Part of the reason FHQ shifted from a simple weighted average to a graduated weighted average (one that progressively discounts polls based on when they were in the field) is that states like Minnesota and North Carolina were unresponsive to a series of new polls that ran counter to where our averages had each state.
Some of that unresponsiveness was remedied with the methodological change, but did not move either state as much as some would have liked. The bubble seems to have burst in Minnesota for McCain, so the North Star state has worked its way back to essentially where it was prior to the conventions -- a strong Obama state (Sure, some of that has to do with the threshold being dropped.).
North Carolina, though, is a bit different. In the Tar Heel state we have witnessed a string of polls that have shown Obama ahead by margins up to 6 points with just a few pro-McCain polls peppered in. Yet, it is still seemingly stuck in the McCain toss up category. Much of this has to do with the amount of information we have in North Carolina. Even with the older polls discounted, there is an awful lot of McCain support inherent in the average. In other words, there are a lot of McCain polls for this recent series of Obama polls to overcome. The Tar Heel state has had around 50 polls conducted this year and none of them (other than the Zogby internet polls) favored Obama until after the Lehman collapse. That's a lot of McCain support in the average.
As far as a magic number is concerned, North Carolina is a lot like Ohio: it is going to take a lot to move things just a little. For the next poll to push North Carolina into the blue, it would have to give Obama a margin of 45 points. That's just not going to happen. But we may continue to see numbers come in under the current average of 2.4 (for McCain) that continue to chip away at that margin. In fact, I think that is likely. Between now and election day, we are likely to see polls that are in the +/-3 point range with some outliers thrown in.
Just for the heck of it let's do an exercise here. What if we lopped off all the polls conducted before Obama clinched the Democratic nomination; everything from June 3 on (FHQ has done something similar before.)? How would that affect things? Once we reweight the polls based on a lower number of days in the period examined, we find that Obama gains, but that McCain's lead shrinks to only 1.7 points (down from 2.4). What is North Carolina's magic number then? Not surprisingly, it drops, but not to anymore manageable a level. It would still require a poll with Obama ahead by 25 points to turn North Carolina blue. Obviously, the the number of pro-Obama polls it would take to successfully chip away at that average and turn it blue would be far fewer in this instance.
This is in line with my thinking about North Carolina. I'm a native Tar Heel and though I'm not there now, I still have family ties to the state. My sense is that North Carolina is a "close but not quite" state for Obama. Sure, I've been out of the state for a while, but North Carolina still feels (And yes, that certainly strays from the black and white we get from the numbers typically leaned on here at FHQ.) like a state that is a continued demographic shift away from becoming less reliably Republican -- at the presidential level -- and more reliably competitive. It speaks to the Democratic tilt of this election that North Carolina is talked about in the same breath with the Ohios and Floridas on the map.
UPDATE: Our discussion has extended beyond North Carolina in the comments to encompass a discussion of much of the South. Scott has taken the Census data on the African American percentage of the population and regressed that on Obama's support among whites in these states. A simple bivariate regression with some rather interesting results.
Here are the states Scott looked at (all have at one point or another shown John McCain and Barack Obama within single digits of each other):
Virginia: 39% (20%)
North Carolina: 38% (22%)
Georgia: 28% (30%)
South Carolina: 25% (29%)
Louisiana: 18% (32%)
Mississippi: 16% (37%)
Below is the plot of that relationship; one that shows a rather high correlation between the two variables. The data above are rank ordered based on the dependent variable (Obama's white support) and are displayed as such below.
|Obama's White Support as a Function of the Percentage African American|
A big tip of the cap to SarahLawrenceScott for a nice addition to our discussion. Kudos!
The Electoral College Map (10/20/08)
The Electoral College Map (10/19/08)
The Electoral College Map (10/18/08)