AP and NBC yesterday belatedly called North Carolina for Obama, making the state’s result the second last to come in. Only Missouri hadn’t been called yet. So how did it become so close? Facing South has a good summary up of the main strategical and political reasons. But I would like to look more specifically at the geography and demographics of the race.

For Obama to win the state required a 12.4% swing (that being the margin by which Bush was elected in 2004). He got a 12.6% swing. Which parts of the state pushed Obama over the line? Where did his efforts of persuasion fall short? What demographics were at play? An in-depth look.

North Carolina was expected to be close in these elections throughout the summer and fall. By late May, Obama had already narrowed the gap down to 5%; by August it was 3%. In late September, polling aggregator Pollster.com first had Obama in the lead.

In the last two weeks of the campaign, however, McCain seemed to be coming from behind. In fact, of the 20 states that were even remotely on the radar in the last two weeks of the campaign, only six moved toward McCain in the last two weeks of the race, and four of those were North Carolina, Virginia, West-Virginia and Pennsylvania – possibly marking a McCain push in the Appalachians. (A fifth was Florida, so you could also speculate that McCain was rebounding a bit in the South).

By election eve, polling for the race was tight as a tick: Obama 48.8%, McCain 48.4%. (Data from pollster.com: 10/20, 10/27, 11/4).

North Carolina being a race at all, nevertheless, is a testament to how successful the 50-state strategy Howard Dean pioneered has been. I mean, North Carolina! No Democratic presidential candidate won here in 32 years. Not just did Kerry lose this state by over 12%, so did Al Gore – who is from neighbouring Tennessee. Bill Clinton, another Southerner, lost this state twice – to Bob Dole by almost 5%. Dukakis and Mondale were trounced, by 16% and 24% respectively. Sort the states Bush won in 2004 by vulnerability, and ten red states were closer than North Carolina (seven of which Obama won). I mean, North Carolina!

Of course the demography of the state has changed significantly in the last decade or two. Even just in the four years between 2002 and 2006, the population grew by 10%. Much of that influx is non-white: while the state’s population grew by 793 thousand, the white population grew by just 455 thousand. The Hispanic population grew by 163 thousand, or 38%. Four per cent of North Carolina residents in 2006 had only just moved into the state in the last year.

You can glean other demographical changes from the American Community Survey that would benefit Democrats as well. For example, the share of those employed in management and professional occupations or in service occupations went up from 45% to 49% in four years. In a state where the white working class trends Republican, this is good news for Democrats. The unmarried share of the adult population (minus widow/ers) went up from 34% to 39%, which is relevant because this is one of the strongest Democratic-leaning demographics – nationally they went for Obama by about 2:1.

However, when the swing to the Democratic presidential candidate is about one and a half times the national swing, demographic change can only explain so much. That goes for the regional variation in how proportions changed too. In Cumberland county (Fayetteville) there was a 21% swing to Obama; in nearby Columbus county there was a 6% swing to McCain! Demographics are probably at play here, but maybe not just in terms of new residents coming in.

Chart 1: The swing by county (click to enlarge)

When reporters write about the changing North Carolina, the first place they always seem to point to is the Research Triangle, the heart of the new economy in the state. And it is true that the Triangle is easily the most Democratic-friendly region of the state. Even John Kerry won here, if by a narrow 1.2%; Obama swept home with a 15% margin of victory. But it was not the part of the state where the swing in his direction was largest. The swings in Metrolina (Charlotte) and the Piedmont Triad (Greensboro) were larger still.

You have to click on the colourful-looking chart above (or here) to see it in full size and find out it’s not actually a nerdy artist’s impression of birthday party balloons. It shows the swing per county: how did the swing in each county measure up to the state-wide 12.5% swing Obama needed to win?

Since not all counties are created equal, the size of the bubbles reflect the number of voters in the county. (All data was taken from the county-by-county election results listing on the North Carolina State Board of Elections website on 6 November, once all counties had been ticked off as having fully reported).

Finally, there’s the small matter of this insane tradition you Americans seem to have of preserving county names that appear to bear no relation to any well-known place inside the county, and for added anonymity seem to be used in pretty much every state. So to help identifying which is what, I’ve coloured the bubbles by region.

The first thing that struck me is that the bubbles appear in roughly reverse logic to party balloons. Biggest ones on top, smallest ones at the bottom. What that means is that basically, the swing to Obama was largest in metropolitan and urban areas, and smallest in rural counties. Not just did Obama get some of his highest shares of the vote in urban agglomerations – which was only to be expected; he overwhelmingly gained the most ground in comparison with Kerry in 2004 there too. His biggest gains were in Cumberland (Fayetteville), Mecklenburg (Charlotte) and Forsyth (Winston-Salem) counties, followed by Onslow (Jacksonville), Guilford (Greensboro, High Point) and Wake (Raleigh). The consequence is that the political divide in the state between urban and rural areas must have increased.

The second thing you automatically look for is how the different regions compare. Here the pattern is much less clear; at first glance the colours seem randomly dispersed. But this is not quite true.

Chart 2: The swing by region (click to enlarge)

For example, eight of the nine counties where Obama actually fared worse than Kerry were either in the Mountains region in the west of the state, or in the Southeast. On the other hand, there was a swing of at least 2.5% to Obama in every county in Metrolina, the Piedmont Triad, the Research Triangle and the Central-Eastern region. Those are the most urbanised parts of the state, but even of the rural and small-town counties in the Piedmont Triad and Research Triangle, all but one showed swings of 5+%.

When you group the results together by region as in Chart 2, the differences simultaneously become smaller and a little more clear. The urbanised central region of the state (Research Triangle, Piedmont Triad, Metrolina) basically all showed roughly the same swing of 14-15%. Note: their election results are different – while Obama swept the Triangle, he still ended up narrowly behind in the other two regions. But the size of the swing to him was similar in all three.

It’s hard to overestimate the mobilisation of new voters at work here. In each of these regions, McCain actually received more votes than Bush had gotten in 2004, but Obama’s new voters swamped them. In Metrolina McCain got 10% more votes than Bush, but Obama got almost one and a half times as many as Kerry, increasing the Democratic vote from 320 thousand to 470 thousand. He got 1.4 times as many votes as Kerry in the Research Triangle, and almost that much in the Piedmont Triad. (Facing South points to instant electoral research showing that North Carolina showed “the greatest increase in overall turnout”, measured against the total eligible voting population, of the country.)

The reason the result was nevertheless such a squeaker is because he underperformed, comparatively, in the mostly rural east of the state and in the mountainous west. On average, for sure, he did much better than Kerry there as well; the swing just fell short of the needed 12.5%.

The reasons for that seem to vary. In the northeast and to some extent the southeast, the problem seems to largely lie in racially polarised rural communities. Especially the northeast has a large African-American population, and encompasses some counties that are hardcore Democratic. In majority-black Warren, Northampton, Hertford, Bertie and Edgecombe counties, even Kerry already got over 60% of the vote (see the NYT-derived “county leaders” map below). It’s hard to add another 10+% of the total vote if you already have that much.

I also say “racially polarised”, however, because the voter mobilisation seems to have worked both ways here. A few days before the elections, Democratic strategist Ed Kilgore worried about “signs of a racial backlash developing – against Obama [..] but also against the heavy early voting turnout of African-Americans.” Is that what was at work here? In the North- and Southeast of the state, McCain actually increased the Republican vote count by 12-15%, presumably all from white voters. In the Mountains out west, in comparison, where few blacks live, McCain increased Bush’s vote count by just 6%.

N.B.: Compare the WaPo map, which shows the winner's margin in number of votes, with the NYT one below, which shows it in percentages (you can click this one to open it in a new window).

In any case, the increased Republican turnout went some way to blunt some of the massively successful ground game of the Obama campaign, which increased the Democratic vote in the Northeast by 33% and in the Southeast by 39%.

It was in the Mountains that the swing to Obama fell short most, though. Even there, there was an overall 7.6% swing to the Democratic presidential candidate, but that would have been far too little to win the state. If the Mountains voters had swung to Obama by as much as their more urbanised fellow North Carolinans to the east, it wouldn’t have taken an extra day to find out who won the state.

That’s not to say that Obama got nowhere up in the Appalachians. It wasn’t just in hippie Buncombe county (Asheville) that there was a healthy 10%+ swing to him (Buncombe in fact underperformed a little compared to counties with similarly sized populations elsewhere). There was one in Watauga county (Boone) as well, where a 6-point Bush lead turned into a 4-point Obama win. Up in Madison (Mars Hill) and Jackson (Sylva) counties, as well as in Transylvania (Brevard) and Henderson (Hendersonville) counties, the swing wasn’t quite the target 12.5%, but still a good 8-9%.

That said, some of the rare failures of Obama to gain any ground came from the western-most tip of the state. Graham, Clay and Cherokee counties, which already voted for Bush by about 2:1, all actually swung further to McCain. There’s a partial parallel here with the county on the western tip of Virginia, which swung to McCain and “voted Republican for the first time since Reconstruction”. It certainly fits in the larger picture shown in the excellent interactive map the NYT has online. Click “voting shifts” to see how the vote by county has changed, across the US, in comparison with 2004, 2000, 1996 and 1992. The Appalachians and Ozarks stand out as the regions where McCain actually gained ground.

Other notable failures of Camp Obama to win over any souls came from the Southeast, where the voters of both Richmond (Rockingham) and Columbus (Whiteville) counties swung somewhat to McCain. That’s interesting for a non-native, non-resident like me, because both counties were about evenly split four years ago, and with 20-25,000 voters each they aren’t entirely out in the sticks either. What’s up with that?

A last variable at work may be partisan as well as racial balance. Of the counties with a larger population (over 60,000 voters), the ones that swung most strongly to Obama were Mecklenburg, Guildford, Forsyth, Wake, Cumberland and Onslow. Aside from heavily Republican Onslow, all these counties had been roughly split between Bush and Kerry in ’04 (with Kerry getting 46-52%). They also all have significant, but not majority, black populations (18-34%). The ones where the swing to Obama was less than average were Catawba, Gaston, Davidson, New Hanover and Orange. All but one are heavily partisan – Catawba, Gaston and Davidson went for Bush in ’04 by more than 2:1, while Orange went for Kerry by 2:1. They all also have a relatively small black population (8-17%).

So you could say that while there was relatively little extra ground to win for Obama in already heavily Democratic counties and the deep red counties generally held relatively firm for McCain, Obama really slaughtered McCain in the toss-ups. (The one exception is Onslow county, which has just 18% blacks and went for Bush by 2:1 and is thus very similar to Catawba, Gaston and Davidson – but where Obama nevertheless made impressive headway).

I might have looked at the data too long, though, and started to overstate these things. After all, in Orange county Obama’s percentage of the vote was still 5% up on Kerry’s, while in Onslow it was up 9%. So these things are very relative. The bottom line remains that Obama supercharged Kerry’s results across North Carolina, and took a state that nobody would have thought a black Democrat would be able to take even just a year ago.