Our friends at Survey USA have been getting some flack in the comments of this post on MyDD for the percentage of black voters they are showing in the North Carolina polls they put out yesterday. I've already broken down their Senate and Governor surveys, and I'll get to their interesting Presidential numbers later this morning.
The reason they're getting flack is that they showed the percentage of the black electorate at 20%, when the 2004 North Carolina exit poll showed it at 26%. Obviously the difference between 20% and 26% is going to make Clinton look better relative to Obama.
I've been using the 26% figure for our NC general election polling but the more I thought about it last night the more the 26% figure just didn't make sense. North Carolina's population is only about 19% black, and it would be pretty unusual for them to be represented at 26% in a general election.
So I e-mailed Gary Bartlett, the director of the extremely well run North Carolina Board of Elections, this morning and asked him for the demographic data on the 2004 electorate in North Carolina. It turns out the actual black electorate in the state that year was a little under 19%.
So the exit poll was incredibly wrong. This is a good example of the caveats with weighting polls to prior exit polls. We will certainly be changing the way we weight for race in North Carolina based on this information, and we are not going to be blindly accepting any past exit polls as fact when figuring out how to weight things in other states either.
I think the figure Survey USA used is pretty reasonable, and hope they don't get any more gruff for that particular aspect of the poll.