As they write (bold face by me):
Some caveats: The poll is a mixture of auto-dialing and online responses that showed Jack Kingston with a huge lead in the U.S. Senate primary runoff, but what matters here is the movement. A week ago the same poll had Perdue ahead by one percentage point.
I beg to, slightly, differ. It's dangerous to over-analyze "change" between two polls using the same error-prone methodology. Essentially, you have two robo-type polls with "change" in both instances lying within the margin of error. In other words, I'd argue there's not much change at all. They remain in a statistical tie.
But -- this is important -- while there isn't much change, that's no fun to write about. That doesn't sell papers, or get clicks, or draw viewers. So of course this gets more attention than it probably, mathematically, statistically, deserves.
You can see more of the poll, with crosstabs, here. You can see the earlier poll here. A few interesting differences emerge in the makeup of the two samples. For example, the previous poll had 30 percent black. The more recent poll has 27 percent black "likely voters" in its sample.
These polls call landlines or reach smartphones to provide a questionnaire. No live interviews, no humans talking to humans (the gold standard of polling). That said, at least they're trying to reach people other than via landlines, though it's preferable to call cells (using humans, as it's illegal to robo-call cells).
Are there fundamentals that favor a Nunn upset? Not really. Still, it's possible, for instance, for very technical reasons, many of the polling models are underestimating black turnout by a percentage point or three. The reasons are nerdy and PhDweeby, and I don't want to spend pixels explaining, but it has to do with Census weighting and the use of 2012 data to estimate 2014 turnout. Those teeny tiny percentage points, however, can make all the difference in the world, in a close race. Would I bet on Nunn? Nope, not straight up, but if you give me 7 points, I'll take some of that action.