Across North Carolina, Organizers and Volunteers are taking a well deserved breather.
Just kidding! One Stop Early Vote GOTV is over, but Election Day GOTV has already begun.
No rest for the weary until November 7th.
2 days and one evening of GOTV are left.
If you are wondering why GOTV is important, you just have to watch this video:
Saturday Voting:
There was a last minute boom of Democratic turnout on Saturday, the last day of One-Stop Early Voting. In total, 201,681 people cast their ballots Saturday: an estimated 112,669 of them voted for President Obama, while an estimated 89,012 voted for Romney, giving Obama an estimated 23,656 vote margin on the final day of early voting. That's 517 votes better than Obama's estimated 23,140 vote margin from the last day of early voting in 2008. As always, turnout may increase as counties process more data.
It's all about GOTV now.
Total NC Early Vote Turnout:
With practically all (but not all) of the early votes in, Early Voting Turnout in North Carolina has easily broken the records of 2008.
The Charlotte Observer has a great early voting summary picture, which is quite simply too good looking not to post:
Total turnout is sitting pretty at 2,742,100, whereas in 2008 it was 2,616,779. The net increase in turnout is 125,321 votes, or an increase of 4.8%.
Click on the picture below for a full sized chart.
These numbers do not just include "accepted" ballots, but also include ballots which were "spoiled" and have other statuses. As a result, the total turnout number above is a bit higher than the 2,706,836 "accepted" votes reported by the North Carolina Board of Elections (PDF File!). But those other 35,264 votes are floating out there in the ether somewhere, and whether or not they get counted could be up to the lawyers and judges in any potential recount.
Overall, an estimated 1,455,277 votes have been cast for President Obama and an estimated 1,286,823 votes have been cast for Mitt Romney, giving Obama an estimated 168,455 vote margin, or a 53.1%-46.9% lead, with roughly 59.6% of the vote in (assuming a final turnout of 4,601,461 "accepted" ballots).
In addition, the final numbers of ACTUAL votes cast for both Obama and Romney are likely to be lower for two other reasons:
1) Undervotes - First of all, there will be some undervotes. In 2008, the NC Board of Elections Reported a "total turnout" of 4,353,739 votes. But there were only 4,310,789 votes cast in the Presidential race, meaning that for whether by mistake or on purpose, 42,950 people who voted left the Presidential race blank on their ballots. We can probably expect a similar number of undervotes this year.
2) Third Party and Write-In Votes - In 2008, Libertarian presidential candidate Bob Barr was on the ballot and received 25,722 votes. Again in 2012, Libertarian presidential candidate Gary Johnson is on the ballot. Libertarian party registration is up, so perhaps Johnson will do a bit better than Barr, but overall we should probably expect a similar number of votes for Johnson in 2012. There will also be some number of write-in ballots once again. For example, 1,686,457 people have "committed" to write-in Jesus on their ballots. But we'll have to wait and see how many of them live in NC and how many follow through on their "commitment." In addition, Gary Johnson or a write-in is the only option available for people who are both racist and anti-Mormon.
Voter Demographics:
827,998 White Republicans have voted in 2012, compared to 757,456, an increase of 70,542 White Republicans. It's not an accident that the number of additional White Republicans voting early is pretty close to the amount by which Romney has cut Obama's (estimated) early vote margin from 2008. However, it's not clear how much this higher early White Republican turnout will ultimately help Romney, because many of these Republicans may be likely voters who would have otherwise voted on election day.
White Democratic early vote turnout is down by 96,252 votes from 2008, but given the likelihood that many of the missing "Democrats" are really Dixiecrats who did not vote for Obama even in 2008, it's unclear that this really does very much to hurt Obama's chances.
Romney is fighting an uphill battle against inexorable demographic trends. As dreaminonempty illustrates, White vote share has been consistently dropping in exit polls every 4 years year since 1992, with a shockingly high .9922 R-squared value.
We can see that drop in White Vote Share in action in North Carolina early voting. Although the raw number of early votes cast by White voters has increased slightly from 1,809,926 to 1,845,605 (a 35,679 vote or 2.0% increase), White vote share among early voters has dropped from 69.2% in 2008 to 67.3% in 2012. As white vote share drops, Romney has to do better among White voters than McCain just to keep up with McCain's 2008 numbers.
There has also been an increase in African American from 702,958 in 2008 to 752,913 now. You don't need to have a math PhD to be able to tell that the increase in African American turnout of 49,955 votes (a 7.1% increase) is greater than the increase in White turnout of 35,679 (a 2.0% increase).
And among other Minority groups, there has been an increase in turnout from 69,431 to 83,016 - an increase of 13,585 votes, or a 19.6% increase.
It also needn't be said, but the number of White Progressives in the Triangle, the Triad, Charlotte, Asheville, etc has only increased since 2008. That's another part of the demographic change that helped turn North Carolina blue in 2008 - and which could help keep it blue in 2012.
How Accurate is My Methodology?:
Obviously the real test of this question will come on Tuesday. It is one thing to predict the past, and quite another thing to predict the future!
But Ari Bronstein suggested a good way that we can at least see whether my numbers predict the past:
If you can do it (and I'm pretty sure your database has all the info you'd need), you may also want to apply your model of likely voters on each side to ALL the 2008 voters (including Election Day voters), and see how close you are to the true final results. It'll give us all a sense of how good your breakdown of the Unaffiliated voters is when we're looking at your bottom line %.
I did exactly that. Here were the results:
Without, for the moment, bothering to adjust for undervotes and 3rd party votes, my methodology comes pretty close to the actual result. It predicts that Obama should have won by about 47,727 votes. In reality, he won by 14,177 votes.
It makes sense that my model should be pretty close - but a bit more favorable for Obama - to the overall 2008 results, because I built the model in the first place based off of 2008 exit polls to roughly fit the known 2008 early voting results. Again, this does not necessarily mean that my methodology will be accurate in predicting the 2012 outcome - predicting the past is a different animal from predicting the future. But since it makes sense that Obama should have done systematically a bit better among early voters than election day voters in 2008, it makes sense that this should result in a prediction of Obama winning by slightly more than he won in reality.
But soon after running these numbers, I figured out something very interesting.
Since I was drawing my data from current voter data, updated just a week ago by the NC Board of Elections, I was using the current 2012 party registration of 2008 early voters.
Then I realized that the NC Board of Elections also makes historical data available, showing the historical party registration and race as of 2008 for people who voted in 2008, rather than showing their updated current 2012 registration.
And so I re-ran the numbers using the historical registration data, rather than the current registration data. These were the results:
We previously knew that my methodology had some swing among White Voters to Romney built-in to the system, because party registration has shifted away from Democrats and towards Unaffiliated voters. But previously we were not able to quantify how big the built in assumed swing of White Voters to Romney was.
The fascinating thing about this is that it enables us to see the change in party registration from 2008 to 2012 specifically among people we know voted in 2008.
When we used the current 2012 registration of 2008 voters, the predicted Obama margin was 47,727 votes.
But when we used the historical registration of 2008 voters, the predicted Obama margin was 113,906 votes.
The difference between those two predictions is 66,179 votes. Those 66,179 votes are the change in the predicted results that are attributable to the fact that some 2008 voters have changed their party registration between 2008 and 2012.
So, given 2008 turnout, my methodology has a built in assumed net swing to Romney of about 66,179 votes. It's not strictly true that all of that swing is coming from white voters, but it's very nearly true. If we simplify and say that all of the swing is coming from white voters, then those 66,179 votes are the same thing as a 2.1% built-in assumed swing among White Voters to Romney.
Therefore, if you think that there will be absolutely no swing of White voters to Romney, you can go ahead and add approximately 66,179 votes (or, with 2012 turnout, about 70,000 votes) to Obama's predicted 2012 margin. Given the fact that elderly Dixiecrats have been replaced in the electorate by younger white voters less likely to vote against Obama simply because he is black, and given the fact that polls do not show a detectable swing of White NC voters to Romney (PPP has Obama at 36% among Whites, and other polls are similar), that is not entirely implausible.
One other thing you will notice is that using the historical data, the undervotes are eliminated - only voters who voted specifically in the presidential race are included in the historical data.
Armed with this new knowledge, we can also look in a bit more detail at how people who voted in 2008 have changed their party registration since 2008:
From the chart above, we can plainly see that the vast majority of all of the change in party registration among 2008 voters who were already registered was from a shift of former White Democrats re-registering as White Unaffiliateds.
Since the current 2012 registration includes undervotes, and the 2008 historical registration does not include undervotes, we need to correct for that undervote difference. If we assume that people of all races and party registrations were equally likely to undervote, then we get the change in party registration shown in the "Change Corrected Assuming Equal Share of Undervotes" column. We can see a decrease of 69,813 White registered Democrats and a corresponding 54,473 person increase in the number of White Unaffiliateds. 54,473 is not exactly the same as 69,813, so obviously there have been some other changes as well - including some White Democrats changing their reported race to "Undesignated."
Because I assign 83% of White Democrats to Obama and only 28% of White Unaffiliateds to Obama, that change in registration means a pretty large shift in predicted votes from Obama to McCain - and that is why there is the 66,179 vote difference between the two predictions.
Testing 2010:
In addition to "testing" my predictions against the 2008 results, I also "tested" my predictions against the 2010 results in the same way. This is obviously very much a counter-factual, since there was no Presidential race 2010. The only statewide race in North Carolina in 2010 was the US Senate race between Elaine Marshall and Richard Burr. A fairly uncompetitive Senate race in a midterm is a very different animal from a hotly contested Presidential election, but this is what my methodology would have predicted:
My methodology predicts that Elaine Marshall should have lost by 48.1% to 51.9%. In reality, she lost by 43.1% to 54.8% (or 44.0% to 56.0% not counting third party votes).
Although no exit polls are available for NC's 2010 Senate race, it is clear that Marshall did substantially worse with white voters in 2010 than did Obama in 2008. That difference in support from white voters explains the difference between what my methodology would have predicted and what actually happened.
For example, in Buncombe County (Asheville), where 87% of the Voting Age Population is White, Obama got 56% of the total vote, and Marshall got just 50%.
And in heavily white Randolph County (Asheboro), where 84% of the Voting Age Population is White, Obama got 28% of the total vote, and Marshall got just 20%.
This highlights the main weakness of my methodology - it does not take into account swings in voter preferences since 2008, except insofar as those changes in voter preferences are reflected by changes in party registration.
So if there is a very large swing among white voters to Romney, rather than just a relatively modest swing among white voters to Romney, then things could look better for Romney than my numbers would indicate.
But again - I cannot see any detectable swing of White NC voters to Romney (PPP has Obama at 36% among Whites, and other polls are similar). If anyone can find any evidence that White voters in NC are shifting substantially to Romney, please let me know in comments.
Prediction:
Tomorrow, I'll post my final educated guess of the outcome in NC, down to the last vote and with turnout broken down by vote likelihood category, party registration, and race. After the election, we can see how accurate it turns out to be!
When I started writing this diary series, I guessed that NC would probably be a bit closer than most people seemed to assume, but I thought that Romney would probably win in the end. But the more deeply I look into this voter data, the less persuaded I am that Romney really has an advantage, and the more I wonder whether Obama may even have a slight edge. Even just a few days ago, as we saw turnout decreasing because of Sandy, I was feeling much more pessimistic than I am now.
Intrade currently has only a 20% chance that Obama will win North Carolina. I legitimately do not know if Obama will win North Carolina or not, but I really and truly think that his odds are much better than 20%. If I had an intrade account, money to waste, and was a gambler, I'd definitely be buying Obama NC stock. Indeed, if I were managing Mitt Romney's blind trust, I think I'd buy him some Obama NC stock!
Likewise, Nate Silver at 538 currently gives Obama only a 21% chance of winning. I'm a big fan of 538 and have been reading Nate Silver since 2008, and I generally think Nate's national forecast will be more accurate than anything else out there. But the problem, at least in the case of NC, is that 538 does not take ground game, early vote, or changes in voter registration into account.
Ground game is not a factor in other southern states where demographics are shifting in Democrats' favor, such as Georgia and Texas. But I think that 538 probably overestimates Romney's margins in those two states as well, for the same reasons of demographic, early vote, and voter registration - Nate currently has Georgia at 54.5% Romney, 45.0% Obama, and has Texas at 57.6% Romney, 41.8% Obama.
It's all about GOTV now.
Early Vote Turnout Charts:
Minority and White Democratic turnout is tracking 2008. Of course, inside that number, there has been a drop-off in the number of White Dems since 2008, partly because some White Dixiecrats have changed their party registration to Unaffiliated or Republican since 2008. And there has been an offsetting increase in Minority voter turnout.
White Republican early vote turnout continues to be high, but of course as we noted above, this reduces the pool of White Republicans who may vote on election day.
Obama's cumulative margin had leveled off at just under 150,000 in the wake of Sandy, but on the last day of early voting Obama's margin noticeably shot up.
From the daily margin chart, we can see that Saturday was just like 2008, in terms of Obama's estimated vote margin. Likewise, the beginning of the early vote period was just like 2008. The drop-off in Obama's vote margin (if indeed the built-in swing of White voters to Romney comes to fruition) came entirely during the intermediate period of early voting.
Convergence with 2008 is evident in the estimated cumulative Obama percentage chart.
Obama's estimated daily vote percentage has very clearly converged more closely with 2008 over the last week.
Previous NC Early Voting Diaries:
Day 1 & Methodology
Day 2
Day 3
Day 4
Day 5
Day 6
Day 7
Day 9
Day 10
Day 11
Day 12
Day 13
Day 14
Day 15
Day 16