|
good nate: still good https://twitter.com/Nate_Cohn/status/790749244332335108
|
#
?
Oct 25, 2016 04:01
|
|
|
|
| # ? Jun 8, 2024 23:40 |
|
|
I have to say, I really like the article they put together about why their model hasn't counted Trump out yet. Seems kinda reasonable, and it's pretty funny how changing any of their assumptions leads them to look more like the various other pollsters out there. http://fivethirtyeight.com/features/election-update-why-our-model-is-more-bullish-than-others-on-trump/
|
|
#
?
Oct 25, 2016 15:31
|
|
|
In his latest article Nate calls Anthony Weiner "tragicomic," which I appreciate.
|
|
#
?
Oct 29, 2016 17:40
|
|
|
Nate Silver: Thus, as of early Monday evening, our polls-only model gave Hillary Clinton an 85 percent chance of winning the popular vote but just a 75 percent chance of winning the Electoral College. There’s roughly a 10 percent chance of Trump’s winning the White House while losing the popular vote, in other words.
|
|
#
?
Nov 1, 2016 03:24
|
|
|
so
|
|
#
?
Nov 1, 2016 04:03
|
|
|
The more I watch 538 the more I think it is just an awful model. Just way too conservative with some of its numbers. Okay, sure Florida/NC/AZ are 50%, that's fine. But how the gently caress is Virginia 85% a week before the election when it looks like this:  IMO, the best method is just use state polls for that state. Then maybe slightly modify based on national polls, but I'd lean towards just ignoring them too. If we are trying to figure out how Virginians are voting, adding in any data based on how people are voting outside of Virginia is just noise.
|
|
#
?
Nov 2, 2016 16:20
|
|
|
poppingseagull posted:The more I watch 538 the more I think it is just an awful model. Just way too conservative with some of its numbers. Okay, sure Florida/NC/AZ are 50%, that's fine. I suppose time will tell though?
|
|
#
?
Nov 2, 2016 16:42
|
|
|
six days until the election, 538 has Delaware with an 89% chance of going blue. No one else has it below 99%. "Conservative" is one word for this
|
|
#
?
Nov 2, 2016 16:44
|
|
|
Yeah, take a look at Colorado, where it's been almost a month since a poll hit that showed Trump with a lead yet they give him a one in four shot of winning there. I can't shake the feeling that this is mirroring what we saw in 2012, where the national polls showed a tighter race than the state polls, and the state polls turned out to be accurate. Also, I think I ranted about their percentages earlier in this thread, so I won't rehash, but I'm from rural Minnesota, and I follow politics here closely, and the idea that Trump has a 1 in 6 shot here is laughable. There hasn't been a single poll taken at any point that's shown him with a lead, and just one from the end of September that showed a tie (which 538 adjusts to +1 Clinton). Their vote share shows Hillary nearly at 50%, and with a 5.6% advantage. How that, with 6 days to go in the election transaltes to a 1 in 6 chance of Trump winning is beyond me. The same argument could be made on the other side for Georgia with the Democrats. 538 gives Clinton about a 1 in 6 chance of carrying Georgia, yet Trump is over 50% in their projected vote share with a 5.2% advantage. Is it impossible for Trump to win Minnesota or for Clinton to win Georgia? No, but each state would need to move by a percentage point in the polls each day from now until election day for it just to be a tossup. And 538 is saying there's a 1 in 6 chance of that happening. I like the way that they present their data, and I like their pollster adjustments, but their percentages are just pants-on-head stupid.
|
|
#
?
Nov 2, 2016 16:51
|
|
|
Azathoth posted:Yeah, take a look at Colorado, where it's been almost a month since a poll hit that showed Trump with a lead yet they give him a one in four shot of winning there.
|
|
#
?
Nov 2, 2016 16:55
|
|
|
And it is important to note that yes, this happens both ways. As Azathoth said, Georgia is the same way in reverse as Virginia/Colorado/Minnesota. That said, they don't cancel each other out when we look at the overall odds of the election. Having more certainty with the current state alignment directly corresponds to more certainty in a Clinton win.
|
|
#
?
Nov 2, 2016 17:00
|
|
|
The biggest issue this year: the quality of polling has been atrocious. There has been fewer overall polls, and also fewer quality polls. For 538's model, that's going to be disastrous.
|
|
#
?
Nov 2, 2016 17:30
|
|
|
Concerned Citizen posted:The biggest issue this year: the quality of polling has been atrocious. There has been fewer overall polls, and also fewer quality polls. For 538's model, that's going to be disastrous. nate's lovely model cant fail, it can only be failed (by polls)
|
|
#
?
Nov 2, 2016 17:42
|
|
|
imo the lesson to learn from this year's 538 is that glorifying predictions and turning them into pseudo-journalism media sites is not a good idea.
|
|
#
?
Nov 2, 2016 17:43
|
|
|
poppingseagull posted:And it is important to note that yes, this happens both ways. As Azathoth said, Georgia is the same way in reverse as Virginia/Colorado/Minnesota. A lot of 538's assumptions about shifts in state-level polling also seem to be based on their past correspondence to one another demographically, whereas this could be the start of a realigning election that leapfrogs the traditional balance of red/blue states relative to the mean. So even though in the past OH and IA have been more favorable to Democrats than NC, the new partisan split we're seeing between college degree + non-college degree upsets that relationship in a way where it's hard to make a correlative statement between them.
|
|
#
?
Nov 2, 2016 17:44
|
|
|
Our Election Model
|
|
#
?
Nov 2, 2016 17:46
|
|
|
My election model is Charlotte McKinney
|
|
#
?
Nov 2, 2016 17:49
|
|
|
An Open Letter to White Girls Regarding Pumpkin Spice and Cultural Appropriation
|
|
#
?
Nov 2, 2016 17:50
|
|
|
https://twitter.com/ForecasterEnten/status/793892208713854977
|
|
#
?
Nov 2, 2016 20:08
|
|
|
God bless you, Harry Enten. I hope you can still find work after this election. Maybe Sam Wang will take you in. Samuel Clemens has issued a correction as of 23:44 on Nov 2, 2016 |
|
#
?
Nov 2, 2016 21:39
|
|
|
dwarf74 posted:It's because they are looking at the movement of all the deep Red hellholes bordering Colorado. Which, yes, state lines are basically permeable and imaginary. But polls of people within those states are not imaginary. If you have every poll in Colorado showing Clinton winning, people are not going to cross the border from Wyoming to flip the state red. It would make far, far more sense to build a model based on heavily polled swing states and use that to infer what is happening in non-competitive states than to try to infer a swing state shift that isn't happening in the polls from a random poll of a noncompetitive state. The few times when a new state legit gets put in play, like Arizona or Utah, then add them back in as "in play" or something, instead of pretending that a poll in North Dakota has any relevance to a competitive state.
|
|
#
?
Nov 2, 2016 22:02
|
|
|
Concerned Citizen posted:The biggest issue this year: the quality of polling has been atrocious. There has been fewer overall polls, and also fewer quality polls. For 538's model, that's going to be disastrous. 538s whole model is built on the polls being terrible. They're going to have the least egg on their face of anyone if there is a massive polling failure.
|
|
#
?
Nov 2, 2016 22:59
|
|
|
Azathoth posted:It's all just endemic of their problem with garbage-tier polls. For the few times that a poll in a seemingly non-competitive state gives something truly insightful or interesting, like McMuffin in Utah or Trump's surprising weakness in Texas, it's just total noise they're dealing with the rest of the time. so basically Sam Wang's model?
|
|
#
?
Nov 2, 2016 23:01
|
|
|
Bip Roberts posted:538s whole model is built on the polls being terrible. They're going to have the least egg on their face of anyone if there is a massive polling failure. It's hard to have a massive polling failure when the polls are flailing wildly from one place to another. How 538 interprets that is on them.
|
|
#
?
Nov 2, 2016 23:43
|
|
|
Concerned Citizen posted:It's hard to have a massive polling failure when the polls are flailing wildly from one place to another. How 538 interprets that is on them. And yet Wang's model is converging at a very reasonable EV number based on what we've seen so far. I mean if Princeton is wildly off in the end I'll come back in here and eat my crow but the simpler explanation is that Nate's model this time around blows and people still feel a need to white knight it
|
|
#
?
Nov 3, 2016 01:26
|
|
|
Chokes McGee posted:And yet Wang's model is converging at a very reasonable EV number based on what we've seen so far. What is Wang's model saying about Florida and North Carolina? Ohio and Iowa? Seems like there's still a lot of EV to give in either direction, just not enough for Trump to win.
|
|
#
?
Nov 3, 2016 01:36
|
|
|
Vox Nihili posted:What is Wang's model saying about Florida and North Carolina? Ohio and Iowa? Seems like there's still a lot of EV to give in either direction, just not enough for Trump to win. leans dem, leans dem, leans repub, repub (iirc) his 75% strike zone is 320 or so iirc but I'm on my phone and too lazy to check. just Google Sam Wang and click the first link or w/e
|
|
#
?
Nov 3, 2016 01:44
|
|
|
Vox Nihili posted:What is Wang's model saying about Florida and North Carolina? Ohio and Iowa? Seems like there's still a lot of EV to give in either direction, just not enough for Trump to win. Sam Wang has Clinton winning NC and FL, losing OH, and IA is a tossup. Clinton win probability 97%. 320 EV median I think. It's been there since the first debate. It never dipped below 80% even at the worst point in September. The Whole Internet has issued a correction as of 02:09 on Nov 3, 2016 |
|
#
?
Nov 3, 2016 02:05
|
|
|
Can we all at least agree that Farai Chideya is the best writer on 538
|
|
#
?
Nov 3, 2016 04:42
|
|
|
https://twitter.com/NateSilver538/status/730251094614528000
|
|
#
?
Nov 3, 2016 05:58
|
|
|
The Whole Internet posted:so basically Sam Wang's model? Tayter Swift posted:Can we all at least agree that Farai Chideya is the best writer on 538 Agreed.
|
|
#
?
Nov 3, 2016 06:01
|
|
|
Wow, it turns out hey do have standards. http://projects.fivethirtyeight.com/pollster-ratings/
|
|
#
?
Nov 3, 2016 17:55
|
|
|
drat he's good
|
|
#
?
Nov 3, 2016 18:52
|
|
|
The supposed unreactionary polls-plus model has dropped 15% in a week. At this I'm hoping it shows a Trump W on election day just so 538 will truly be shamed.
|
|
#
?
Nov 3, 2016 20:28
|
|
|
poppingseagull posted:The supposed unreactionary polls-plus model has dropped 15% in a week. At this I'm hoping it shows a Trump W on election day just so 538 will truly be shamed. go read the articles about how polls-plus works. no need to report back, either
|
|
#
?
Nov 3, 2016 21:48
|
|
|
is there any way to see what 538 would say if they took the polls at face value and didn't reweight/unskew them? I think it's interesting how adamant people like Sam Wang are about how you shouldn't second guess the pollster's methods.
|
|
#
?
Nov 3, 2016 22:17
|
|
|
 
|
|
#
?
Nov 3, 2016 22:39
|
|
|
we need a custom smilie of a shaking nate silver head imo
|
|
#
?
Nov 3, 2016 22:58
|
|
|
Lutha Mahtin posted:go read the articles about how polls-plus works. no need to report back, either Where do I find said articles? I am only going off what Nate himself said on one of the podcasts I've listened to where he described the models. He said the polls-plus is the least reactionary.
|
|
#
?
Nov 4, 2016 01:14
|
|
|
|
| # ? Jun 8, 2024 23:40 |
|
|
poppingseagull posted:Where do I find said articles? I am only going off what Nate himself said on one of the podcasts I've listened to where he described the models. He said the polls-plus is the least reactionary. http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeights-2016-general-election-forecast/ Presumably what that poster meant is that all of 538's models are, at this point, going to be converging on the behavior of their now-cast and weighting new information much more heavily. If there's significantly divergent behavior between the three models over the next few days, it's a sign there's something wrong with one of them.
|
|
#
?
Nov 4, 2016 01:35
|
|
