest weighted polls being adjusted to Trump +1, Hillary+8, Hillary+5, Hillary+4, Hillary+5 based on the pollster's partisan lean, with the raw results being Hillary+2, Hillary+9, Hillary+5, Hillary+5, and Hillary+6. Just taking those five polls, adding up their overall margins and dividing by five, it puts the adjusted number at Hillary +4.2% and the raw number at Hillary +5.4%. While I think that unskewing polls is fraught with danger, I've always found their attempts to correct for overall partisan lean to be worthwhile, as it seems to help prevent cherry-picking of polls. From watching their updates, it seems that their model thinks that the polls this cycle are slightly biased towards Hillary, and I think that 1-2% is pretty much on target with what they are saying about the national election, so I don't find it surprising that they're seeing a 1.2% correction towards Trump in New Hampshire. we're out of egg mcmuffins sir
|
|
#
?
Sep 24, 2016 06:08
|
|
|
This deserves to be out of the main thread, but I don't wanna dig for the polls thread. Anyway, here's the exact reason why the LA Times poll is bad https://np.reddit.com/r/PoliticalDiscussion/comments/53gcwh/polling_megathread_week_of_september_18_2016/d8137d9 quote:Trump is not gaining at all.
|
|
#
?
Sep 25, 2016 09:29
|
|
|
Reason further down that Reddit thread, the poster also says the young black Trump guy makes over $75k a year. I don't know how they are able to dig out that data, but  if that poll's results are being thrown way off because of some rando guy.
|
|
#
?
Sep 25, 2016 19:51
|
|
|
Adar posted:This deserves to be out of the main thread, but I don't wanna dig for the polls thread. Anyway, here's the exact reason why the LA Times poll is bad I'm super confused how Reddit works and why that that post was removed.
|
|
#
?
Sep 25, 2016 19:56
|
|
|
Anywhere that includes the LAT/USC thing,that while really interesting for a pile of reasons, as a poll to be related to other polls in the creation of a national average is being completely disingenuous. I'm looking at you, 538.
|
|
#
?
Sep 25, 2016 20:00
|
|
|
Schnorkles posted:Anywhere that includes the LAT/USC thing,that while really interesting for a pile of reasons, as a poll to be related to other polls in the creation of a national average is being completely disingenuous. The sorts of analysis 538 is doing fundamentally puts them in a damned if you do/damned if you don't type situation WRT being too picky-choosey with polls or being too ideologically stringent with what polling methodologies to accept.
|
|
#
?
Sep 25, 2016 20:07
|
|
|
Schnorkles posted:Anywhere that includes the LAT/USC thing,that while really interesting for a pile of reasons, as a poll to be related to other polls in the creation of a national average is being completely disingenuous. While I'm not a huge fan of the 538 re-weighting of polls, their "house effect" weighting does mean that the LAT poll basically counts as a tie race for the average rather than Trump +5 or whatever.
|
|
#
?
Sep 26, 2016 01:17
|
|
|
Bip Roberts posted:The sorts of analysis 538 is doing fundamentally puts them in a damned if you do/damned if you don't type situation WRT being too picky-choosey with polls or being too ideologically stringent with what polling methodologies to accept. However, this poll's methodology presents a unique opportunity to be flawed in non-obvious ways. Some quick googling and some even quicker math tells me that their poll should include roughly 90 to 100 African American male respondents, and if it's true that some iterations of the poll really includes only 2 African American male voters without a college degree, then it really calls into question their entire methodology.At a glance, the poll appears to be one of the larger ones based on sample size, but it's behaving like one of the smallest. The idea of re-interviewing the same respondents throughout is really interesting as a concept, given how trend from one iteration of a poll to the next can be more informative than the overall results of the poll itself, but if they aren't able to get enough responses to accurately weight the poll, it calls the viability of the whole model into question, at least at the sample size they have. In a poll with more traditional methodologies, this would be a transient issue with a single poll iteration, but in this poll, it is a systemic problem. I'd really love to hear someone at The Upshot or 538 react to this.
|
|
#
?
Sep 26, 2016 01:37
|
|
|
there's no reason to include it at all. the model isn't really built to accommodate radical polling methodologies and it's not like there's a surfeit of national polls
|
|
#
?
Sep 26, 2016 03:33
|
|
|
theflyingexecutive posted:there's no reason to include it at all. the model isn't really built to accommodate radical polling methodologies and it's not like there's a surfeit of national polls The real problem comes from the fact that the 538 model is really only built to deal with polls that are flawed in the ordinary ways, like "Our likely voter model assumes that white voters without a college degree will turn out at 80% this election cause Trump's gonna make polling great again! USA! USA!" or "Our sample size is less than 100 people" or "We only gather results by use of paper forms hand-delivered to respondents by our dedicated team of street urchins, not newfangled gadgets like the telly-phone". Accounting for crap like that is why places like 538 exist, and if they can't, then you may as well just total up all the polls and average them out, since it's essentially what they're going to be doing anyways. The flaw in the poll's model could be overcome by various tweaks, such as having a much larger pool of respondents to try to prevent one single person from having such an outsized effect or not relying on the notoriously inaccurate responses concerning who people voted for in the last election, but it's far too late for that now. If anything, it seems like they tried something novel and it ended up having a whole bunch of unintended consequences, which is to me a reason to rethink the polls fundamental assumptions and iterate to make it better, not discard it interely. I'd like to see it come back in 2020 if they can address the problems.
|
|
#
?
Sep 26, 2016 04:25
|
|
|
Bip Roberts posted:The sorts of analysis 538 is doing fundamentally puts them in a damned if you do/damned if you don't type situation WRT being too picky-choosey with polls or being too ideologically stringent with what polling methodologies to accept. 538 is unique in all the models in some of the polls they choose to accept. It's the source of most of their "uncertainty" in relation to models like the upshot.
|
|
#
?
Sep 26, 2016 04:37
|
|
|
Also my understanding of 538's overall methodology could be wrong, but realize that the LAT/USC poll is a wild outlier, and more importantly, not really a poll. They do poll "unskewing" based on the overall average of all polls and polls distance from that on average [I think. Harry's described it and that's my recollection. This could be wrong and it might change how I feel about it, but I'm not sure what other methodology that isn't just wild unskewing allows you to do what they're trying to do.] I.e. the poll, because they choose to allow it, is still effecting the overall picture and how they choose to "unskew" polls. It's muted, sure, but still there. e: Words, it's not weighting. 
Schnorkles has issued a correction as of 04:44 on Sep 26, 2016 |
|
#
?
Sep 26, 2016 04:40
|
|
|
I don't think the methodology nor poll itself should be blacked out, just not included in a model unequipped to handle it, especially when it's unknown what parameters it needs for optimal functioning. A fundamental aspect of polling is that you're constantly pulling in new subgroups of the population and that randomness serves to reduce the margin of error. I think (but don't have enough quant knowledge to confirm) a tracking poll would have that set initial MoE and then a compounding MoE for every day's cohort. This would be reduced if they just reported numbers averaged over the week for the entire pool, but that defeats the purpose of a NUMBERS EVERY DAY! poll from a publicity standpoint. Plus I'm wagering that if you're included in this poll, you have a sort of pressure to turn into a pundit and (accurately or not) change your likelihood of support every week. I also think that people aren't good at self-reporting probability of voting. They also acknowledge the "I don't remember for whom I voted" effect, but have no mechanism for correcting this. I think this poll is really terrible as a traditional poll, but really excellent for the other question they ask, which is the expected winner question. I also think this will be a great resource for developing swing voter theories; my personal one is that someone who changes their mind week to week is too stupid and ADD-ridden to actually complete the voting process
|
|
#
?
Sep 26, 2016 05:24
|
|
|
Anyone looked at those state by state google consumer surveys? Some of those results are down right wacky. I think I saw a couple like Hillary +1 in Arkansas and +11 in Kansas yet it somehow gets a B in pollster ratings. 
|
|
#
?
Sep 26, 2016 05:33
|
|
|
Schnorkles posted:538 is unique in all the models in some of the polls they choose to accept. It's the source of most of their "uncertainty" in relation to models like the upshot. When you do meta-analyses, you have to make the call as to what observations best fit your analytical model. Sometimes you tweak the model to allow for a greater range of data, sometimes you throw observations out because they don't meet your criteria. Shook Nate is leaning hard on the former to include a bunch of no-name pollsters with obfuscated data adjustments (LV screen hacking, illegitimate crosstab extrapolations, arbitrary turnout weighing, insufficient subgroup breakout analyses, enthusiasm weighing) so that 538 updates more frequently than RCP or NYT. Why else would he post a whiny defense of USC/LAT? He gets to piggyback on their NUMBERS EVERY DAY system. Why else would he include the Ipsos state breakouts when no other aggregator will? He gets 51 bonus data points including state "polls" of states that no pollster would ever waste time calling and every time one of those breakouts posts a weird result because it isn't crosstabbed and is barely three digits of LVs, Kellyanne or even orange hitler himself tweets about it. Shook Nate is either too dumb to understand that a lot of polls in his model aren't statistically rigorous enough to include or is willfully disregarding scientific accuracy for clicks
|
|
#
?
Sep 26, 2016 05:45
|
|
|
I would love nothing more than the end result of this election making all pollsters and indeed Nate look like garbage idiots. I don't think it'll happen but oh man that would be great.
|
|
#
?
Sep 26, 2016 05:52
|
|
|
|
| # ? Jun 8, 2024 18:25 |
|
|
I think the narrative that Nate is pushing his model to maximize clicks and horserace is underestimating what brand of sperg Nate is.
|
|
#
?
Sep 26, 2016 06:29
|
|