I'm not really invested in defending Nate Silver or in the results of the presidential election, I just think statistics are cool.
Being a liberal doesn't make you bad at stats. The degree to which 538's models reflect personal value judgments is almost certainly minimized, so dismissing them out of hand because they come from Nate Silver and use cutesy animal drawings doesn't really "own the libs."
"But 2016" they gave Trump more of a chance than anyone else did, and besides they've since updated their models to avoid those specific problems.
And, I'm sorry, even a really really shitty statistical model is going to be a better election predictor than your opinion on what the "mood on the street" is or whatever.
Of course, Trump's voter suppression campaign is a huge wild card in all of this.
Agreed, 538's models are the best around, they got the 2018 midterms almost dead on.
People seem to ignore the fact that statistical modelling is only going to be as accurate as the statistics they're based on are, and in 2016 polls were all suggesting a sure Hillary victory.
Nates a fucking awful pundit, statistics aren't everything, and 538 should stop trying to do sports predictions, but that doesn't make his models bad.
Also, I respect 538 for leaving all their election predictions up as they were on the day of the election. It's a lot better than the Chapo Hosts who like to retroactively pretend that they've never gotten anything wrong.
Glad to see this here. There were a lot of people on the old sub that seemingly thought if you state that there's over 50% chance Trump loses and then he wins, you got owned and statistics are meaningless and probability is a spook.
https://xkcd.com/2370
My one gripe with 538 is that in 2018 they were defending individual donations as a valuable source of forecasting data for campaigns, but in in 2020 when that turned out to be a boon for Sanders, they made a complete 180 on that idea. I mean, Sanders did get defeated, so they weren't wrong, but that stuck with me.
Otherwise, yeah. Statistics are better than anecdotes.
Everyone should ignore their punditry. Their whole brand is stats and so to the extent that one is going to pay attention to 538, one should only pay attention to their models.
People here often have an understanding of polls and statistics that you would usually find under (and in) Trump's tweets.
538 did quite a good job in 2016.
the polls are probably accurate, but there's probably an unaccounted "election fraud" bump to push things towards trump. To be fair, its not in nate's control, but its definitely worth keeping in mind
Yeah exactly, and that could vary wildly state to state. The other big thing is mail in ballots. A relatively huge percent get thrown out, again varying state to state. Dems are mailing in votes more than anyone. This could mean them losing 5% of their vote; even before them getting fucked with.
state by state predictions
For the record, this is how 538's model works.
Those result from the state by state simulations. They represent odds to win the Electoral College, not the meaningless national popular vote.
And, I’m sorry, even a really really shitty statistical model is going to be a better election predictor than your opinion on what the “mood on the street” is or whatever.
"buT i HaVen'T sEEn aNy BidEN yArD SIgnS!"
Wasn’t he like 50 for 50 in 2012 with Obama? And got most of the senate right too?
His models are better than almost anyone else, but also stats don’t determine everything and it’s similar to the complaints sports people have about statisticians. You have to have a feel for real life as well. Polling is a tool, not the only tool.
Trump very well could still win, but I don’t think people are gonna vote for the president that got covid which is who he has become.
To be fair, his model does include many more factors than just polling.
338 Canada is also doing stats for the American election and they provide similar odds to what I've seen on 538. 338 hasn't been around as long but they've been surprisingly accurate in predicting the last few provincial/federal elections in Canada.
https://youtu.be/R5r157vk7IE
different time
EDIT: Ben Burgis went to high school with Nate Silver lol
Being a liberal doesn’t make you bad at stats.
Okay, but did you see this one? https://fivethirtyeight.com/features/why-donald-trump-isnt-a-real-candidate-in-one-chart/
I swear this is said to death on every Nate Bronze thread. stats and punditry are two different things.
Two points on this article:
- Trump announced his candidacy on June 16, 2015, this was posted on the same day of jun. 16, 2015...I mean apart from Bill Mitchell, did ANYONE think that at this point think he was a serious candidate? He only started steam rolling polls by July 2015. This is literally preaching to the choir clickbait
- Favorability is probably a good campaign marker (that being said I haven't seen any article discussing it in detail) but Trumps ability to find and build a base even with massive unfavourability amongst the general population is probably what made him such a strong outlier contender for that election
This is probably "patient zero" in terms of why everyone should ignore 538's punditry.
Who was saying Trump was a good candidate on the day Trump announced his campaign?
I just don't really understand the point of these models. If you glance at the swing state polls, it's pretty easy to reach the conclusion that Biden will probably win but Trump has a chance. These models, even if they're well-designed, are just a very snazzy and attention-grabbing way of saying the same thing. They produce a percentage instead of saying "probably", but even they downplay that (538's headline prediction is just "Biden is favored to win the election") because they know the precise number is pretty meaningless and that most people aren't very good at interpreting probabilities anyway.
Probabilities are generally only useful when you're dealing with repeated events, since that way you can actually check whether the probabilities are accurate and you can use them to build up predictions about what will happen on a larger scale. Like if I run a hospital and I know that the next patient who comes in has a 30% chance of having covid, there's not much I can do with that information since I need to be prepared either way. But if I know that 1000 people come into my hospital every day and each one has a 30% chance of having covid, I can use that to work out what resources will be needed for covid care, and I can check at the end of the day whether the 30% figure was correct.
Also an annoying thing with the 538 model is that they build in loads and loads of tiny adjustments just to give themselves stuff to talk about, and so that they can answer "yes!" when people pester them with "OK, but does your model take into account this weird thing I just thought of?" Many of these adjustments will have minimal impact on anything, but they make it harder to understand the model, make it more likely that there will be some kind of bug in it, and also I think this gives people unrealistic expectations about what actual useful models of things are like.
If you glance at the swing state polls, it’s pretty easy to reach the conclusion that Biden will probably win but Trump has a chance
Well yeah, it is basically this, but it quantifies it and gives it more statistical legitimacy. Mathematically speaking, if you can work out the probabilities in each state, that immediately leads to the overall probability, but using an actual model lets you put a real number on it. It also lets you take into account things other than polling data in a way that's consistent and reduces the kind of personal bias that plagues punditry.
Biden's going to win because boomers like misery and he is human misery, in ways Trump just isn't. They hate aspiration, dreams, principle, hope and those who fight for anything, that's why they like Biden and despised Bernie. They know their time is running out either way, which is why they desperately simp to never trumpers hoping to get more boomers to blockade their own base from having anyone that isn't a GOP hugger in power.
This is a really bad take. The idea of statistics presented here are akin to my aunt who claims her astrology horoscope was right when anything happens.
This is mostly a critique in regards to the rigor of the statistics and the way they are framed in OP's post. I do appreciate good statistics and value modeling, for it to work you have to free yourself from you fitting your ideas into your regressions.
The idea of statistics presented here are akin to my aunt who claims her astrology horoscope was right when anything happens.
How? Is there a specific issue you have with the rigor of their model?
Yes. When you dive deep you notice wild inconsistencies with the way the statistics are done and tuned. They also generalize over data sets which don't deliver the necessary information and both fail to account for past elections and for systematic reasoning about future elections.
My biggest pet peeve is that often information about spatial heterogeneous aspects gets marginalized and not explanatory homogeneous data is used in its stead. Though what at least 538 did correct is to look at individual states instead of just nation wide trends (which is pretty naive).
My understanding is that their model uses states polling. Their national estimates are not based only on national polls
I've never noticed wild inconsistencies in 538 results but admittedly I've only ever done some loose verifying of their numbers in my head (being math brained - assuming you're the same based on this comment). Gimme some ideas where to look? Im genuinely interested to see.