Progress Illinois

Several weeks ago, I published an article here at Progress Illinois that used a regression model developed by the anonymous blogger "Poblano" to examine how increased youth and minority turnout could affect Barack Obama's chances in the general election this November.

Poblano had launched the website FiveThirtyEight about two months earlier and had quickly begun making a "name" for himself among various election watchers. Many were impressed with the depth of the data presented on the site and the complexity of his analysis. But when he used his model to predict the results of the Indiana and North Carolina primaries, pollsters and analysts really began to take notice.

Last Friday, in a brief blog post titled "No, I'm Not Chuck Todd," Poblano unmasked himself as Nate Silver, a 30-year-old Chicagoan and managing editor of Baseball Prospectus.

What's fascinating about Silver is that FiveThirtyEight is not his first renowned statistical creation. He is also well-known in baseball circles for having developed PECOTA (Player Empirical Comparison and Optimization Test Algorithm), a projection model that does a remarkable job predicting how specific teams and players will perform in a given season. As The Chicago Reader's Michael Miner recounted earlier this year, "Silver’s system named six of the eight [2007] playoff teams, and it had a seventh, Cleveland, finishing a game out of first. PECOTA called the NL Central for the Cubs with an 86-76 record. They won it at 85-77."

Now that he's out from under his pseudonym, we're going to be seeing a lot more of Silver's commentary. For instance, he has a column today in The New York Post (and not in the sports pages). He also plans to contribute ongoing election analysis to The Guardian. I interviewed him yesterday to get a sense of why he decided to reveal his identity, how he finds the time to manage these two statistical juggernauts, and what happens to FiveThirtyEight on November 3.

JK: How did the idea for FiveThirtyEight strike?

NS: It was borne partly out of frustration with the alternatives out there such as RealClearPolitics. They have the averages in each state, but they treat all polls equally, compared to what we do: look at which polls are more reliable and weight those more heavily. And the fact that you don't have a lot of polls in certain states, so you need something like our demographic model to help compliment and balance out those results. Being able to look at things more rigorously and more dynamically was really the motivation, because it seemed like there were different kind of pieces of it elsewhere on the web, but no one was doing it in quite the right way. So that was kind of the rationale: If no one else is going to do this themselves, then I'm going to have to build it on my own.

JK: When did you start building the model?

NS: I launched it in March and began some preliminary work on it back at the end of February. It was probably a couple weeks worth of work, maybe a couple of hours at a time.

JK: Is it fairly low-maintenance work at this point? Or are you constantly tweaking and calibrating the model?

NS: When I first debuted it, it was kind of in beta test mode, so I was kind of playing around with a few things. But I wanted to get that part done early, because you can't change your methodology in the middle of the election. Until about a month ago, I was still fiddling with it, but now it's pretty much as is. You search for new polls. You type those in, press the button, go have a cup of coffee -- it takes about five minutes to calculate -- and then you have to cut-and-paste a bunch of tables. But that's only about a 10 or 15-minute process every day.

JK: There are folks out there who have commented that, while FiveThirtyEight is fascinating to look at and play with, the numbers don't really mean anything because we're so far away from Election Day. How do you respond to that?

NS: Well, we account for that kind of uncertainty. Take Hillary Clinton's projection right now. She's winning 300-and-some electoral votes. But we don't assign her a 100 percent probability of winning. Or even 80 percent. It's still more like 60 percent because the model knows that the longer you have to go until the election, the more uncertainty there is around those numbers. So a state where Obama might be down by 10 points or something right now, it still gives him a 10 percent chance or so of winning that state, especially if it doesn't have a lot of polls, because we don't really know what's going to happen between now and then.

But as you get closer to Election Day, that margin of error will get smaller and smaller and so you'll see those states that are pink or light blue will start to become solid red and solid blue. And then there will be the states where you have really close results, which could still be half a dozen or more.

The fact there is uncertainty this far out is all kind of built into the model, in other words.

JK: When you're talking about the site, you often say "we." Is it just you or has anyone else been helping you out?

NS: I have a friend who helped me out with the design. I have one other blogger coming on soon. But almost all of it's been me. I'm trying to expand the team a little bit, just so I don't exhaust myself.

JK: Why did you originally start to work under the pseudonym Poblano?

NS: There was no good reason except that I'd done some blogging at DailyKos under that name so it was how people knew me. When I signed up for my DailyKos account it was kind of accidental. My friend had used it in another forum and I thought it was a cool name, so I kind of hijacked it from him. But I never expected it to build up to more than a handle. And at first, it is kind of fun to do it anonymously because people are guessing who you are.

JK: But then you revealed your identity last Friday.

NS: Well, eventually you want to get your name out there in the public discourse and change the way people are maybe looking at these things. It helps to have your real name associated with that. If I hadn't done so I would have been sacrificing a lot of opportunities to help promote the site. For example, the fact that I can talk to you and have a little more credibility in that way.

JK: What exactly is your position at Baseball Prospectus?

NS: There are really almost two or three different things. My claim to fame is that I created projection system called PECOTA, which determines how every baseball player is going to do next year -- you know, Derrek Lee is going to hit .310 with 30 home runs and that kind of thing. There are a lot of people who do that. But the way we do it is looking at a player like Lee and finding other players who are similar historically. It's proven to be the most accurate system of its kind on the market. It has a pretty big following -- fans and baseball players really like it. And folks in the front office actually use it and give us all kinds of questions about it. So it's a good brand name in that particular kind of niche market.

I also do writing for Baseball Prospectus on a weekly basis, where I try to take on a contemporary baseball topic. So I'm used to writing on a pretty quick turnaround, which is a very useful skill when you're blogging. Especially with election news, which tends to break a lot faster than baseball stuff.

And the last thing I do is to help manage the company. We're a partnership and I have the title of managing partner. So there's a lot of sharing responsibilities. Baseball Prospectus is a pretty good brand and I've been involved in trying to help manage that brand, which is another thing that gives me a sense of what works in terms of internet content and what doesn't.

JK: What are some examples?

NS: I like to keep things very visual. I also know that if you have content that comes up three or four times a day, it's better than having a bunch come up all at once, because you get behavior on the part of the customer where they're checking the website constantly and don't know when stuff is going to go up. I keep a kind of weird schedule, so I update things sometimes at eight in the morning and sometimes at three in the morning.

JK: On the statistical side, how do PECOTA and the FiveThirtyEight model compare in terms of complexity?

NS: The baseball stuff is more complex, I'd say, by a factor of three. But it's because I've been doing it since the original version back in '02 or '03, so it's been constantly tweaked and revised for five or six years now.

There are some things you can actually cross-apply. For example, when I took Michigan and looked at a congressional district there and then looked at other congressional districts in Pennsylvania or Ohio that are similar to it -- that is kind of a version of what I did with PECOTA, where I looked for similar players in terms of their power and their height and weight and everything else.

JK: As you juggle these two projects, what does your average day look like?

NS: The last month or so, it's always been a fire drill. You kind of figure out what's more pressing -- whether there's a news article I want to respond to or whether I've got an article due on the baseball side or whether we have an urgent business matter. I wish it were more regular. But I'm always trying to do twenty things at once.

JK: Has building traffic for FiveThirtyEight been a gradual process? Were there any big milestones that helped build your audience?

NS: The North Carolina primary seemed to be pretty big. We put out a projection there that seemed very bold at the time. We had Obama winning by 17 -- he actually won by 15. So we were close enough. And Indiana we saw as being really close -- basically a toss-up. In that case, we were actually more accurate than the polls were and that kind of got a lot of attention.

JK: You seem to have a pretty healthy community of commenters on the site.

NS: I've been encouraged by the commenters. Sometimes on political sites the comments can go kind of haywire. I'd actually like to get more people from Republican-leaning blogs, but it's mostly been people from the left -- probably three-to-one or so. I'd like to kind of get more of a following of people who maybe I don't agree with personally, but are intelligent voters.

JK: You said in your blog post on Friday that you've developed some relationships with bigger outlets. What are the details and how did these arrangements come about?

NS: It was just people asking if I'd like to write for them. I'm going to be writing semi-regularly for The Guardian. And the rest I'm still trying to sort out right now. I kind of need to figure out how much capacity I have. But I think, at least with The Guardian, which is an outlet that's really growing a lot -- its U.S. traffic is similar to the Tribune's -- that's an outlet I'm very pleased to be working with.

JK: And finally, what's going to happen to FiveThirtyEight after the election is over?

NS: Well, hopefully there'll be a recount. [Laughs]

I think, particularly if Obama wins, people are going to stay relatively tuned in to politics. So many people are going to want to see what he does, especially with the fact that he will probably have an overwhelmingly Democratic Congress. There's going to be a lot of attention paid to it.

There are other things we can try -- perhaps creating power ratings of senators. We could use different metrics to see: if some senator co-sponsors legislation does it increase its chance of being passed? Stuff like that. And there are issues like gay marriage, which is an initiative in California. We could look at whether it's really true that it's a landmine for Democrats electorally.

So there are other issues where you can look at polling data and look at demographics and other metrics and try to approach things that way. I think there will still be room for intelligent, somewhat quantitative, political commentary.