Where We're At in the Democratic Primary

A few days ago, FiveThirtyEight published a very useful piece that essentially laid out the landscape for the 2016 Democratic primary. They used some polling data, fundraising, demographics, and Facebook popularity to figure out how we can expect each state to vote if Hillary is leading by twelve points, and alternately how we would expect them to vote if she's fallen into a tie with Bernie--i.e., the first projection shifted over by twelve points, plus or minus rounding errors. And what this lets us do is convert individual state results into an estimate of how the race stands nationally, both with actual results and with polls.

So, for example, they show Nevada as a pro-Hillary state, one that should go to her by about 3 points even if the race is tied. If in fact Nevada ends up as basically a tie tomorrow (and we have very little idea what will happen there; there have only been three polls, all low-quality ones, in recent weeks), that's a fantastic result for Bernie. Even if it comes out as Hillary +6%, the most pro-Hillary of the three recent polls, that's a very good result for him. Conversely, Hillary's got something like a 25-point lead in the South Carolina polling average, which would be even better than the +23% win they project for her with a national 12-point lead. It's to be expected that different states will give different signals, because these projections are by no means perfect. (It's like extrapolating election night results on the assumption of uniform swing: it's a good approximation but never quite holds.) Indeed it kind of looks like a decent guess for Nevada is the national-tie scenario, while a decent guess for South Carolina is the Clinton lead scenario.

One interesting thing, though, is that the two states that have voted so far both say the same thing about the race. Iowa was a tie, falling almost precisely between the two projections (+7% for Hillary if she's up big, +6% for Bernie if it's a tie). Meanwhile, though Sanders won big in New Hampshire, by 22%, he didn't win as big as he should have if the race is a tie, which would've been 29%. Both states, therefore, imply a Clinton lead of roughly 6 or 7 points nationwide. Which confirms my feeling that those results were decent for Clinton, even though they looked pretty lousy. To be sure it's remarkably that Sanders is within 6 or 7 points nationally, but that's still a pace that has Hillary winning. If that pace holds, we should see Hillary winning by nearly 10 points in Nevada and by 16 or 17 in South Carolina; more likely we'll see a similar average result between those two states but highly disparate individual signals.

In any case, it's just something to keep an eye on, as a guide to interpreting each state's results.

Winning the Small States Would Give Obama Options

One interesting feature of the FiveThirtyEight model is that it doesn't just calculate a point estimate for the popular-vote margin in each state, it also calculates a margin of error for that estimate. And this isn't simply some assumption based on how far in the distance the election is; rather, it's derived from the data available in that state, with I think the most important component being the amount of polling available in each state. The intuition is fairly simple: if a state has been polled to death, we have a better sense of the state of play there, and ought be more confident in our predictions regarding that state, than if there have only been a handful of polls. Part of what that means for 2012 is that Ohio, which has been polled to death, has a rather small error bar around its forecast, as can be observed from the "Competitive State Summary" in this article.

The fact that, in particular, Ohio has a much more confident forecast than Iowa or New Hampshire means that the order of the states by Obama's likelihood of winning them is not the same as Obama's order of states by Obama's expected margin in them. So if you draw up a path-of-least-resistance map based on the 538 odds, you add in all the 90%+ states, and then Wisconsin, and then Nevada, and you're at 253 electoral votes, and then Ohio and you're at 271 and Obama wins. But if you construct a path of least resistance map based on the 538 margins instead, after Nevada's 253 you next toss in Iowa and New Hampshire, getting to 263. Now, since the next state on your list is Ohio, which would've been the tipping-point state anyway by the other method, Iowa and New Hampshire are redundant, and so the path of least resistance is probably  basically the same by this method: you add on Ohio, get up to 281, and then ditch Iowa and New Hampshire as superfluous.

But that doesn't mean the fact that Obama's margins in Iowa and New Hampshire are larger than that in Ohio couldn't be relevant. Consider this map:

The dark blue states are those Obama leads by at least 3 points on the current FiveThirtyEight forecast, which total 253 electoral votes. The light blue states are, well, Iowa and New Hampshire, the two other states Obama leads by more than he leads Ohio. The red states are those Romney leads in, and the purple states are those Obama leads in but not by more than in Ohio. If you only give Obama the dark blue states, then he must add Ohio or Colorado and Virginia in order to be victorious. But if you tack on Iowa and New Hampshire, giving him a base of 263, then Ohio or Colorado or Virginia would do it. Ohio would get him to the highly-superfluous 281, Virginia would get him to 276, in which case he could jettison one of IA and NH, and Colorado would put him at 272.

Winning these small states, which for whatever it's worth happen to be the two most important primary states, in other words, gives Obama an extra degree of flexibility. They would allow him to win with his Midwest firewall, his Southern firewall, or his Western firewall, any one of the three. So while Iowa and New Hampshire are superfluous for purely path-of-least-resistance purposes, they matter a great deal in terms of the number of relatively easy paths to victory available to Obama.

Comparing Nate Silver and Ladbrokes

From time to time I like to look at the odds that inTrade, the big political gambling/prediction site, is giving various Republicans to win the nomination. But I think I actually prefer the odds at Ladbrokes, a major British betting site. They don't just have numbers on the nomination and general election: they've also got things like Iowa caucus odds. And, by a happy coincidence, FiveThirtyEight just released their first Iowa prediction, where the bottom line is percent chance to win. So what happens if we compare them? Well, for any given odds you can calculate an implicit percent chance to win, so if we do that with the Ladbrokes odds and compare them to 538's numbers we find that... Ladbrokes likes every single candidate more than Nate does. This, of course, is because the bookies like to give you shorter odds than you deserve, so that they make money in the end. Their implicit probabilities of winning Iowa add up to 124.8% between the eight candidates. But if we divide out by this number, and scale it down to a total of 100%, we find a fairly strong correlation between the two methodologies. Nate gives Newt a 49.6% chance of winning, compared to 28.2% for Ron Paul, 10.6% for Romney, 5.2% for Rick Perry, 4.1% for Bachmann, 1.6% for Santorum, and 0.7% for Huntsman. Ladbrokes, by contrast, gives Newt 46.4%, Paul 26.7%, Romney 11.4%, Perry 5.2%, Bachmann 4.7%, Santorum 3.2%, Huntsman 1.2%, and Gary Johnson 0.4% (he's not in the 538 forecast, I think). The big takeaway is that Nate likes the front-runners, Gingrich and Ron Paul, a lot better than Ladbrokes, giving them a combined 77.8% chance to win against 73.1% at Ladbrokes. Conversely, Ladbrokes is slightly overselling the odds of a Romney, Perry, or Bachmann win, and vastly overselling Santorum and Huntsman (giving them 5% combined instead of 2.3%; that's more than double!). Do not bet on Rick Santorum to win Iowa on Ladbrokes; they're giving you awful odds.

Of course, you shouldn't actually bet on Newt or Ron Paul either, because Nate only likes them better than Ladbrokes' oddsmakers once you scale down to a sum of 100%. Overall, though, it's a pretty impressive alignment: the correlation between Nate's odds and the Ladbrokes odds is around 99.9%. (This is impressive for Ladbrokes, by the way, since we assume Nate knows what he's talking about in advance.)