SXSW: Nate Silver Discusses Data Bias, the Strangeness of Fame

Nerval's Lobster writes "Nate Silver feels a little odd about his fame. That's not to say that he hasn't worked to get to his enviable position. Thanks to his savvy with predictive models, and the huge readership platform provided by The New York Times hosting his FiveThirtyEight blog, he managed to forecast the most recent presidential election results in all 50 states. His accuracy transformed him into a rare breed: a statistician with a household name. But onstage at this year's SXSW conference, Silver termed his fame 'strange' and 'out of proportion,' and described his model as little more than averaging the state and national polls, spiced a bit with his algorithms. "It bothered me that this was such a big deal," he told the audience. In politics, he added, most of the statistical analysis being conducted simply isn't good, which lets someone like him stand out; same as in baseball, where he made his start in predictive modeling. In fields with better analytics, the competition for someone like him would be much fiercer. He also talked about, despite a flood of data (and the tools to analyze it) in the modern world, we still face huge problems when it comes to actually understanding and using that data. 'You have a gap between what we think we know and what we really know,' he said. 'We tend to be oversensitive to random fluctuations in the data and mistake the fluctuations for real relationships.'"

Comments

silver is honest

[SlippyToad]

I think Silver stands out because unlike too many modern American politicians, he is interested in the facts, and not what bullshit he can use the data to support.

So it's not so much that he's done a fantastic job figuring all this out, it's just that he's fucking honest about the results unlike a certain perpetually-deluded political party I'm sick of naming.

Re:silver is honest

[Hatta]

You can't deadlock them, you can only deadlock us. The Democrats and Republicans have a lot more in common than they have differences. The rich people who control the Democrats have much more in common with the rich people that control the Republicans than either have with any of us. The worst thing that could happen to the Democratic party is for the Greens to win some major elections.

If the Democrats are 95% wrong, and the Republicans are 99% wrong, that means that 90% of the time they are completely unopposed in doing the wrong thing. If you vote for either Democrats or Republicans, that's what you're voting for.

Re:silver is honest

[fuzzyfuzzyfungus]

I think Silver stands out because unlike too many modern American politicians, he is interested in the facts, and not what bullshit he can use the data to support.

So it's not so much that he's done a fantastic job figuring all this out, it's just that he's fucking honest about the results unlike a certain perpetually-deluded political party I'm sick of naming.

Arguably, it isn't really politicians who he differs from most meaningfully. Sure, there are a lot of politicians living in absurd contrafactual fantasy worlds; but that is(unfortunately) mostly a product of the fact that they are acting as representatives of people who do exactly the same thing... Pandering is a nonfactual enterprise in the sense that it may involve telling people the most insane lies, if that is what they want from you; but it is an eminently empirical exercise in the sense that you must constantly strive to better understand what people want to hear, so that you can better pander to them.

Where Silver, and his data-driven compatriots, really differ from the traditional is with the 'pundit' class. Pundits are selected pretty much entirely for their ability to tell emotionally compelling stories, with minimal reference to data, and provide marketable column inches and cable news minutes. The better ones, to their credit, are masterful in engaging audience emotions, weaving stories, and other affectively gripping flimflam. However, they tend to be somewhere between extraordinarily weak and overtly hostile to the idea that 'data' rather than 'feelings' can actually provide excellent information about the world, particularly if you use this crazy 'math' stuff that the nerds are always going on about.

Pundits make good TV(and, very conveniently, can offer viewers everything from lowbrow talk radio shouting matches to middlebrow 'public intellectual' posturing with little more than a change in tone and presence or absence of a thesaurus, unlike stat-heads who pretty fundamentally lean on nontrival math); but the kind of suck compared to statistical models.

Re:silver is honest

[Hatta]

But they do. Warrantless wiretapping, eternal copyright, the war on drug users, the ever increasing militarization of the police, the for-profit prison industry, and a DOJ that cares more about Aaron Swartz than about John Corzine, etc, etc,. The Democrats and Republicans agree on most of the most harmful policies that afflict this country. And when they pretend to disagree (e.g. Rand Paul on domestic drone strikes), it's all for show.

Re:silver is honest

[SlippyToad]

Just to pile on, when you say "both sides do it," you are implicitly refusing to deal with the actual topic at hand, which is for example "budget" or "national security," or whatever.

So when you do that, you are basically throwing up your hands and saying "who can know such things?"

It's fucking lazy. Very, fucking lazy. I don't have much time to argue with people too lazy to at least delve into the elements of a topic. You obviously are.

Science is rare

[Hatta]

Very, very little in this world actually happens because the data suggests its a good idea. People make decisions based on their comfort level, tradition, who their friends are, etc. Suggesting that we should listen to the data disempowers the powerful. It's 2013, and the principles of evidence based medicine were only developed 20 years ago, and are still not widely used in practice. We're going to have to wait centuries before evidence based public policy becomes the norm.

Re:Science is rare

[Hatta]

The thing is you can have mountains of data (which we do). The problem is asking the right questions.

The problem is that those in power dont want to ask the right questions because it will force their hand. e.g., we have mountains of data that show that Cannabis is less harmful than most over the counter drugs. Certainly less harmful than alcohol. So we 75,000 of us get together and ask the President a question. "Why can't we regulate Cannabis like alcohol?"

His response amounted to "Cannabis is a dangerous drug", and he never mentioned alcohol once. He didn't even bother dismissing the question, he failed to acknowledge its existence.

Why would he do that? It's obvious, if he actually considered the question, he would have been unable to come up with an answer that would be consistent with our current policy. He knows that, and he doesn't care that the policy is demonstrably harmful.

That, my friends, is the face of evil.

Optimism

[sjbe]

We're going to have to wait centuries before evidence based public policy becomes the norm.

I think you are being optimistic. Very optimistic.

Deadlocked into debt

[sjbe]

Deadlocking them produces the best feasible outcome.

Except on the periodic occasion when we need them to do actually something. You know, like not endlessly raise the national debt because they want to promise everything but don't want to have to tell the voters they have to actually pay for it someday.

"Right" is frequently just opinion

[sjbe]

Both sides are not equally wrong, but that doesn't mean both sides are right. The Republicans are wrong 99% of the time. The Democrats are wrong 95% of the time. Why can't we field a candidate who's right even half the time?

Because "right" is for better or worse often a matter of opinion. There is no single objectively right answer for many questions. Think abortion or gun control. Lots of opinions on both sides but there is never going to be a single "right" answer. At best there might be a consensus but probably never a unanimous one. Even for questions where a single objectively right answer may theoretically exist, there often is insufficient data to figure out what that answer is. (for example what is the optimal tax rate)

Valid Criticism

[Capt.Albatross]

Cathy O'Neil (Mathbabe) offers a well-argued criticism of Nate Silver when he stepped beyond his area of expertise in his recent, popular book, '"The Signal and the Noise: Why so many predictions fail – but some don’t"

http://mathbabe.org/2012/12/20/nate-silver-confuses-cause-and-effect-ends-up-defending-corruption/

According to Ms. O'Neil, Mr. Silver fails to recognize situations where bad models are deliberately used to game that system.