This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
Yes, but for a non-repeatable event it’s also very easy for a pollster to say they were right. After all, even someone who predicts a 95% likelihood of A winning can say “well, the 5% likelihood of B winning happened to be the outcome in this scenario, my forecast was in fact entirely correct” and this is completely unfalsifiable.
They were all wrong, but Nate was less wrong ,so that makes him the winner in this regard. His model was more accurate.
No, Nate isn't "less wrong" because 95% chance of winning and 70% chance of winning don't actually have a meaning in this context. How could you even make such a judgement? How do you know that if we had access to 100 different universes with the exact same 2016 race, Hillary doesn't win 95% of them?
It's absurd to claim that Nate Silver's model was more accurate because it gave marginally better odds of a Trump victory; that's not even getting into the fact that absolutely no new polls were available to Nate that showed a tightening race - this was purely Nate fudging the numbers because he knew something was off (something he used to constantly do with his sports prediction spreadsheets he made his bones on)
I disagree. By this logic, no polling can be considered useful or skill does not exist in terms of polling. Obviously one cannot redo the election hundreds of times or split off many universes.
Nate wasn't doing polling, he was placing odds on election outcomes. Why you think "that logic" has anything do with the potential accuracy of any given poll is beyond me.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I respect the thrust of this argument in general, but Nate Silver specifically came the closest to predicting Trump's victory out of the major pollsters. Most pollsters just look at the headline probabilities but fail to properly take conditional probabilities into account. They looked at poll after poll and did the fairly standard "average everything, find the STDEV, there's your confidence interval, 99% clinton victory." What made Nate Silver special is that his model accurately identified the sorts of universes in which trump was likely to win by finding out the ways in which various poll results and errors were correlated. That allowed him to more accurately assess the possibility of a systematic underpolling based off of purely statistical guesswork-- he didn't need to understand why the polls could have been (and ultimately were) biased for clinton, he just had to set up his model so it spit out that possibility on its own.
Based on the reality we live in, it's probably true that even Nate's estimate was wrong-- that it wasn't rolling a 5+ on the six-sided election day dice that gave trump the win, but that underlying factors put trump's win probability somewhere north of 50%. Given the data available though, Nate was the most effective poll-aggregator available.
Prediction markets are like a super-nate. Aside from each individual user having having access to all the same tools nate did (and the retrospective + incentive to use them), every vote on every market is a sort of poll, and all the people playing arbitrage force the markets to take into account conditional probabilities. They're still not going to be "right," all the time-- they're not even necessarily going to outperform your average pundit. But as a casual observer without inside knowledge, following the markets is a dominant strategy over basing your worldview on any particular pundit or basket of pundits. It's like the "always buy SPY" investment advice. Rare people, in rare cases, can consistently outperform the betting markets. But without some very convincing reasons, you shouldn't assume you're one of them.
I don’t disagree with you, my objection is really primarily aesthetic and partially on principle. A fund manager makes a bet, he doesn’t just decide that there’s a 70% chance of Boeing stock doubling in the next year, he gambles that it will. This counts as ‘making a decision’.
Nate is the actually worse than the bank research analyst who defends his buy rating on a stock that took a huge dive by saying that all he actually meant was that it would most likely do well, and then shrugs, because at least the latter has a position.
Nate should be confident enough, at least, to have a headline prediction that says “X is probably going to win”.
More options
Context Copy link
More options
Context Copy link
Sure, but this doesn't make sense in the context of prediction markets. Prediction markets host hundreds of predictions. We can look at the history of those predictions and see how well calibrated they are.
I don't believe the claim that prediction markets are "not accurate" would bear scrutiny.
More options
Context Copy link
Is probability even well-defined for a one-off event? It's not like we can random sample the multiverse on how the election actually went. At the same time, nothing is absolutely certain (supervolcano as October surprise!).
Maybe it makes sense from a Bayesian perspective: given the current knowledge of the system state (polls, voter registrations, demographics, maybe even volcanology reports) we can estimate the probability of a specific outcome. But a frequentist view seems nonsensical, even if a lot of predictions seem to present themselves that way.
One-off events are intractable. Kelly does not work on them.
More options
Context Copy link
I completely agree, the frequentist view is nonsensical. This is why forecasters need to be nailed down to a specific outcome (or ‘I don’t know / it’s too close to call’ but this has to be acknowledged as opting-out).
That's my main problem with Nate Silver's modelling.
There should be large error bars around the prediction that slowly close in as the predicted event approaches.
It shouldn't be "X% Trump, Y% Kamala," it should be "X% Trump, Y% Kamala, Z% irreducible uncertainty."
The logic is "if the election were held today then here's the probability." But... the elections won't be held today. That's the whole point of the prediction for a future event, and I think it behooves them to acknowledge that uncertainty is inherent to the modelling process.
If they'd included that back when it was Trump vs. Biden, the conserved probability would have accounted for Biden suddenly dropping out and wouldn't have broken the model instantly. Also helps reflect the chance that one of the candidates dies... which also almost happened.
And if Nate trusts his model, there's a ton of money to be made in the prediction markets.
What would this irreducible uncertainty mean for an event with a binary outcome? I think Silver already accounts for increasing uncertainty as he propagates his current prediction into the future (what he calls forecast vs. nowcast).
Error bars would make sense around the expected vote percentage. Of course the probability distribution over vote percentages becomes broader as you look into the future, and perhaps he does show that to paying customers. But in the end you still have to integrate over that when the layman asks for the probabilities of who wins the election. And that still amounts to two numbers that sum to 100%.
Evaluating a predictor's performance seems straightforward to me via the usual log-likelihood score. Record the final outcome and take the log of the predictor's probability for that outcome. That score can then be summed over multiple different elections, if you like. (Not sure though if I'd call that scoring rule particularly frequentist.)
But the outcome ISN'T really binary, is it?
Biden dropped out, Trump could have been killed by that bullet, and then we'd have a whole new ball game. The "Trump vs. Biden" model almost certainly didn't include a variable for "the Candidate abruptly drops out" and I doubt assassination risk was plugged in either.
And the fact that it tries to 'call' an election months out but has to adjust radically to new info is why I call it 'gimmicky.'
Taleb had his own discussion of this a while back, and this is the best summary of it I've found.
https://towardsdatascience.com/why-you-should-care-about-the-nate-silver-vs-nassim-taleb-twitter-war-a581dce1f5fc
Sure, in this trivial sense it is not.
I've heard this same argument from other Taleb-readers and got the impression that "irreducible uncertainty" means something entirely different than just a "none of the above" outcome.
If it is just that, then the same arguments apply: When you make the final probabilistic prediction, you integrate over the uncertainty, resulting in three numbers that add up to 100%. After recording the actual outcome, take the log of your predicted probability for this outcome, and that's your performance score.
Of course, if Silver rounds down the probability of "none of the above" to 0% for convenience and it still occurs, he'd technically incur a score of -Inf, which should tank his credibility forever. But I find that a boring technicality.
From your linked article:
Taleb is wrong here. Under this standard, no reasonable predictor could survive any real world application, as it would have to be trashed after the first mistake. And those that do survive would be hopelessly overfitted to past data.
The article got it precisely the wrong way round. Classical models such as logistic regression are trained and evaluated using their full probabilistic predictions.
It is only thresholded to a deterministic choice when used as input to a human decision, where the prediction is appropriately weighted for costs of false positives vs false negatives etc. (which you cannot do if it were a deterministic prediction in the first place).
How's that bad? I'd call that perfectly rational behaviour.
Perfectly rational behavior would probably be saying "I don't think I can accurately predict outcomes this far in advance."
So they add in the caveat "if the election were held today here is what the model says about the odds."
But the election isn't being held today. They know that, the audience also knows that but will still read the model.
Without a gimmick they have nothing to sell.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
There's not a ton of money to be made if you believe the odds are 50/50. Prediction markets give Trump 60/40 odds, while Nate's model gives 50/50 odds. If your bankroll is $1M, then it's only rational to bet 167k, for an expected value of 40k. Not nothing, but not a ton of money either.
That also ignores other costs, like counterparty risk. Nate also has to deal with reputational risk: people might value his published models less if they thought he was making bets on markets that were influenced by his models. Since that's his main source of actual income, a bet would be substantially negative EV for him.
There's reputational risk for having his model diverge too far from the prediction market's call, if the markets end up looking more accurate.
And I've seen him offer various bets before.
I like Nate generally, but I end up with the feeling that the Presidential Election model is a bit too gimmicky for my tastes. As stated, he should display some factor that accounts for the inherent uncertainty of a long-term prediction, rather than making confident-seeming prognostications which get aggressively revised as new information comes in.
He's not calling his shot well in advance, he's just adjusting to the same information everyone else gets as it comes in. Credit for the model being reasonable, but what new information is it giving us?
There is no new information. We have no reason to think his model is even particularly good. The only thing that it really brings to the table is that he both has a consistent reasonable process that doesn't give much room for human bias to creep in, and that he doesn't abandon it the second it makes partisans mad. This is, apparently, a really hard thing to do.
Or, a bit more optimistically: it actually does provide information that can be new to people. E.g., post-debate the model showed that Biden was going to get walloped because voters just weren't willing to re-elect someone senile, while many Democrats were doing their best to convince themselves otherwise. It's somewhat fair to say, "oh, so all the model does is tell you things that are blindingly obvious?" Yes, with a significant caveat: seeing what's in front of your eyes is a very difficult thing in politics.
"All models are wrong. Some are useful."
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
You have to look at their predictions in aggregate. If they predict 20 elections with a 95% chance for party A, and A wins 19 of those 20 elections, then yes they were accurate.
Even if that 1 election was a landslide for party B, the prediction method is accurate. People who say otherwise just aren’t accepting that it’s a percentage chance and not a poll.
More options
Context Copy link
More options
Context Copy link