Princeton Election Consortium

Innovations in democracy since 2004

Outcome: Biden 306 EV (D+1.2% from toss-up), Senate 50 D (D+1.0%)
Nov 3 polls: Biden 342 EV (D+5.3%), Senate 50-55 D (D+3.9%), House control D+4.6%
Moneyball states: President AZ NE-2 NV, Senate MT ME AK, Legislatures KS TX NC

Nate Silver’s move

July 24th, 2013, 11:04am by Sam Wang

By now most readers know about Nate Silver’s move from the New York Times to ESPN and ABC. ESPN purchased the FiveThirtyEight domain, suggesting a significant commitment by Silver. The move fits with Silver’s roots in sports, as well as his outsider status with regular journalists. ABC will provide excellent broadcast opportunities during election seasons. And I’m sure that money played a major role.

My take is very similar to that of Matt Yglesias at Slate, minus a little of the anti-pundit/journalist invective. Basically, poll aggregation is not that hard to a statistically-minded professional. In my view, prediction is not hard either, except for the step of identifying true predictive factors. Silver’s model includes components of low probable added value. For instance, the econometric components almost certainly give only a low-resolution echo of the polling measurement.

I note that in July/August, here at PEC we had Obama’s re-elect at 91%. This was based on a random drift model, which is clean and is known to lack redundancies. For those who want more complexity, Drew Linzer (Votamatic) has perhaps done the best job of combining opinion and econometric data. Josh Marshall values the FiveThirtyEight model, but I think he’s wrong for the reasons I have given. (He also neglected PEC’s rather good record in the 2012 campaign…but that’s life out here in the world of unpaid blogging). Last August I wrote generally on the value of political-science-type forecasting models.

In my view, Yglesias (and Ezra Klein too) put their finger on what Silver added: storytelling. From 2004 to 2010, I ran a simple site with very little daily essaywriting. My traffic was low. Adding the daily column drove traffic up tremendously. The fact is that readers like the narrative. This is Nate Silver’s contribution: finding a way to make the numbers into a compelling play-by-play. Again, it’s the sports guy in him.

What I find amazing is that journalists don’t use numerical facts in this way. In 2012, one of my favorite themes was singling out a political writer whose narrative was contradicted by known facts (Peggy Noonan, David Brooks, and John Dickerson come to mind). In future campaigns, I hope that kind of Politico-style b.s.-ing will be more constrained by quantifiable information, as Nate and I do.

Update: TIME’s James Poniewozik has an excellent analysis of this last point: the advent of using high-quality data analysis to make journalism better.

Tags: 2012 Election · Uncategorized

25 Comments so far ↓

  • bks

    Are you going to give the 2014 midterms the PEC treatment? I realize there’s a lot less grist for the mill.

  • Amitabh Lath

    So… there’s a vacancy at the NYTimes for someone who can write from a mathematically literate perspective while running a poll aggregation prediction engine.

    Perhaps what they need is a Princeton professor who has talents and interests in this area.

    Seriously, Prof. Krugman seems to pull it off effortlessly. You guys could commute in together on staff meeting days.

  • Sam Wang

    bks – probably. I wonder how much value I can add beyond what simple aggregators do. In 2010 I pointed out the impending GOP takeover of the House…as did a number of people.

    One interesting question comes to mind: is it possible to extract time-series data on the national dynamic? In other words, a day-to-day House/Senate tracker that says where the overall campaign is headed, like the Presidential EV snapshot from 2012? That could be interesting, though sharp moves like the first Obama-Romney debate are unlikely.

    Amitabh Lath – that’s very nice of you to compare me with PRK. One difference is that to my knowledge, he does not run a full-time research group.

  • MarkS

    It’s certainly true that “poll aggregation is not that hard to a statistically-minded professional.” The real story is how few statistically-minded professionals are working in mainstream media, and the stunning innumeracy of the vast majority.
    (Note that even Yglesias gets it wrong when he says that you and Silver average polls.) I see no signs that this sad situation is getting any better.

  • pechmerle

    SW, what I hope PEC will do for 2014, as it did in 2012, is show us which House and Senate races our contribution dollars will have the most effect on.

  • pechmerle

    Poniewozik’s is a very good piece. But note the sad fact that TIME put it under its Entertainment category. Let’s hope that’s not really how they think of numerate political journalism.

  • Jim

    Wow…now he gets to “debate” with the likes of Skip Bayless and Stephen A. Smith. He may be making more money, and more power to him, but this is definitely a step down on the journalistic food chain.

    • Sam Wang

      What, again?

      There are ways to address that – doing the same analysis that I did in the NYT, year by year. You are right, I should get back into the fray.

      Those guys use analysis methods that add a lot of noise. I am frankly baffled as to why they continue to make that point. The effects of gerrymandering are conspicuous simply by inspecting election results in NC or PA. Leave it to specialists within a field to take an obvious truth and make these kinds of arguments.

  • A New Jersey Farmer

    Remember that Nate started his career doing sabermetric reports on baseball. Many of the older baseball people still want to use their intuition instead of the data. As goes baseball, so goes politics.

  • Pat B

    I dropped by to see if you had commented on Nate’s move and, voila, you had. I do hope you’re around for the 2014 midterms. You may have to predict the dog vote because by then the country may be in their hands (paws.)

  • ottovbvs

    I much preferred your approach to Silver’s during the election. I’m counting on you being there in 2016. Most punditry is totally worthless (at least at the level it’s conducted today…Brooks is not exactly one of the great names from the 30’s/40’s and they often got it wrong despite dirty martinis with FDR.)

    • Amitabh Lath

      To top it off, the 2012 presidential election was a pretty easy call (PEC had it in May 2012). It just showcased how math-challenged the pundit class as a whole can be.

      I bet as a group few if any of them even has single-variable calculus under their belts. They really are wedded to the “it’s anyone’s ballgame!” storyline, damn the actual score.

    • Sam Wang

      Thank you. In my view, my approach is more defensible, more transparent, and more accurate. However, he made a major contribution: he wrote play-by-play in a way that people liked and wanted.

      I also agree about the general value of punditry. Amazingly, they get paid to do that. Basically, they are entertainers. It’s the distinction between news and analysis (me and FiveThirtyEight) and opinion (the bloviators).

  • Amitabh Lath

    The daily play-by-play creates an impression that every bit of statistical noise (new poll in OH! GDP moves half a point! Unemployment unchanged from last month!) is somehow important and deserves comment.

    Silver’s performance was measured as much by the amount of traffic driven to his website as by his accuracy. Maybe that’s why he included all these econometric parameters in his estimator. It gave him something to write about.

    The truth is that the electorate in aggregate changes opinions slowly except for shocks like the 47% comment or debate 1 (and I remain skeptical about those also).

  • E L

    I haven’t dropped by in quite a while. I think you’re spot on about what made Silver compelling. I know from my experience that a trial lawyer must tell the jury a “story” to be successful. Now, if we can just combine your statistic ability and good sense with JK Rowling’s story telling talent….

  • Mark Sillman

    Very very late comments:

    I think you hit the nail on the head with “storytelling”. I thought that in 2008. fivethirtyeight provided better reporting-from-the-field than anyone. He and friends’ reports on the Obama field campaigns in Colorado, in Pennsylvania still resonate with me 5 years later.

    But I also thought that Nate provided one statistical feature that I didn’t see anywhere else: an estimate of the real uncertainty for results on election day, even for a candidate with a good lead in the polls.

    He did this based on an amalgam of previous election results – including, for example, how Dukakis’ 13-point lead evaporated in 1988. As a result he had Obama just a 60% favorite in summer of 2008. Yes, Obama’s final poll numbers were not that different in June – but who could foresee, at that time, McCain’s surge in August (bringing him to 50-50 and beyond in the statistical nowcasts), the disaster of Sarah Palin and the meltdown of the economy?

    I’ve never seen an effective refutation, either, of Silver’s estimate of election uncertainties minths in advance.

    Uncertainty is one thing that mathematicians can attempt to provide – pundits never do.

    • Sam Wang

      Uncertainty is a good thing to report. However, Silver habitually overestimates true uncertainties. That could be a good thing for public consumption.

      As for a statement about uncertainties far in advance, perhaps see the draft of my paper. Also read my essays on true prediction on this site. Basically, Silver’s low certainty is equivalent to much larger movements in political races than have historically occurred.

    • Dan Kleinman

      I believe the guy who gave us those great reports from Obama/McCain field offices across the country was Al Giordano, not Nate. Al had one of the best blogs covering the 2008 elections, The Field.

Leave a Comment