Have yall never played poker/thrown a dice/anything random? When he says libs can relax its because the ODDS of winning are in his favor, but they are still fucking odds. Trump can still roll a 6 and win. If trump wins a) if he predicted that biden would have won, you'd get angry. b) if he predicted trump would have won, thats a shitty prediction that has little basis in the data and even if he was right nobody would listen to a guy who guessed right by chance!!
Think that i win if a coin lands thrice on heads. It's a 12.5% chance i win. Would you bet for me? No. Would you be surprised if i win? Also no, i still had a chance.
The chances lie in the fact that many ppl will vote on a whim based on how they feel one particular day, and you cant know all the data or how reliable it is. He isnt covering his ass, he is acknowledging that he cannot know with utmost precision. Its not a political/emotional thing, its how math works.
Punditry is trying to explain what the last minute deciders are going to do, which is fundamentally about predicting the thoughts and behaviors absolute psychopaths (undecided voters), and is therefore pointless.
you cant predict what a single undecided is gonna do, but a swarm of them are easily predictable.
If that were true all of 538's predictions would say 100% chance, not 60% chance. My post was agreeing with your premise btw. I've been downbeard on chapo dot chat for defending Nate's statistics (while mocking his punditry).
yes, the reason it's not 100% is cos you only know about a group of psychopaths, and only days before they cast their ballots. the way the rest of the mass is positioned and shifts is pretty chaotic and difficult to predict, the only thing not difficult to predict are the chances.
I feel like we're talking past each other. It's fine. I agree that odds exist and getting three heads in a row doesn't prove that it wasn't a 12.5% chance to happen. Correctly observing that an unlikely event is still possible isn't proof that you're full of shit.
ok i dont knwo if/what we are arguing about, im just giving more statistics facts, hahahah
Its very hard to communicate odds with people, because for how much hate Nate Silver got in 2016 the outcome was still comfortably within the polling margin of error so you could say that he was "right" in some way
Even if Trump only had a 1% chance of winning in 2016, that 1% could still very well happen. Things that have a low probability of happening happen all the time.
again, that's a gamblers fallacy, jsut cos something improbable happened, it doesnt mean other improbable things wont happen. you dont "recharge your luck" basically
yeah the other day I was playing fetch with a young family member and the number of times the ball hit something and bounced right back to me was alarming. In one throw, the ball looped around the bowl of a bird bath, bounced off the top of a grill and hit a corner of a shitty fire pit that sent it over a fence and back at my feet.
but things like that happen all the time
People kind of understand probabilities by themselves, but if you let the probability get anywhere near any other percentage, most peoples' brains segfault and fall over. So while people understand a 75% chance of winning a double coin flip, and understand that getting 75% of the votes would be a landslide victory, something funny happens and they can't tell which is which when you tell them there's a 75% chance someone will win an election.
It’s almost as if you can’t actually magically divine the outcome of these things before they happen.
Which Nate Silver knows, which is why he'll never release a prediction for something consequential to be >90% or <10%.
he was. that's how probabilities work. he could have studied the situation better but there is a ceiling to how well you can predict chaotic things.
yeah the only factor that was noticeably butchered or ignored in predicting the 2016 outcome was level of education, which turned out to be a major indicator
but that 1 thing doesn't make Nate Silver a loser
True, what makes him a loser is being a neoliberal pundit pretending to be unbiased
Compound odds are what really throws people off. Most people can conceptualize the odds of something happening once. When you have to predict 15-20 different outcomes (basically the number of state races that are anywhere close to being competitive), which are all conditional on each other to a varying extent (correlated demographics), AND you have to apply some sort of "secret sauce" tilt factor to all of your data (because polls are all guesses/models themselves and you have to make a judgement on their accuracy), it gets much more complicated.
The analogy I use is Russian Roulette, 1 in 6, but still, you ready to put your life on it?
This analogy doesn't work because the Biden chambers are also loaded with bullets.
At least if you get shot in the head, you don't have to experience full fascism. Shot in the leg? Here's your
lanyardmedical bill. (wear it proud and pay it off as we declare you can no longer discharge your medical debt)
It me, the big brain frequentist who's talking about relative frequency probabilities for a single random trial.
I came here to argue with you about the election being a random event so talking about probabilities without breaking it down into what the events/info Nate's uncertain about are and their probabilities/effects are being pure ass covering but I guess if there really is a significant proportion of "whim" voters who could vote either way and might change their mind in the line and after casting that covers everything.
Not really convinced these whim voters do exist, I remember reading somewhere that "undecided voters" were a myth/greatly exaggerated and elections are swung mainly on turnout, but either way good post, thankyou for sharing.
it's not that the election is really that random, most things are deterministic. if i knew about every single person in the country i would be able to predict almost any election, especially now that we are missing only one day till then.
i am not making a statement in politics, but in statistical science and mathemathics.
we do not know all people in the country, and therefore every poll can only rappresent the group of people polled. there are both factors of uncertainty in the people polled (is the population we've asked rappresentative of the whole or is it skewed cos we used skewed methods, such as only phoning landlines; or, have i interviewed enough people? if i ask ten people there coudl be a chance that be coincidence all 10 are trump supporters, if i ask a million it's less likely by the law of large numbers) and in how to interpret the results (the people who say they prefere biden may be less likely to vote/be able to go to a poll). these are the real unknowns that condition the odds of an election, polls come later.
the statement "trump is polling at 55% in, say, texas" doesnt mean that trump will win texas in 55 universes out of 100. it says nothing about that fact. instead, i could be sure that there is only a one in a thousand chance that his polling will be outside 54 and 56% which will make sure he will win, or i could be uncertain and say only in 3 cases out of 10 will trump be outside 50 and 60% of the votes, which leaves a real margin for biden to win.
for example, look at this https://i.imgur.com/g3TEuw8.png
the area under the curve is the chance that that particular result happens. it peaks at 55 for dems and 45 for rep, but all the red combinations are the very possible chances that a rep will win. less likely yes but still possible. in this case we have low certainty
https://i.imgur.com/6Spzlcx.png here instead, the election result is nearer to 50-50, but it’s not as close as the other one, since the blue area is way smaller. this is a situation where we have high certainty in our polling.
Strong agree on uncertainty existing in the analysis Nate's making or even that it's possible for Nate to make, and acknowledging that uncertainty being a good thing.
What I'm trying to say is that putting numbers to that uncertainty without even talking about what the sources of that uncertainty are, how you're putting numbers to those sources, and how you're compiling those numbers into a final percentage, is pure arse covering.
yes, but in the end he runs a business. he probably worries about uncertainties way more than pure result numbers, but the betting market and the news want those and not the mathematically important data.
So, I did a little reading on how 538's numbers work. Apparently their polling aggregation and weighting is actually done to try and make the polls better predict the election? They're done to make the polls better predict future polls? and the uncertainty comes from how "swingy" polls are, or how much they change between polling.
Seems whack.
Have they showed their working on how they came up with their election win chances?
this is still all in the norm for a statistician. if you just have the numbers of the polls, that is often not enough to be certain of what the result will be. for example, you could notice that in 2016, candidates in the primaries who wanted m4a got 10% more voters in a caucus than expected (not a real fact, i made it up). therefore, you could input that observation by counting m4a proposers 10% more in caucuses in 2020. the why doesnt need to be important, and the method doesnt need to be rigorous, but if you repeat this process for hundreds of observations for decades you reach a point where your predictions have to be getting better, cos you are trying to figure out the underlying working of society in a way or the other. it's called making your model better lol
"swinginess" is what i talked about i think? (you'll have to explain yourself better because swinging just means how much they change in my mind, lol.) there is nothing out of the norm here either. he is a normal statistician and if his methods were scientifically flawed, there would be tons of papers about it. but tbh, i think they cant show their methods, otherwise people would just steal those. hm...
Their "swinginess" was calculated state by state and based on how much the polling values changed between each polling, so if a state was polled in January to be 40 Biden 60 Trump, in February to be 50 Biden 50 Trump and in March to be 40 Biden 60 Trump again they'd put it down as having an average swing of 10% and mark swinginess accordingly.
I couldn't find any info on how (if at all) they accounted for length of time between polls, two different polls with alternating biases, etc.
I'm not arguing against the discipline of statistics or anything here. I'm saying that I'm suspicious Nate is doing really bad stats that don't say what he says/implies they do.
Most statistics, especially public facing statistics about/around electoral politics is about torturing data until it says what you need it to, and there aren't enough academic statisticians in the world to write papers refuting it and I've never met anyone who bothers unless there's a specific narrative they need to push back against, it's like fact checking politicians or debating with a gish galloper. I haven't yet seen anything to convince me Nate's doing anything revolutionary.
idk i dont want to be a fangirl, but the guy is doing everything correctly. any secrecy he has is justified since it's his secret formula he wont share. everything seems in order. that swinging is part of the imprecision, if it changes from month to month, it changes day to day, by unpredictable amounts. you can only know by more or less what order of magnitude the swing could be, which is our imprecision i was talking about. a lot of people could change their mind in either diretion, but that's unlikely, which is why the peak remains where it is.
538 isnt particularly revolutionary, it just has a good record and reputation. plus, believe me but criticizing 538's record is really low hanging fruit for anyone wanting to make a paper
People literally only understand 3 percentages in the context of probability: 100%, 50% and 0% if you say something has a greater than 50 -60% of happening people cannot fathom a world in which that event does not occur.
that's litterally the job of an statistician, idk what you are on about. they provide coverage of how the situation expands, they do polls, tell people with electorialism anxiety brain-rot how the things are looking, and are payed by both news and betting markets. in a capitalist world, his job makes sense. what do you expect him to say, "no, i am 100% correct everytime"?
honestly why should you look at polls if you want one side to win and all you'll do is voting
The audience of this information isn't the average person though. People who work on these campaigns find the information useful
Imagine if we would answer any question at work just like Nate...
This might be where the disconnect comes from. STEM nerds, when reporting data like this, are allowed to openly state that they are uncertain to their bosses where uncertainty actually exists. People in more canonically proletarian jobs are forced to pull solutions out of their asses and tell their bosses what they want to hear or get fired, no matter how challenging the task.
The thing is his odds are wrong. Trump is most likely going to win.
It might take the Supreme Court handing it to him, but he will probably win.
oh yeah, that for sure, but libs are still worried about what the elctorial college will look like
The makeup of the electoral college will be what the Supreme Court decides on. Trump will most likely end up with 270 in one way or another.
ok, how do i tell you this dude, this is not what he is predicting? you cant put numbers on those things. stop trying to look smart like that, it's like considering "what if i get up and punch you" in a game of chess. i know you are right but this is not what any of us is talking about
He's predicting who will win what states based on the official vote count in each. The official vote count in some state will likely be in dispute leaving the SC to make a decision that ultimately determines it. We will never know the actual vote count, so it doesn't matter. In your own words, "you can't put numbers on those things".
If he were around to predict Gore winning in 2000, would you say he got it right only for it to be overruled by the Supreme Court? No, because we don't know who won Florida.
We will never know the actual vote count
mh?
also yes i would say he got it right. it's not up to him to know what will happen after people cast their votes. his job ends there. he doesnt make any prediction on the matter of policy, or eventual behaviour of a singular politician. you ask him "how will the people vote", he gives you an answer, then if tomfoolery happens, it's not his job to know.
mh?
Not sure your confusion here, do you know the exact number of people who went into the voting booth in Florida in 2000 with the intent to vote for Gore? No one does.
oh ok that. i suppose it's more of an exception than a rule.
But that's exactly what's going to happen. There will be some dispute about a few thousand votes in Pennsylvania or some shit, a decision will be made to settle that dispute, and Trump will win it. We'll never know how many votes each person actually got. A prediction of some "actual vote count" is completely useless and unverifiable.
Then his odds aren't wrong, that isn't what the model is predicting votes. It's predicting who gets more votes in actuality. Predicting what the recorded votes will be is a separate prediction and difficult because it's hard to know exactly what will happen
It’s predicting who gets more votes in actuality.
Then it would be predicting something that's literally impossible to verify. There is no "in actuality" without knowing the intents and situations of every single eligible voter in the state.
It's a completely useless prediction to make.
Who got more votes "in actuality" in Florida in 2000? Does "in actuality" take into account the people who accidentally voted for Pat Buchanan? How many people actually did that?
There is no “in actuality” without knowing the intents and situations of every single eligible voter in the state.
Polling is an attempt to figure this out. Polls have a margin of error but are generally accurate.
Who got more votes “in actuality” in Florida in 2000?
If votes aren't counted properly it's impossible to know. All we can do is estimate the actual count, and know the official count that gets reported.
The 538 probability clearly doesn't factor in the SCOTUS tilting the scales in Trump's favor. A much more useful statistic might be an estimated probability of Biden winning in such an enormous landslide that the SCOTUS couldn't possibly ratfuck the election. Such a probability could give us an approximation of the complement of the true probability of a Trump victory.
Nah I don't agree with the 10% in his model, but in no way is Trump favored. He's likely going to lose by a decent margin too. I think people here underestimate how much Trump is hated by most people.
i wish i was as blessed by ignorance as you, go forth starchild <3
