2016 Postmortem
Related: About this forumI Give Up on 538, Im Going to the NYT
538 is getting too frustrating to follow. Its not that I think they are up to something, its just their numbers dont seem to make much sense. It seems like no matter what polls come out, clinton falls. She can have a poll come out today showing her up by 18 points but she will fall by 2 percent at 538. Her average lead is 8 points nationally and she is far ahead in every swing state but she had an 82% chance of winning? Bullshit. I dont know if they are too conservative with their numbers or afraid of underestimating trump like they did in their primary commentary but it seems like their model sees it as almost a tie unless clinton is up by 40 points on average.
The NYT odds arent drastically different than 538. They have her 90% compared to 538 at 83ish% but they at least make sense. If a poll comes out showing her ahead by 12 points in florida, they dont drop her odds of winnby by 3 points for seemingly no reason.
http://www.nytimes.com/interactive/2016/upshot/presidential-polls-forecast.html?_r=0
saidsimplesimon
(7,888 posts)what seems like madness? Drive down expectations, complacency, GOTV has never mattered more to me.
geek tragedy
(68,868 posts)and probably permanently--
Tommy_Carcetti
(43,154 posts)I know they probably explain in great detail, but a Reader's Digest explanation?
geek tragedy
(68,868 posts)the economic data points to a 50/50 race, so it will dilute any polling advantage either candidate has
Polls-plus combines polls with an economic index; polls-only does not.
Polls-plus will include a convention bounce adjustment; polls-only will not.
Polls-plus starts by assuming that likely voter polls are better for Republicans; polls-only makes no such assumption. Both models revise this assumption as more data becomes available.
Polls-plus subtracts points from third-party candidates early in the race, while polls-only does not.
Both models employ a regression that is based on demographics and past voting history. But polls-only weights the regression less and places less emphasis on past voting history.
Polls-only accounts for more uncertainty than polls-plus.
Polls-plus and polls-only will tend to converge as the election approaches.
DemocratSinceBirth
(99,708 posts)DanTex
(20,709 posts)They have similar techniques (there really aren't that many ways to average polls), they've made some methodological choices that result in different estimates, but both are teams trying to come up with the best estimate possible.
MFM008
(19,803 posts)with 538 ever since they said the Sea hawks were a sure winner in the Super bowl for the second year in a row .
Maybe they just didn't see Carroll make the stupidest call in football history in the last minute of the game.
yallerdawg
(16,104 posts)Carroll's call was unpredictable.
Godhumor
(6,437 posts)538's model is programmed to be fluid and respond to data along with trends and continuation in those trends. This makes their forecast jumpy.
I do this kind of work for a living, and my issue with 538 is that long term forecasts are supposed to be relatively stable unless a major shift happens. The whole point in a forecast is that you are confident in what you predict and don't need it to move around unless something completely unforeseen happens.
On that basis, PEC is my favorite forecaster, as Sam Wang hangs his hat on his predictions and sticks by them.
Doctor Jack
(3,072 posts)I can understand the importance of trends but the effect that 1 poll can have on their prediction for election day is drastic. I thought the whole point of 538 was to see the long term prospects not to get lost with each individual poll. Im not sure what has happened since 2008 and 2012 but 538 seems to be different this time around.
Godhumor
(6,437 posts)Last edited Fri Aug 26, 2016, 02:24 PM - Edit history (1)
And not being cynical, that is explicitly what they're doing. No one talks about models this far out on the news if there isn't some shocking shift in probability. That is, again, why they added the NowCast, as it is literally not useful for anything except as a publicity tool.
For their more traditional models, by assuming trends continue in the future, they get shifts. They can also get shifts by baking in expectations for candidates than moving the needle only when they underperform (I.e. in NY Clinton should win by 30. Two new polls come out, one with her at +31 and one with her at +25. Fantastic numbers in both cases! But Clinton's win probability may actually go down because the model assumes she should be +30, so the +31 doesn't help her. However, the +25 is a significant "underperform" and may indicate a shift away from her in broader support.). It is pretty dubious, and the academics who do this work without the PR aren't exactly enamored with Nate's methodology right now. He will end up being right, which is all anyone will remember, but he is making a path to getting it right that maximizes exposure.
I like Nate and his team, I quote them often. Just not the best model in the world this cycle.
Imperialism Inc.
(2,495 posts)The polls have been tightening slightly so when he runs the 10k trials of the simulation Clinton wins less often. Yes, he does trend line adjustments and that effects the specific number the models come up with but if the question is about the direction of her chances (they went down) that isn't really relevant. Its pretty simple; her lead went down in the polls so her chances went down in the model. Surely any reasonable model should behave that way, unless her lead is so large that the margins of error in the polls don't matter (and remember his model relies on state level polls first and foremost) .
Maybe I'm missing something though?
MohRokTah
(15,429 posts)Players in sports that start a downward trend will typically continue that trend even when things look up for a while due to an aging populace of players.
This has a tendency to under predict in a political trendline, so when Hillary trends down for a time, they upturn is never handled as well as it actually is performing.
Adrahil
(13,340 posts)to not just follow sites that make you feel better. Besides, 538 STILL shows HRC as the heavy favorite.
Demsrule86
(68,469 posts)However, 538 does different sort of polls that measure different things so you have know what the purpose of the Poll is.
mrJJ
(886 posts)Source: Princeton Election Consortium
As of August 26, 12:03PM EDT:
Snapshot (100 state polls): Clinton 340, Trump 198 EV Meta-margin: Clinton +5.4%
Clinton Nov. win probability: random drift 92%, Bayesian 95%
Senate snapshot (49 polls): Dem+Ind: 51, GOP: 49, Meta-margin: D +2.0%
Link: http://election.princeton.edu/
romana
(765 posts)Silver can't get over the fact he underestimated Trump in the primary, so he's overcompensating for that. It's biasing all his output now.
Thrill
(19,178 posts)How many times does this guy have to be right before you learn this
Doctor Jack
(3,072 posts)He has only done this 2 times before. Nothing wrong with looking at other analysts. He isnt the end all be all of statisticians
Thrill
(19,178 posts).
jamese777
(546 posts)Called 50 states out of 51 states plus D.C. correctly in 2008 and 51 out of 51 states plus D.C. correctly in 2012. His "batting average" then is .990.
That's pretty good for 102 times "at bat."
Until he blows one big time, I'm sticking with 538 Blog.
MohRokTah
(15,429 posts)you tell me.
He blew 2010 and 2014.
doc03
(35,296 posts)jamese777
(546 posts)Show Trump with a slight lead over Clinton: USC/LA Times Daybreak Poll (44.3-43.6) and UPI/CVoter (48-47) Both have Trump up by less than one point. Since these are legitimate polling organizations, they impact the national polling calculations at 538 Blog.
Also a new Mason-Dixon Poll in Florida shows the race tightening slightly there: Clinton 44%/Trump 42%.
Just the normal statistical "regression toward the mean"ups & downs of a long campaign season.
Doctor Jack
(3,072 posts)And really up or flat over the last week, yet 538 has her down over 10%. When I wrote the original thread this morning, I thought it was ridiculous that they had her at 83%. Now she has fallen another 3 points? Why? Because the LA Time tracking poll is moving around at random, as per usual. Again, I call bullshit.
jamese777
(546 posts)shows a drop for Hillary from +12 to +5, so expect more fluctuation.
https://www.yahoo.com/news/clinton-leads-trump-5-points-reuters-ipsos-poll-192022350.html
vadermike
(1,415 posts)Geez So is she ahead or is trump really gaining on her that much I find it hard to Belleive he has gained that much on Reuters but then again who knows Rasmussen has her ahead lol that la times and upi are both wacky But god forbid if this is a trend Trump wins no matter what he does and then we are in serious trouble which means his voters aren't budging and Hillary is actually losing voters which doesn't make sense Anything is possible in this fucked up election What scares me is its October 25th and its like 50/50 and we are just barely ahead God help us if that happens , hopefully not I am worried about voter suppression tactics etc I still think she will win but it might not be a landslide which is fine A win is a win I just hope it's not too close
vadermike
(1,415 posts)There's going to be wild fluctuations and even when Hillary goes down in the polling Trump doesn't really go up notice that ? He's actually below 40, like 36 etc So it could just be fluctuation
bagelsforbreakfast
(1,427 posts)CVOTER gets "C+" and LATIMES "no rating" as far as reliability on 538.
Ipsos "A-" has Clinton +3
Quinnipiac "A-" has Clinton +9 (adjusted; +10 non-adjusted)
For the life of me I can't see how Clinton is only up +3 = they must have called the wrong country (Russia?)
wildeyed
(11,243 posts)It does not change once the cycle starts. You can agree or disagree with the methodology, but you cannot argue with the math just because it is saying something you don't like.
His predictions may or may not be right, but at this point he is just feeding numbers into his system. There is no bias.
http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeights-2016-general-election-forecast/
Also, I am not sure what you are so nervous about. It is still August. Clinton is ahead by huge numbers and the momentum is in her favor. We still need to GOTV our butts off, but I am not losing sleep over whether she has an 83% chance or 90% chance of victory. Both are pretty good odds, and neither changes my role in the process. Did I mention how important it is to GOTV? Seriously. Do some phone banking or register some voters. That always does wonders for my election anxiety issues
ecodeathmarch
(34 posts)I don't work for this site or anything, I just think it's the best, simplest one. been using it since '08. Nbd just FYI
www.electoral-vote.com
megahertz
(126 posts)Currently has Clinton 53.6/Trump 46.4
It's the main poll site I follow. Occasionally it's slightly slow to load (for me, anyway).
For poll geeks, the "how it works" info is in the FAQ.
Past Presidential elections:
2004: The final forecast published on the morning of the election predicted that Bush would receive 51.5% of the popular two-party vote, an error of 0.3 percentage points (result 51.2%).
2008: On Election Eve, it predicted that Obama would receive 53.9% of the popular two-party vote, an error of 0.2 percentage points (result 53.7%).
2012: On Election Eve, it predicted that Obama would receive 51.3% of the popular two-party vote, an error of 0.7 percentage points (result 52%).