Nate Silver runs for re-election as America’s President of Data

Nate Silver vs. the pundits is being billed as math vs. language, as data vs. opinion, as science vs. art, as robot vs. human

Silver at the 2011 World Series of Poker

Nate Silver is a statistician—a baseball numbers nerd who became a politics numbers nerd in 2008 when he correctly predicted election outcomes in 49 of 50 states (he was off in Indiana, which went to Obama by 1%). He also correctly called every ’08 Senate race, and has subsequently risen to fame as the Nostradamus of U.S. politics. Lately though, he’s become a target.

Silver’s FiveThirtyEight blog, licenced to the New York Times, has been heavily favouring an Obama win from the start. Whereas individual polls find the candidates in a dead heat, Silver’s analysis (as of this moment) says that Obama is an 80.9% favourite. Silver also happens to be an Obama supporter, a fact which he discloses in the name of transparency, but which he says bears no influence on his predictions, which are the result of statistical modelling, not of his personal political persuasion.

Conservative critics aren’t buying it, and have been characterizing Silver as a partisan warrior disguised as a pocket calculator. Though they’re sure his numbers are way off the mark, they fear a self-fulfilling prophecy wherein swing state voters, otherwise woo-able by Mitt Romney, will submit to the seeming inevitability of Nate Silver’s blog and vote Obama.

This is silly, but not shocking. I’m a little more surprised by the other Nate Silver war. Not the battle between Democrat and Republican, but between political columnists and social media mavens who are casting the ruckus as a grudge-match between old media and new. Nate Silver vs. the pundits is being billed as math vs. language, as data vs. opinion, as science vs. art, as robot vs. human. If his predictions come true, then we must welcome our new Big Data overlords. If not, then chalk it up as a win for the old school, for ink-stained politicos and cable news hosts whose news savvy and gut instincts are worth more than any fancy computer wizardry.

Don’t believe it.

What Nate Silver offers is a highly informed estimate. While we lose ourselves in discussions about who seemed more “presidential” in the last debate or whose gaffe generated the most humiliating meme, Silver looks at numbers. He looks at myriad polls, demographic analysis, historical data- anything that has shown some predictive power and that can quantified. But just because Silver uses numbers, that doesn’t make what he does math.

His algorithm, his “secret sauce,” is created by making decisions about which data to favour, about how much weight to give each poll and input. These decisions are a new kind of editorial judgment. Might they be influenced by Silver’s political bias? Sure. But even if Silver wants Obama to win, it’s much more important to him that he win. The numbers that Silver’s algorithm spits out are not answers, just one man’s predictions. We value them only because Silver’s predictions were so good in the past. But if a Vegas oddsmaker is wrong too often, the gambling man will find a new handicapper.

“I’m sure that I have a lot riding on the outcome,” Silver told Buzzfeed. “I’m also sure I’ll get too much credit if the prediction is right and too much blame if it is wrong.”

What Silver’s saying is that giving Obama an 80% chance of winning is also giving him a 20% chance of losing.  Roll snake eyes, and the dice aren’t wrong- you’re just unlucky.  But dice odds are set.  Each number has an equal and persistent chance of coming up.  Silver’s goal as an oddsmaker is to define an incredibly complex problem as a predictable, and thus solvable problem. It’s not.  Silver doesn’t know anything empirically- he’s just really, really good at guessing.  Politics, like poker (which Silver plays) are filled with imperfect information and unknown factors.   Even the best players hold their breath on the river.

Follow Jesse on Twitter @JesseBrown

 

 




Browse

Nate Silver runs for re-election as America’s President of Data

  1. Roll snake eyes, and the dice aren’t wrong- you’re just unlucky. But
    dice odds are set. Each number has an equal and persistent chance of
    coming up.

    ***

    This is at the very least worded improperly. The number on each die has an equal chance of coming up, which makes for different probabilities when you combine more than one die. When rolling two dice, for example, the chances of rolling 7 are far greater than rolling two or tweleve.

    • Improperly worded. I mean to say that each number on one die has an equal chance. Thx!

  2. Silver himself says ( on Charlie Rose .. when Charlie stopped blathering long

    enough for Silver to say anything) that he is not an Obama supporter. Because

    of his current position, he does not vote in presidential elections. He says his

    own views place him somewhere between Gary Johnson and Mitt Romney !!

  3. Silver just on tonight”s CBC As It Happens (1st half hour, about 10 min in).

    Here’s how I understand he adjusts current polls – he looks at previous track record of a pollster, and adjusts for bias /error based on previous polls.

    But, it seems to me that this approach will eventually fail if the individual pollsters do the same thing (which it seems to me they are doing now) – take their raw data and adjust for prior bias (in effect it would eventually get corrected twice – once by the pollster, then by Silver).

    In other words, refried refried beans eventually overpower the special sauce.

    • No doubt at some point he will walk into the wall. He says so himself.
      But (he says) that’s not what it’s all about … not about being right or
      wrong … it’s about probabilities. The fun part of it all is the Boys on the
      Bus defending their position as interpreters if the campaign spinners.
      As though they were immune.

      • Yeah, I don’t agree with “it’s all about probabilities” when you engage in the social “sciences”. Especially with forecasting voting patterns of individuals who can change their minds. In today’s day and age with all the feedback loops.

        Mind you, with all of the gerrymandering in the U.S. , and limited choices (Rep or Dem) can seem to narrow it down considerably.

  4. Worth noting that Silver’s methodology is a ‘secret sauce’ in that his methods are proprietary. These analyses on the other hand just look at the pools and has non-proprietary methods:http://horsesass.org/?p=47473

    Much more scientific..and fitting of the ideals of openness many of us aspire to than Silver’s work.

  5. The difficulty with Silver’s work is that he builds large, heavily customized, complicated models that work extremely well until they don’t. Because he is constantly tweaking his model to try and account for every possible variable it’s very vulnerable to edge cases where one unaccounted for element skews the whole result. It’s the reason why in coding we occasionally take everything we learned from one solution and rewrite it from the ground up.

    Silver doesn’t appear to do this, he just keeps adding to his model to the point where eventually it’s too complicated to even rewrite well. As consequence you get things like his model’s complete failure at forcasting the British vote or that when he passed on his baseball forecasting model they had to essentially toss it out as too complicated, hard to work with, and vulnerable to edge cases.

    Of course I still regularly read his blog because the work is fascinating and there’s a good chance it’s accurate, that and if it fails it will be spectacular. :D

Sign in to comment.