nate silver interview

In the alternate room…did we just lose the sound? no, it’s back.

538 was originally procrastination? plus frustration. “esp fox news”

“not a big fan of polls” bcause over-interpreted.

what were early polls/coverage missing? oversimplification, reductive. from a data geek’s perspective best natural experiment.

race? it’s complicated, can’t treat any race as monolithic: hillary’s success with latinos, but obama in the final still got overwhelming support from hispanics.

in advertising, they say demographics is dead; spotlight analysis, dividing into tribes based on “life perspectives” (that got a lot of mention in The Big Sort) voters as individuals, falling into clumps. some explanation of baseball stats experience, no “typologies” short-sighted (missed name, hillary’s campaign guy?) not seeing forest for trees. most americans don’t like to think abt politics issue by issue? american public underestimated in its sophistication. interesting idea for qualitative project: what really motivated you, esp people who went against type.

had always been thinking about politics in the background. 80/20 is the rational thing to do if you’re a normal person, but he’s all about the crazy meticulous detail. again relating to baseball; differences in the margin v important, millions of dollars, etc. “puzzle comes together a little bit at a time” over a long season, baseball or election. “you kept me kinda calm during the election” (which was, yes, my experience.)

appalachian, ancestry discussion “american” as equivalence for “redneck” defining difference between poor white in kentucky vs. (missed other state, am subbing in oregon)

how much is (baseball guy) worth this year? blah blah blah 🙂 more blah blah blah re:baseball.

economic crash & voters? haven’t had this situation in the modern era. fearful, 9/11-type even in terms of pessimism but also high approval of prez. at some point will be an intersection. better get moving! what’s the grace period before he starts getting blame…august 2010? right before midterms! but public is even more pessimistic than economists, so

polling zip codes with high foreclosure rates? some states might not be gaining reps in the next census. lowest migration in non-wartime. unemployment is key variable for the public. “jobless recoveries” gdp could be positive, but could still get blame bcause of unemployment. (which seems fair to me)

missed part of the q. how do you compare ’48 or ’32 to now with all the changes? ref to Black Swan. have to be aware of shocks that aren’t in yr datasets. bubbles seem like they should’ve been obvious. none should be that shocked, a few economists, but otherwise “worries in the back of your head” “making up explanations when we see random noise”

missed question again. guy writing about experience on a campaign, fieldwork/organization stuff. (and that was absolutely fascinating on 538) where candidate is not being more important. who’s working, who went home. (I remember seeing those, loved ’em.) reporter was breaking up w/girlfriend, driving from SF to St louis, then went all over the country w/photog. oh, so that’s what the story was. some things you just have uncertainty in a range and then the qualitive to see whether you should look at the top or bottom of range. (I think i mangled that paraphrase.) chance & luck & personal crap (in baseball), human factors.

ibm analysis of workers, like his baseball stuff. are you going to go into HR? no, kinda sckeptical. how you evaluate job performance. gates (foundation?) and eval of teachers. suspicious that poorly executed objective eval is worse that well executed subjective, counter-productive in the long term.

computer science, wouldn’t have to hire a programmer. economics? (his BA) in a lot of fields, master’s doesn’t help. if i knew how to write some of my own code, a lot of improvising, a lot of time with bad tweaking of blog templates.

would you put genome on the internet? (MOST RANDOM QUESTION EVAR) surprised at how mcuh people are willing to share every detail of their life now. kinda in email generation (is 31), avg 15-yr old girl sends 2500 txts/mo. if i actually spent time on fb, would never see light of day bcause obsessive.

what’s yr strategy going forward, given being most notable political (???)? trying to work out funding, would be nice to have more infrastructure. him & shawn and some pro bono programmers. would be nice to have actual designer, better presentation of data. (veen tackling this would be rad) tried some oscar predictions, ny mag making the generalization “you can predict anything” looking at variables over 30 years. like a poll with a very high overlap. (not really paying attn, oscars for me is almost as blah blah blah as baseball.) computer confusion of designing model, “keep working on your fucking model”

tell us abt yr book. on forecasting & modeling in general, going to talk to people in different fields. including fashion design “quasi-scientific” extraterrestrial, with only one data point. at the stage where he has an outline, has’t written anything “will be interesting, I hope”

statistical probability of being a terrorist? baseball is easy because it’s been recorded 99% accurately, and no cost in being wrong (or low cost), real world not so much, like looking for one needle in a pile of needles.

epidemiology, including crime, drug use, will probably put terrorism in that group for writing purposes. (oh, that slate series on reasons why no attacks)

twitter question: predicting whether flight will be later. jet blue punished because of flying out of jfk; relative to airports you fly out of, context importance. another good dataset?

huh? oh, another baseball question. issues of motivation in re: structures of sports.

audience question: how did you face challenge of collecting data thruout country? not terribly complicated, steal off other polling sites, then got results directly from polling co’s. problem of evaluating whether a poll is real. also lots of data at us census site, but with clunky interface. “fundraising primary” data publically avail.

q: what do you read? how much time? why 538? number of electoral votes, could be 539. reading a lot about the markets irrational exuberance. nixonland. (that was FABULOUS. read it mostly laying on the beach on the river last summer.) read a lot of books halfway thru, but that’s ok.

q: dealing with bad data? try to balance with regression based on demographic data. at a point where you have enuf good data that bad data doesn’t add anything at all. 7 or 8 national tracking polls, dozens of state polls, everyday.

q: if mccain had won? netflix challenge? smaller stimulus, more tax cuts. more interesting if hillary had won. email doesn’t work in the white house, and they’re trying to solve the economy!

q: have you read richard finnow’s book? how to look at unquantifiables? look at dems in repub states & vice versa; “overperforming” like sibelius in kansas

q: what part of business community could most profit from his type of analysis? if he knew, wd be working for a hedge fund or something. data not that grate akshully, because co’s not entirely honest. the way inflation is calculated, basket out of date; unemployement, most basic metrics not very precise. deal with that before getting “cute.” interviewer: one of the challenges of media is figuring out how to do what nate is doing. who is not interested in doing econometrics unless he took a year off to actually learn it. baseball & political polling in a sweet spot, solvable, complex but not THAT complex.

q: prediction markets? is the electoral college an abomination? not a huge fan of prediction markets. electoral college is good for his website. 🙂 (missed a bit looking at a cool graphic on someone else’s screen)

q: missed it. somebody who runs a prediction market? frustration with journalists. “blame data when things go wrong” referencing bradley effect.

q: basketball data, not recording all the good stats. how does his analysis improve data capture? creating a market for the application of all that stuff. sports data geekery.

q: thoughts on impact of young political bloggers? have to have some impact, just because of traffic. make people more accountable. problem of groupthink. mccain folks surprised at how negativity didn’t work, thinks one reason is pushback of blogs.