Like thousands of others, I’m a huge fan of the things that Nate Silver is doing over at FiveThirtyEight.com — bringing serious statistics to the forefront of this year’s political analysis makes for much more interesting thought and perspective, and he’s certainly achieving that. (For good bio pieces on Silver, wander over to read this and this.) The sheer data he presents is like candy to me, and while Silver does a great job of finding the true gems in his statistical analyses and shedding light on them, I’ve found myself mining into his data to try to find other interesting tidbits.
Take the 10,000 election simulations that Silver runs every day, using the incredibly complicated model he’s developed to reflect historical outcomes, current polling data, and inter-state trends. Each day, he presents a chart with analyses of the scenarios that come out of those 10,000 simulations (it’s the chart on the right of the page, titled “Scenario Analysis”), and as interesting as each line item is, more fascinating to me is the information lurking just below the surface. For example, yesterday’s run resulted in 9,497 instances of Obama winning the popular vote, and 503 instances of McCain doing the same, both of which are interesting given how the race is shaping up. But below that, we learn that 199 simulation runs resulted in Obama losing the popular vote but winning the Electoral Vote (a la Bush in 2000), as opposed to 76 runs in which McCain pulled this off — numbers that by themselves aren’t that meaningful, but when combined with the winning-the-popular-vote numbers, show a stark reality.
How? Again, in 503 of the 10,000 simulation runs, or 5.03% of the time, Silver’s model predicts McCain winning the popular vote — but we now know that in 199 of those runs, Obama still wins the Electoral Vote (and thus the election), and in another 76, Obama wins the popular vote but McCain wins the election. So the reality is that McCain only wins in (503 – 199 + 76 =) 380 simulation runs, or 3.8% of them, a 1.2% decline from the apparent number. Conversely, Obama wins the popular vote in 9,497 simulation runs (94.97%), but he loses the Electoral Vote in 76 of those, and he wins the election while losing the popular vote in another 199; that translates into winning the election in (9,497 – 76 + 199 =) 9,620 simulation runs, which is an increase to 96.2%. That’s pretty amazing, and if Silver’s model is correct, it’s a testament to the well-known Obama campaign strategy of concentrating on electoral vote numbers as closely as it does those of the popular vote (a strategy that carries over from the campaign’s intricate understanding of delegate counts during the primary).
Similarly, take the “McCain loses OH/FL/PA, wins election” stat, which occurs in none of the 7,137 runs. Here, that numerator is certainly interesting, but for me the denominator is far more so. Trying to figure out why that number — 7,137 — is different than the 10,000 simulation runs we’ve been talking about up until now led me to the answer that sheds light on a really eye-popping fact: Silver’s model has McCain losing three of the biggest battleground states in 7,137 of its 10,000 runs, or 71.37% of the time. That’s fascinating, and a piece of information that’s not discernible from any of the more traditional polling numbers that you see on the web. (Of course, that’s not to say it’s necessarily correct, just that Silver’s model predicts that to be the likelihood of McCain losing Ohio, Florida, and Pennsylvania.)
If you’re not spending quality time on FiveThirtyEight.com every day, you’re really missing out on a chance to view this year’s election in an entirely different light. (And I haven’t even begun to laud the On the Road series of posts on 538, which are each just plain awesome.) Only time will tell how prescient (or not) Silver’s analysis is, but from the cheap seats, he certainly looks to know his shit.