|
Polling is only going to get shittier every cycle. By 2024, polls will all be "poo poo, I don't know."
|
# ? Nov 5, 2016 18:12 |
|
|
# ? Jun 9, 2024 00:26 |
|
exquisite tea posted:If early voting trends are not matching the polls, then the problem has more to do with the initial assumptions you made in your model, which can be better controlled for than creating some kind of arbitrary fudge factor for early voting (which we cannot know for certain until Tuesday). Like for example, not including sketchy landline only polls with a high MoE. For models constructed solely off of polls, it's GIGO. But it may not be just because of garbage landline-only polls- CNN/ORC recently found Trump up in NV, and they're not exactly Gravis or ARG. The bigger problem with the polls, right now, is that many voters (Hispanic, low propensity) are not making it through LV screens or being contacted at all due to polls being largely conducted in English. And these are voters which we're being told by the Clinton campaign are turning out in massive numbers in FL. On some level, it comes down to the fact that Clinton's campaign is turning out voters who would be missed by LV screens early, and it's not Nate's job to determine how good or bad a pollster's LV screen is. This is, however, shown in the EV numbers we're seeing. There's *some* predictive power there- NV-based journalists have read the tea leaves correctly several times in recent memory, beating out the polls. Maybe Nate can add EV numbers in some way to Polls-Plus, because that model is full of subjective crap that doesn't improve the model. a patagonian cavy has issued a correction as of 18:21 on Nov 5, 2016 |
# ? Nov 5, 2016 18:17 |
|
Patter Song posted:Polling is only going to get shittier every cycle. By 2024, polls will all be "poo poo, I don't know." p sure we're there right now
|
# ? Nov 5, 2016 18:19 |
|
joepinetree posted:Actually, Brier Scores are used precisely for that reason: Thanks for the link, it was a really interesting read.
|
# ? Nov 5, 2016 18:37 |
|
Lol wtf is this supposed to mean
|
# ? Nov 5, 2016 19:29 |
|
poppingseagull posted:Thanks for the link, it was a really interesting read. Now, keep in mind that those were only for election day predictions. If they tried to find who was right, earlier, Drew Linzer correctly picked all states in July: http://votamatic.org/2012-forecast-tracker/
|
# ? Nov 5, 2016 19:32 |
|
drat, Nate's really losing his poo poo: https://medium.com/@and0/election-update-how-our-model-works-8acb97202277#.vq6xkud27
|
# ? Nov 5, 2016 20:10 |
|
someone (who's not lazy like me) should make a chart showing all of the times every site changed its predictions and how often they were on the correct side
|
# ? Nov 5, 2016 20:14 |
|
androo posted:drat, Nate's really losing his poo poo: https://medium.com/@and0/election-update-how-our-model-works-8acb97202277#.vq6xkud27 holy poo poo this is amazing
|
# ? Nov 5, 2016 20:15 |
|
androo posted:drat, Nate's really losing his poo poo: https://medium.com/@and0/election-update-how-our-model-works-8acb97202277#.vq6xkud27 lmao
|
# ? Nov 5, 2016 20:17 |
|
Mukaikubo posted:i agree but for different reasons As it happened, you agreed for exactly the same reason as I did.
|
# ? Nov 5, 2016 20:23 |
|
Patter Song posted:Polling is only going to get shittier every cycle. By 2024, polls will all be "poo poo, I don't know." One of the recent 538 podcasts touched on that a bit, I think. What with more polling companies moving to in-house stuff on account that funding from universities and newspapers are going down. Not sure if I'm remembering that right.
|
# ? Nov 5, 2016 20:45 |
|
https://twitter.com/NateSilver538/status/794972961019232256
|
# ? Nov 5, 2016 20:48 |
|
Admitting that your model isn't a very good predictor of reality doesn't strike me as a particularly smart business decision.
|
# ? Nov 5, 2016 20:54 |
|
Samuel Clemens posted:Admitting that your model isn't a very good predictor of reality doesn't strike me as a particularly smart business decision. i think you'll find that by "election forecast" he meant "summary of what polls say"
|
# ? Nov 5, 2016 20:56 |
|
https://twitter.com/NateSilver538/status/794994593574113282 maybe you might think that huffpo article is wrong but i'm not sure how criticizing nate silver is inherently irresponsible edit: he's so mad he's makin typos https://twitter.com/NateSilver538/status/794996711315951616
|
# ? Nov 5, 2016 21:15 |
|
https://twitter.com/NateSilver538/status/794994593574113282 lmao edit: https://twitter.com/NateSilver538/status/794996711315951616 double edit: fuuuuckk you ^
|
# ? Nov 5, 2016 21:16 |
|
androo posted:drat, Nate's really losing his poo poo: https://medium.com/@and0/election-update-how-our-model-works-8acb97202277#.vq6xkud27 God drat those graphics are brilliant.
|
# ? Nov 5, 2016 21:19 |
|
tankadillo posted:https://twitter.com/NateSilver538/status/794994593574113282 he's so mad he's barley making sense anymore https://twitter.com/NateSilver538/status/794997715604213761
|
# ? Nov 5, 2016 21:23 |
|
nate silver's going to have a stroke on tuesday and trump's going to teabag his corpse
|
# ? Nov 5, 2016 21:23 |
|
Sebadoh Gigante posted:God drat those graphics are brilliant. Haha I thought that snappy tweet was about it and my heart almost skipped a beat.
|
# ? Nov 5, 2016 21:25 |
|
Concerned Citizen posted:https://twitter.com/NateSilver538/status/794994593574113282
|
# ? Nov 5, 2016 21:26 |
|
This seems like the place to ask people's opinions on this: Given 538's model is far more in line with the betting market's predictions for the probabilities of a Clinton/Trump win, and people here seem to think that Nate's model is overly conservative on Clinton, what's driving the underestimation of the Clinton winning probability in the betting markets? Motivated reasoning?
|
# ? Nov 5, 2016 21:28 |
|
CottonWolf posted:This seems like the place to ask people's opinions on this: pants making GBS threads
|
# ? Nov 5, 2016 21:29 |
|
Concerned Citizen posted:he's so mad he's barley making sense anymore Replace 'liberals' with 'polls', and this is what I imagine Nate's current state of mind to be.
|
# ? Nov 5, 2016 21:30 |
|
CottonWolf posted:This seems like the place to ask people's opinions on this: I assume there's a natural aversion to putting down money for little gain. I mean if you put down a dollar on the stock market and make a dollar five back in a month that's great but with betting entertainment I assume once the return gets below 10% there is a natural attraction to the minority odds just because of the potential for large gains on small money.
|
# ? Nov 5, 2016 21:39 |
|
|
# ? Nov 5, 2016 21:43 |
|
Bip Roberts posted:I assume there's a natural aversion to putting down money for little gain. I mean if you put down a dollar on the stock market and make a dollar five back in a month that's great but with betting entertainment I assume once the return gets below 10% there is a natural attraction to the minority odds just because of the potential for large gains on small money. That makes sense. Now I think about it I can buy some sort of sigmoidal curve there, especially with loss aversion. If I have to put down my entire life savings to make £100 on basically a sure thing bet, aversion to the infinitesimal risk of losing my life savings is far larger than to the attraction of an almost certain £100 gain.
|
# ? Nov 5, 2016 21:44 |
|
CottonWolf posted:This seems like the place to ask people's opinions on this: Is it? predictwise is giving Clinton an 86% chance of victory. Much less than Yang's but similar to the New York Times. Also I seem to remember markets have trouble accurately pricing highly improbable events, and often severely misprice them.
|
# ? Nov 5, 2016 21:44 |
|
CottonWolf posted:Given 538's model is far more in line with the betting market's predictions for the probabilities of a Clinton/Trump win, and people here seem to think that Nate's model is overly conservative on Clinton, what's driving the underestimation of the Clinton winning probability in the betting markets? Motivated reasoning? The odds on prediction markets are determined by the individual participants, who are by and large not rational actors. The horse-race narrative pushed in the media has a large impact on how people feel about this election, which in turn is reflected in their willingness to bet. Add in the fact that some people bet on a candidate because they want them to win rather than because they expect them to and that humans generally overestimate small probabilities while overestimating large ones, and it's easy to see why the markets are flawed. It's the whole reason smart investors can actually make money off them.
|
# ? Nov 5, 2016 21:47 |
|
Squalid posted:Is it? predictwise is giving Clinton an 86% chance of victory. Much less than Yang's but similar to the New York Times. Also I seem to remember markets have trouble accurately pricing highly improbable events, and often severely misprice them. The averaged betfair/predictit odds are sitting at around 77% at the moment. Here. e: Actually just the betfair odds by the looks of things, I thought it used both. CottonWolf has issued a correction as of 21:52 on Nov 5, 2016 |
# ? Nov 5, 2016 21:48 |
|
Schwarzwald posted:As it happened, you agreed for exactly the same reason as I did. poo poo, sorry, misinterpreted what you wrote then! Also that HuffPo article is hot garbage and whoever wrote it should probably be smacked with the "Actually, You Are Kind Of A Stupid Blogger" nerf bat, but it's hardly irresponsible and it's hilarious how much Nate's loving melted down over it. And this is coming from someone who instinctively errs on the side of higher uncertainty in estimates. 99% is just flat out bullshit- it's basically saying that systematic polling errors like have happened a number of times in recent history are impossible because somehow sample size can correct for systemic error. It's not right. I'm pretty sure Nate's overestimating uncertainty and how correlated errors in state-by-state results are going to be, but I would much prefer people err on the side of wide error bars than narrow ones. This is p. much the same model that has worked pretty okay the last few national elections. It didn't magically start being bullshit. It just so happened that this election cycle presented a list of conditions that gives his model heartburn and causes it to throw out wide error bars. It's not Nate "unskewing polls"- that's juvenile Huffpo idiocy that's an actual example of the clickbait people accuse 538 of. Nate'll probably go back to the drawing board and change his model once his blood pressure drops back into the triple digits in December. I dunno! Nate is a bit "funny" but I like him still. He isn't the be all and end all of forecasting, but he never was. He's just a dude who knows some about stats who is good at communicating them to the public. That's not a skill set to be thrown out. edit: also quote:@NateSilver538 15m15 minutes ago This is why I get kind of baffled by the people accusing Nate of being super arrogant and insisting he's always right. He says this kind of poo poo all the time...? Mukaikubo has issued a correction as of 21:57 on Nov 5, 2016 |
# ? Nov 5, 2016 21:55 |
|
takes a special kind of perspective to think "i've been wrong before" is an actual display of humility
|
# ? Nov 5, 2016 22:08 |
|
Nate is having an... interesting meltdown on Twitter right now to say the least
|
# ? Nov 5, 2016 22:08 |
|
CottonWolf posted:This seems like the place to ask people's opinions on this: The 538 probability likely has an effect on betting markets, too.
|
# ? Nov 5, 2016 22:39 |
James Garfield posted:The 538 probability likely has an effect on betting markets, too. I put $50 on the election on PredictIt based on the assumption that this was undervaluing Dem chances.
|
|
# ? Nov 6, 2016 00:15 |
|
https://twitter.com/NateSilver538/status/795041766877646848 gonna be real interesting to see what nate does once 538 gets the plug pulled on it sometime between wednesday and the end of the year e - gently caress it i dunno why its not showing up C. Everett Koop has issued a correction as of 00:22 on Nov 6, 2016 |
# ? Nov 6, 2016 00:20 |
|
CottonWolf posted:This seems like the place to ask people's opinions on this: -shitloads of bad polls -race tightening even in good polls -loads of dumb money flowing in at the last minute
|
# ? Nov 6, 2016 00:21 |
|
Samuel Clemens posted:Add in the fact that some people bet on a candidate because they want them to win rather than because they expect them to and that humans generally overestimate small probabilities while overestimating large ones, and it's easy to see why the markets are flawed. Wouldn't people betting on who they want to win rather than who they expect to win make them more in line with polls, rather than otherwise? Unless betters are inherently more pro Trump than nonbetters.
|
# ? Nov 6, 2016 00:27 |
|
|
# ? Jun 9, 2024 00:26 |
|
"@ryangrim Fight me. I'll be in the parking lot in front of Denny's in an hour, be there or be square."
|
# ? Nov 6, 2016 00:30 |