1:58 pm – This time I’m early. Keynote interview with Nate Silver of FiveThirtyEight.
2:04 – Baker’s cheatsheet here. The Numerati (his book) is about statisticians storming into different industries.
2:04 – Silver says 538 was used to procrastinate from other works; borne of frustration at the news media
2:05 – Polls were too much a part of the narrative; he is not a big fan of polls, which are overinterpreted, room for “common sensical expertise” to interpret polls in a way that is more intelligent and meaningful
2:06 – Obama had trouble with working class voters in, let’s say, PA, but not in Oregon or Minnesota which are also working class states.
2:07 – The momentum in the primary dictated by the calendar; states Hillary was bound to do well in voted late; early states were tilted towards Obama. Obama’s primary win partly due to the randomness of the primary calendar.
2:08 – People too quick to assume things are attributable to racial voting patterns. Look at Hispanic vote for Hillary then for Obama.
2:09 – Baker: Demographics is dead; categorize into “behavioral tribes.”
2:10 – Silver: Look at voters at individuals not groups; there are an infinite number of typologies with gradations of like-ness; says Penn’s microtrends is basically BS (YES!)
2:11 – The Silver baseball algorithm is TOTALLY different than his political model. The key to success in both is “good habits.” Being really meticulous. If you really want to solve a problem… go beyond the 80/20 rule; actually push into the hardest 20% and talk about it in a very interesting way. Decisions are made on the margins and simple 80%/20% isn’t enough. Being able to forecast 2% better can save $3-4 million on a $50 million payroll. Detail matters.
2:15 – Baker: did you come upon a “5% difference” that changed your outlook on a state? Silver says in Appalachia people say their ancestry is “American,” a proxy for “redneckness.” This makes a poor voter in Kentucky different than a poor voter in Wisconsin. (“American” voters went heavily for Clinton and McCain more so than national trends.)
2:17 – Baker: “What is Manny Ramirez worth this year?” Silver: He’s a $12-15 million a player and is getting paid twice that. People underestimate how fast players start declining; when a player is in his 30s much more likely to get injured (and lose their will). People assume things won’t change as much as they do.
2:22 – Wifi down. Baker asks about economy; when will people blame Obama. People don’t blame Obama now but in 12-18 months they might. The thing Obama has going for him is non-experts believe the economy will be worse than most economists do. So room to outperform expectations. However, recent recoveries have been jobless recoveries and if the current one follows this pattern Obama could be hurt.
2:23 – A recency problem. Most economists only use the last 10-20 years for modeling but don’t go back further than that to explain current trends.
2:24 – Internet bubble, deleveraging of banks *should* be that predictable; none of us should be that shocked that there were these problems. A few people who did predict this including Schiller and Rubini.
2:25 – Sean Quinn was his correspondent in the field; used to be a field organizer for Jon Tester & Brian Schweitzer. Not enough perspective on what is happening on the ground (organization, field work, data mining, etc.) We’re mostly about the numbers but Quinn’s reporting was useful in differentiating themselves. “Oftentimes it’s where the candidate is NOT that’s important. Obama campaign was working 24 hour shifts in Tallahassee and McCain campaign went home.
2:27 – Quinn was from SF. Picked up a photographer and chronicled the swing states. He was never able to work this into his statistical analysis. One side was anecdotal, the other was quantitative.
2:28 – On the morning of the election, we predicted 98.9% chance of Obama victory. If the 1.1% result came true, were we wrong or unlucky? Likens it to baseball. “What if the player finds Jesus?” Totally random, unexpected things can happen for personal reasons.
2:29 – Q: will you get into human resources? Silver: No. Statistical analysis can’t replace bosses.
2:31 – Q: What would you get a masters degree in that you don’t already? Silver: Computer Science so I wouldn’t have to hire a programmer and could tweak Blogger templates. (Feel your pain!)
2:32 – Q: Would you put your genome up on the Internet? A: Probably not. Goes into privacy. He has a compulsive personality and if he were on Facebook, you’d probably never see him again.
2:33 – The question he faces now if he should take VC financing or keep bootstrapping? (for FiveThirtyEight — a blog!) Always a hard decision. But would be nice to bootstrap.
2:34 – Will he branch out into other areas? He is doing more stuff on the economy. Being asked to predict the Oscars. Did okay, but not that great. Looked at 30 years of Oscar history — based on winners of other awards (Golden Globes, etc.) Those are usually pretty reliable. Academy tends not to like comedy or edgy pictures — likes “big pictures,” epics. Got Best Actor wrong — predicted Rourke over Penn. Said 20% chance of Penn victory, so maybe just unlucky. Said personal factors could have worked against Rourke, who was kind of a jerk to his peers. Also the Prop 8 factor.
2:37 – In Best Supporting Actress, he and everyone else thought Cruz would win not his model. “Never come out with a forecast you know to be wrong. If you know you’ll be wrong, keep working on your f*ing model.” LOL.
2:38 – Now going to talk to people in different fields, everything from fashion design (which can be scientific) to if an asteroid will hit Earth.
2:39 – Going to get into national security? Statistical probability of someone being a terrorist? A: The thing about baseball is you have a perfect data set; the real world is not like that. Not like looking at a needle in a haystack, but looking for a needle in a bunch of needles.
2:41 – Thinks it’s impressive we have not had a major attack since 9/11.
2:43 – Baseball more predictable than basketball, football b/c it’s less self-centered and everyone takes their turn.
2:44 – Audience Q&A. How did you solve the data problem? Polls not as reliable as baseball data. Silver: Baseball collection not that difficult to do. Fundraising primary was a predictor. Obama collected twice as much money as Hillary in Colorado, predictor of caucus vote.
2:46 – Q: What does 538 stand for (duh). Silver says # of votes in EC but misstates that DC would get an extra vote and it would go to 539. Actually Utah would get the extra vote. What do you read? Reads a lot of books halfway through.
2:49 – What would happen if McCain won? A: smaller stimulus package; more interesting is what if Hillary had won? Obama was inexperienced; maybe having someone in the WH before might have helped (Hillary) Do you think the Netflix challenge is interesting? Doesn’t get to this question, but it’s a good one.
2:50 – Have you read Richard Fenno’s work on Congress?
2:52 – Doesn’t business also have perfect data? Disagrees that business has good data; companies cook the books and report their own results. If he knew, he’d work in hedge funds. Even gov’t economic statistics are imprecise. Improve the data quality before applying better predictive modeling.
2:54 – Should the EC be abolished? Silver takes the 5th. It’s good for his website. LOL.
2:55 – Silver thinks the stock market has behaved irrationally despite the fact that it’s a top prediction market.
2:56 – Prediction markets better for information that is hard to quantify like the Oscars. For politics, statistical analysis and prediction markets fairly well matched.
2:57 – Irish dude says he runs a prediction market (InTrade?) Uses the Hillary NH example. Blame data when things go wrong, when things go right, you were right.
3:00 – Pollsters getting more responsible about including crosstabs. Amen.