Nate Silver: "**** you, we did a good job"

[NumbersNerd]

nydailynews.com/news/national/ny-nate-silver-fires-back-at-fivethirtyeight-critics-20201105-d5ho3nfkmrc2lo7tsags6jtdi4-story.html

Nate Silver says: "If they’re coming after FiveThirtyEight, then the answer is **** you, we did a good job."

What do y'all think about that?

[lfromnj]

Bullsh**t, you gave all the D hack pollsters A grades, if the data coming was garbage you should have called those pollsters out, instead you spend half that time railing against Trafalgar who although not great was probably a touch better than the rest. If you don't change those grades because their metholdogy is correct then you are doing your job wrong..

[Technocracy Timmy]

He knows his industry will take a hard hit after this and is trying to salvage it.

[SevenEleven]

Shambles.

[McGarnagle]

Bad on the margins, but if most of the uncalled states go to Biden, relatively good on predicting the winner.

[Absentee Voting Ghost of Ruin]

Pretty sure his "snake" came out accurately, which is pretty impressive given how inaccurate the polls he was working from turned out to be.

[lfromnj]

Pretty sure his "snake" came out accurately, which is pretty impressive given how inaccurate the polls he was working from turned out to be.

Florida?

[McGarnagle]

Pretty sure his "snake" came out accurately, which is pretty impressive given how inaccurate the polls he was working from turned out to be.

Florida?

Except for Florida

[It's Perro Sanxe wot won it]

Pretty sure his "snake" came out accurately, which is pretty impressive given how inaccurate the polls he was working from turned out to be.

Florida?

Except for Florida

And probably North Carolina

[lfromnj]

Also lol ME02 too.

[Buffalo Mayor Young Kim]

This happens in baseball analytics, where Nate started out, allot too. ‘The thing you said had a 30% chance of happening happened, where’s your math now nerd?’

[kwabbit]

To be fair, it's hard to build an accurate model off polling if the polling is inaccurate. Again, Silver had the best model, giving Trump much better odds to perform as well as he did than the Morris' model. On a separate note, it always felt like Morris let his partisanship affect his modeling decisions. He excluded what turned out to be valid pollsters like Trafalgar and Emerson on the grounds of having unreliable data. I feel like this had to do with them giving good results for Trump. This was defensible, but I feel like he would not have excluded junky polls if they were giving good results for Democrats.

[Mike Tyson]

What a joke. WI, the likely tipping point state, was off by ~8 points

[Lief 🗽]

Embarrassing. I assume ABC is just waiting for the counting to be done before they fire him?

[GeneralMacArthur]

To be fair, it's hard to build an accurate model off polling if the polling is inaccurate. Again, Silver had the best model, giving Trump much better odds to perform as well as he did than the Morris' model. On a separate note, it always felt like Morris let his partisanship affect his modeling decisions. He excluded what turned out to be valid pollsters like Trafalgar and Emerson on the grounds of having unreliable data. I feel like this had to do with them giving good results for Trump. This was defensible, but I feel like he would not have excluded junky polls if they were giving good results for Democrats.

Trafalgar wasn't a valid pollster.

People need to stop saying that all the Republican hack pollsters were good just because they were closer to the actual result.

I could make a map where I just give Trump every single state by 5 points, and then point to FL and say "look I had Trump winning FL by 5 and he won it by 3!  I beat all the fake news pollsters!"  or "look I had Trump winning WI by 5, and he only lost it by 1, that's way better than the fake news pollsters who had Biden +9!"

The fact of the matter is that the mainstream pollsters were wrong, but they were mostly wrong directionally.  Trump got way better turnout than expected so you basically shift every poll 4-5 points in Trump's favor and you're close to the real results.  That's why Nate's snake map ended up being just about right, with GA the only outlier.

On the other hand, Trafalgar was just predicting a Trump win in every swing state based on nothing other than quantified "gut instinct" about "shy Trump voters."  So you had totally wrong numbers like Trump +2 in MI, +2 in PA, +1 in NV, +5 in GA, +3 in AZ, and so on.  Those are wrong numbers, and there's not any rhyme or reason to their wrong-ness.  They're just random Trump-friendly numbers that Trafalgar got not through any scientific methodology, but rather by starting with the final result they wanted, doing a poll, and then fabricating priors to massage the results until they got to their desired outcome.

Why does this matter?  Because if all polls were like Trafalgar, polling would be worse than useless.  It means all Trafalgar basically is is one guy's gut instinct translated into numbers.  It's no more useful than an SN2903 map.

[kwabbit]

To be fair, it's hard to build an accurate model off polling if the polling is inaccurate. Again, Silver had the best model, giving Trump much better odds to perform as well as he did than the Morris' model. On a separate note, it always felt like Morris let his partisanship affect his modeling decisions. He excluded what turned out to be valid pollsters like Trafalgar and Emerson on the grounds of having unreliable data. I feel like this had to do with them giving good results for Trump. This was defensible, but I feel like he would not have excluded junky polls if they were giving good results for Democrats.

Trafalgar wasn't a valid pollster.

People need to stop saying that all the Republican hack pollsters were good just because they were closer to the actual result.

I could make a map where I just give Trump every single state by 5 points, and then point to FL and say "look I had Trump winning FL by 5 and he won it by 3!  I beat all the fake news pollsters!"  or "look I had Trump winning WI by 5, and he only lost it by 1, that's way better than the fake news pollsters who had Biden +9!"

The fact of the matter is that the mainstream pollsters were wrong, but they were mostly wrong directionally.  Trump got way better turnout than expected so you basically shift every poll 4-5 points in Trump's favor and you're close to the real results.  That's why Nate's snake map ended up being just about right, with GA the only outlier.

On the other hand, Trafalgar was just predicting a Trump win in every swing state based on nothing other than quantified "gut instinct" about "shy Trump voters."  So you had totally wrong numbers like Trump +2 in MI, +2 in PA, +1 in NV, +5 in GA, +3 in AZ, and so on.  Those are wrong numbers, and there's not any rhyme or reason to their wrong-ness.  They're just random Trump-friendly numbers that Trafalgar got not through any scientific methodology, but rather by starting with the final result they wanted, doing a poll, and then fabricating priors to massage the results until they got to their desired outcome.

Why does this matter?  Because if all polls were like Trafalgar, polling would be worse than useless.  It means all Trafalgar basically is is one guy's gut instinct translated into numbers.  It's no more useful than an SN2903 map.

Trafalgar is not a good pollster. Trafalgar is a valid pollster. They do not fabricate data. Part of a pollster's job is to make assumptions about the electorate, and Trafalgar's 'Shy Trump' shift is an assumption about the electorate, however crude and unscientific it is. While I don't think it's cause of a shy Trump effect, it does seem the low propensity WWC voters that flock to Trump do not answer polls, and places with a lot of WWC have had repeated systematic bias in favor of Democrats in polls. Trafalgar's adjustment has made their polling the Midwest more accurate than gold standard polls for the past 3/4 cycles and eventually that has to be recognized. Banning them from your polling average is frankly a boneheaded partisan decision at this point, no matter your opinion of them.

Bottom line, no matter how sh**tty a pollster's methodology, if they have repeated accuracy, they need to be given credit for that.

[Pyro]

[Gustaf]

He certainly did a better job than most Atlas posters at predicting this.

[One Term Floridian]

He deserves to lose all the money he has for screwing with our expectations. Never giving him a click again.

[John Dule]

People need to stop saying that all the Republican hack pollsters were good just because they were closer to the actual result.

I'm sorry, but taken out of context, this quote made me laugh my ass off. Holy crap, lol.

[Santander]

538 both deserves credit for being more sober-minded than most of its peers, and also criticism for its significant shortcomings.

[Tekken_Guy]

Yeah the models did really screw up. However, I think polling is a very hard industry. Definitely not a fan of his attitude here, though.

[AncestralDemocrat.]

You can redeem yourself Nate by downgrading PPP and Quinnipiac to D rating.

[Figueira]

Do people here really not understand that he isn't claiming to be a psychic?

[Statilius the Epicurean]

People need to stop saying that all the Republican hack pollsters were good just because they were closer to the actual result.

I'm sorry, but taken out of context, this quote made me laugh my ass off. Holy crap, lol.

It's true though. Throwing a dart at a board blindfolded might have hit on Biden +4, but that doesn't mean the methodology behind the prediction was sound.

Nate Silver changed Trafalgar to a partisan Republican pollster because it emerged they were doing polling for Republican candidates without telling anyone. Their polls were still in the model. In fact on one of the model talk podcasts Nate explicitly said that he might not trust Trafalgar but they weren't removed in order to balance out the D house effect of other pollsters that might be missing something, if they were missing something.

Either way I think 538 did pretty well. The model said Biden was favoured because he would win even if there was a 2016-style polling error. That happened and Biden won.