The Clear Liberal Bias of Math
Perhaps the most deliciously ironic element of the election was watching all the wingnuts convince themselves that Nate Silver (always him and not Nate Cohn, Sam Wang or anyone else for some reason) was using mathematical voodoo to make it falsely look like Obama was winning and then watch him be almost dead on in his projections. On Facebook, someone said, “After tonight, Nate Silver now has the right to end every sentence with the word ‘motherfuckers.’”
In reality, Silver called every swing state correctly, 9 for 9. And when all the votes are counted, he’ll be off on the popular vote by .3% or so. And just as Silver said, Gallup and Rasmussen were way off. Not a bad track record, especially after his equally accurate projections in 2008 and 2010. In fact, all of the major poll aggregators and analysts were damn close in their projections. I guess math actually works, eh?
Drew Chambers of Unskewed Polls went into almost total radio silence on election night. Nothing on Twitter, no updates to his site, nothing. To be fair, he’s probably busy working on unskewedelections.com, where he’ll explain that if you unskew the votes by reducing Democrats by 5% and boosting Republicans by 5%, Romney really won and is now the true president. Come out, come out wherever you are, Dean. You can’t hide forever. You’re going to have to admit that the “thin and effeminate man with a soft-sounding voice” just showed how irrational and delusional you were all alone.
Here’s the punchline: The morning after the election, Jonah Goldberg published an article complaining about Silver’s “numbers racket.”
Now, I have no idea whether Silver’s model is the psephological Rosetta Stone some hope — or fear — it to be. And no one else does either.
The truth is that any statistician can build a model. They do it all the time. They make assumptions about the electorate, assign weights to polls and economic indicators, etc., and then they wait for the sausage to come out. No doubt some models are better than others, and some models are simply better for a while and then regress to the mean. But ultimately, the numbers are dependent on the values you place on them. As the computer programmers like to say, garbage in, garbage out.
I’m not saying Silver’s just lucky or shoveling garbage. He’s a serious numbers guy. But so are the folks at the University of Colorado’s political science department whose own model is based on economic indicators. Its Oct. 4 findings predicted Romney would win, as did many other models.
They couldn’t all be right.
Actually, we do know, Jonah. We know because he’s been right every time so far. And for you to publish this after he was proven right and all those right wingers who’ve been criticizing him were proven wrong is taking the idea of ducking into the punch to a whole new level.
dingojack:
November 8th, 2012 at 12:12 pm
Since Jonah Goldberg seems piss-poor at psephology, perhaps he should be ostracised.
Dingo
marcus:
November 8th, 2012 at 12:17 pm
You fucking go Nate! And… He has a book! Which I’m sure along the way demonstrates why Goldberg is a shithead and why the U of C didn’t get it right. I just started it myself. It’s called “the signal and the noise, why so many predictions fail- but some don’t”. Find it at your local independent bookstore (please). Take that motherfuckers!
Sastra:
November 8th, 2012 at 12:22 pm
Ok, just my personal preference here, but I do think that’s the kind of voice which says “motherfuckers” the best.
eric:
November 8th, 2012 at 12:30 pm
Yup, as the Daily Show (and probably lots of other wags) have put it: the election results are in, and statistics beat punditry in a landslide victory.
Not being one for a lot of swearing, I think Silver could go a bit more understated than just calling his critics mfs. Just walk around with a shirt that says “Who’s your Daddy?” for a few days. :)
Didaktylos:
November 8th, 2012 at 12:30 pm
It’s all because the voter suppression operations were a resounding failure …
Bronze Dog:
November 8th, 2012 at 12:31 pm
My brother’s taking a statistics course this semester and he’s been on Fark, trying to teach some trolls what probability means. Even if probability favored Romney this one time, that wouldn’t prove Silver’s methodology wrong. It’d take multiple trials, and if the results didn’t conform to the general odds when viewed as a whole, then there’d be cause to say he’s wrong. F[rell]ing statisticals, how do they work?
It seems American society is getting steadily more innumerate.
sailor1031:
November 8th, 2012 at 12:32 pm
“’m not saying Silver’s just lucky or shoveling garbage. He’s a serious numbers guy. But so are the folks at the University of Colorado’s political science department whose own model is based on economic indicators.”
How long will it be, do you think, before the ‘folks at the University of Colorado’ give up their obviously flawed models? Wanna bet they’re still pushing them in 2016?
F:
November 8th, 2012 at 12:34 pm
tbp1:
November 8th, 2012 at 12:34 pm
Glad to see most of the comments over there rake Goldberg over the coals.
dingojack:
November 8th, 2012 at 12:37 pm
F – nice to see you’re emulating the ‘brightest’ and the ‘best’ of the RW pundits, then.
:D Dingo
dave:
November 8th, 2012 at 12:41 pm
If youre not aware of it, the hashtag #natesilverfacts is pretty hillarious. Essentially Chuck Norris jokes for the geek-set.
glodson:
November 8th, 2012 at 12:41 pm
Nate Silver, and all the other guys that used math to make highly accurate predictions based on sound principals, should do a victory lap around the Fox News studios. They deserve it.
garnetstar:
November 8th, 2012 at 12:42 pm
Didaktylos @5 is right. Here’s Chambers agreeing.
Although he calls it “my belief that a nearly equal percentage of Democrats and Republicans would turn out in the actual election this year”, what he means is that he was counting on “five or six or seven percent” of Democratic voters being disenfranchised.
hunter:
November 8th, 2012 at 12:53 pm
1) Goldberg is a proven idiot.
2) Dogpile on Silver from the right because he’s gay.
Reginald Selkirk:
November 8th, 2012 at 1:03 pm
Of course maths are liberal – all that integration going on.
Reginald Selkirk:
November 8th, 2012 at 1:07 pm
Top 25 Nate Silver Facts
Some of these are pretty good.
Area Man:
November 8th, 2012 at 1:11 pm
I’m not sure if Goldberg understands that bad models produce bad results, and good ones produce good results. It is not a legitimate critique of Silver to point out that someone else had a model that didn’t work well. This is reminiscent of climate change denialists (usually the same people here) who dismiss future predictions of temperature change merely on the basis that they’re generated by “computer models”. They are convinced that this alone is reason to ignore them without any attempt to show why any particular model is wrong.
For what it’s worth, the UC guys were not relying (at least primarily) on polling data. They went looking for economic correlates over the last several elections, which creates the obvious possibility of over-fitting. If they’re serious academics, and I’m sure they are, they’ll be ditching that model. All the models based on poll aggregates (that didn’t try to unskew them) were highly accurate.
tassilo:
November 8th, 2012 at 1:16 pm
Over on Real Climate, there is an interesting post (http://www.realclimate.org/index.php/archives/2012/11/trying-to-shoot-the-messenger/)comparing attacks on Nate Silver’s projections with attacks on climate science. They have been enduring a barrage of number illiterate attacks, from the same sources, for years. As they say in the post, Nate Silver was vindicated in a day, global warming will take a little longer.
Btw, the CU Boulder folks seemed no more dogmatic about their model than Nate Silver was about his, and I’m fairly sure they don’t have a political ax to grind. They acknowledged from the start their model could be wrong. Model runs were calibrated with data from previous elections, but there was not ever a guarantee that it would accurately predict the outcome of this one.
Michael Heath:
November 8th, 2012 at 1:34 pm
For a taste of secular conservative authoritarian thinking, it’s fun to peruse the comment posts at the Wall Street Journal. One could build a decent defense that perhaps it isn’t religious thinking that makes people incapable of coherency, but instead authoritarian thinking. [I know, these two are not mutually exclusive.]
I bring this up here because the pre-election articles on the polls had those WSJ commenters not only criticizing liberals like Nate Silver, but also claiming that the Wall Street Journal and Fox News were rigging their polls. That both were also participating in a liberal conspiracy to help President Obama get reelected.
At least religious conservatives get some bliss via their fantasies, but the crowd at the WSJ is just plain bitter and hateful.
scienceavenger:
November 8th, 2012 at 1:45 pm
The UC model was a joke from the start. It had Romney winning Minnesota but losing Nevada. Not in a hundred years. I offerred every GOPer that forwarded me that crap a 10:1 bet against it happening. I got no takers.
umlud:
November 8th, 2012 at 2:31 pm
What I like about Nate Silver’s 538 blog is that he is pretty open about discussing topics of polling bias and how to deal with it. I’m not certain, but it seems that his style of statistics is (or is similar to) Baysian statistics, and – as such – it really isn’t the same kettle of fish that Goldberg is talking (out of his ass) about.
If Silver does use Baysian statistics (and it seems like he is and that he’s using it correctly), then comparing it to more standard methods of statistics is akin to comparing apples to cherries: both are fruit, both grow on trees, both are red (or have varietals that are red), both make tasty pies, and both are major cash crops in Washington. However, beyond those points, there is little (credible) comparison that one can make between the two. Seems to me that Goldberg (and so many so-called pundits) need to actually learn what statistical methods actually are, how different methods deal with problems of bias, how surveys done over time are assessed differently than a single survey, etc.
The only major problem that I see with things like this is that, if people start to really pay attention to sites like the 538 blog, then that large-scale inspection of polls will then cause a skew in the polls, thus reducing the reliability of the polls (and possibly the assessment of the polls).
laurentweppe:
November 8th, 2012 at 2:34 pm
Yeah and any bricklayer can biuld a wall.
The point is that you’re neither a bricklayer nor a statistician, so you refrain from pretending you have any expertise in walls or statistical models, Jonah.
DaveL:
November 8th, 2012 at 2:35 pm
Are we supposed to be surprised that a model based on asking people whom they’re going to vote for, and adjusting those polls depending on how accurate they’ve been in the past, turned out to be more accurate than a model that measures things not directly related to voting?
umlud:
November 8th, 2012 at 3:09 pm
But it can also be conservative, I mean look at all that differentiation! And we know that everyone is susceptible to imaginary numbers… After all, knowing how to distribute who is rational or irrational is sometimes difficult, but I think that this is – at its root – a product of the divisions that have become constants. In some ways, this is now a simplified political world: the GOP are reaching a demographic limit, based on their axioms and proofs of being a “real American”. That set of people has reached its maximum, and – unless they can change their identity – their dimensionality is only going to shrink to non-significance.
;-)
yoav:
November 8th, 2012 at 4:24 pm
In an unpredicted attack of journalism Fix Noise host Megyn Kelly asked Karl Rove if he wasn’t just using math, as a Republican, to “make himself feel better.
Chiroptera:
November 8th, 2012 at 4:30 pm
umlud, #24: After all, knowing how to distribute who is rational or irrational is sometimes difficult….
We’re doomed. However many rational people there are, the set of irrationals has a greater cardinality.
bksea:
November 8th, 2012 at 4:33 pm
I have this vision that all the right wingers have a tattered copy of The Secret on their bookshelves. They think they can change the polls by wishing them to be different.
cjcolucci:
November 8th, 2012 at 4:49 pm
laurentweppe makes an important point. To the extent that what Goldberg says is true — and taken in isolation (which would include forgetting that this is Jonah Fucking Goldberg), there would much to be said for it — there’s simply nothing for Goldberg to say. Serious numbers guys build lots of models. Some work better than others. What works now might not work later. Goldberg, however, can’t shed any light on the subject. So why is he wasting his time and ours?
iknklast:
November 8th, 2012 at 4:52 pm
Didn’t you know, math’s always correct when you’re adding the ages of prophets to predict the age of the earth; it’s wrong when you’re predicting the ‘wrong’ person will win the election.
baal:
November 8th, 2012 at 5:10 pm
“using mathematical voodoo” <– For people who don't actually understand Science or math (or reality as a concept for that matter); it can seem an awful lot like magic especially when you can state something with certainty when it looks like that item is in the 'could go either way' category. Every now and then I find myself saying, "you know, that is knowable" or "that isn't a knowable".
TL;DR – Science is the best magic.
umlud:
November 8th, 2012 at 6:48 pm
Chiroptera @24: I didn’t know that theorem. Sounds sound, though. :D
gwangung:
November 8th, 2012 at 7:46 pm
Serious numbers guys build lots of models. Some work better than others. What works now might not work later.
Yup. And if you know the models well and why they work the way they do, you could predict what factors cause them to be better or worse.
Of course, that takes a little science and math……
Ichthyic:
November 9th, 2012 at 1:36 am
Gallup and Rasmussen were way off.
I’m going to remember that.
I’ve often relied on Gallup poll data.
now I will always be highly suspicious of it.
Nibi:
November 9th, 2012 at 8:30 am
Reginald Selkirk
And, even worse, homomorphisms.
alwayscurious:
November 9th, 2012 at 10:10 am
People vote on more than just economic indicators!?! Is this truly news to anybody?
The warning flags were out about Gallup two weeks before the election. Most of the articles I read weren’t very informative, but basically of the tone that Gallup was more likely wrong than prophetic. And as a result, would likely be changing their algorithm as soon as they figure out how it went wrong.
blf:
November 9th, 2012 at 10:25 am
And we mathematicans commute using a “secret” language, like
∀μ, μ ∉ ℜ, ℱυ∑ⲕ(μ)
where ℜ is reality.
(Apologies to those whose installed fonts aren’t sufficiently rich…)
zippythepinhead:
November 10th, 2012 at 10:41 am
What was most fascinating about Nate’s data is that he was able to put a probability of winning any particular state. With that data (and assuming it is correct) it is straightforward to create a probability distribution of electoral votes. That is a fundamentally powerful statistical method that most people won’t get who aren’t math inclined. It is that distribution that predicted Obama had a 97% chance of winning more than 270 EV. My own calculation on election day confirmed that and said that Obama had a 50% chance of winning more than 315 EV, which ultimately he did.