2016 Postmortem
Related: About this forumWhy Are 538s Numbers Always Much More Favorable to Trump than Other Sites?
Im not saying 538 is right or wrong but I am confused, so maybe someone with more expertise in statistics can explain it. 538 currently has trumps odds of winning at around 24%. Every other site, of which there are many, have him between about 10% and as low as 1% chance of winning. Why does Nate Silver peg Trumps odds as 2.5 to 24 times more likely to win than any other projection site?
geek tragedy
(68,868 posts)"systemic polling bias" or something like that.
that would usually show in turnout models/likely voter screens.
Awsi Dooger
(14,565 posts)Polls can be brutally wrong. I supplemented my income for years via that knowledge and betting on politics, after studying a decade's worth of polls. I developed PAN (Partisan Adjustment Number) that essentially provides a handicap to each state based on which way the polling typically errs.
I used to mention PAN frequently on this site a decade or so ago. New Jersey had a pronounced blue adjustment because polling tended to be overly friendly to Republicans. Alaska and Georgia had huge red PAN adjustment because polling invariably overstated the Democrat by many points.
That still holds up, particularly in Alaska. I gave some examples a few days ago. There are many more. The 2008 Alaska senate polling was dreadful. The final two polls had Begich up by 8 and 22. The RealClearPolitics average was +10.3 for Begich. Uh, not quite. Begich rallied to pull out a victory over troubled Ted Stevens by 1.3 points.
I always have to chuckle when Alaska polling is taken at face value, despite all the evidence.
Nate Silver largely ruined political wagering. I'm very happy that my heyday was prior to his arrival. Win probability in relation to poll margins is now far better understood, by oddsmakers and betting markets, than when I started. Nate takes polls at face value but he does incorporate play in his number since he realizes that far more often than not polling error in a presidential race will be in the same direction from state to state, just like the 2014 senate midterm polling error understated Republicans in the high majority of states.
These days my prized PAN theory is not as viable because far more state polls are being taken. Previously you'd have one firm butcher the model in one cycle and then show up again 2 or 4 years later and sloppily apply the same pathetic model to the same state. Betting odds were based on one or two flawed polls. It was glorious.
triron
(22,020 posts)Thanks for bringing this up.
Sometimes I feel as if Nate has some kind of horse race agenda.. Maybe he likes MSM?
Lefthacker
(264 posts)Because he includes some suspect polling and I believe he makes allowances for occurrences such as the bogus email crap.
lovemydog
(11,833 posts)It has to do with which polls he uses, which he discards. And how he actually rates unpredictably pretty high (as in, polls don't tell you everything). You can tell I'm not a statistician, lol. Maybe someone else can explain it much better.
sunweaver
(75 posts)Scroll down the page: http://projects.fivethirtyeight.com/2016-election-forecast/?ex_cid=rrpromo
and you will see how they come to their probabilities.
"Our win probabilities come from simulating the election 10,000 times, which produces a distribution of possible outcomes for each state. Here are the expected margins of victory. The closer the dot is to the center line, the tighter the race. And the wider the bar, the less certain the model is about the outcome."
getagrip_already
(14,838 posts)He has three layers of analysis - the now cast (if election held tooday), polls only forecast which try's to look forwards taking trends into account, and polls plus.
polls plus is the most conservative, and uses things like the economy, polls from like states, stability, and a lot of other "stuff" he thinks are relavent to predicting things.
Also, he uses some "junk" polls that other aggregators don't use. For example, that la times poll that is very pro republic.
So thats a bums rush explanation.
Dem2
(8,168 posts)For instance (as I assume you know), LA times gets adjusted >4 points to the Democrat.
budkin
(6,716 posts)These are still great numbers for Hillary and still show her on a solid path to victory
sunweaver
(75 posts)I follow 538, and I found these links for you in less than a minute. Read everything he's written about his process - something we are asking people to do about other news.
Here's another page to look at:
A Users Guide To FiveThirtyEights 2016 General Election Forecast
http://fivethirtyeight.com/features/a-users-guide-to-fivethirtyeights-2016-general-election-forecast/
redstateblues
(10,565 posts)So he can point to that if he is wrong.
regnaD kciN
(26,045 posts)I think he puts greater store in the volatility of undecideds, whereas other models assume they won't break much differently than the general results. In a race like this one, with a large undecided vote, that's probably the safest choice.
Also, he has a theory (probably correct) of "correlated errors," where, if a given pollster is off in one state, they're likely to be off in the same direction in other states, instead of being off toward one candidate in one state and the other in another state. That, once again, adds a layer of volatility to any predictions.
Basically, I think of 538's odds as a "worst-case scenario" for whichever candidate is leading.
UTLonghorn
(24 posts)this election season. Like "geek tragedy" mentioned, his model does account for "systemic polling bias" that others don't , but 538 has really grown from its 2008 and 2012 days, and at the end of the day he needs the internet traffic, so he is trying to sell the election as being closer than it is, imo. I have actually found myself following another 538 writer on twitter, Harry Enten, more than Silver lately. Enten seems to be less invested in the horse race narrative and I think has a more realistic perspective on the election.
DLCWIdem
(1,580 posts)They are more conservative.
ProudToBeBlueInRhody
(16,399 posts)Nate is too wrapped up in seeming impartial.
0rganism
(23,970 posts)in theory it can go either way - better chance for a tRump win, better chance for a Hillary blowout
Doctor Jack
(3,072 posts)It does seem like in 2008 and 2012 he was more willing to be bold and he was very accurate. If he is so conservative that his numbers dont mean anything, isnt that also a problem?
Awsi Dooger
(14,565 posts)The major candidate polling in late cycle 2012 was 95% or higher combined for Obama and Romney. This year it's generally in the high 80s.
There is also far greater uncertainty in the third party vote, whether the Johnson and Stein voters will remain loyal or, if not, which way will they break?
If Nate wasn't being cautious this time it would make little sense.
Also, watch for polling to change significantly later this week. Pollsters begin "pushing" late undecideds to declare in the final days. Instead of high 80s combined between Hillary and Trump I suspect we'll have many polls closer to typical level of low to mid 90s between the two major candidates.