Okay, what I should have said was that more than likely it's not statistical noise. Better?
I guess. Honestly I don't know at what point something becomes less likely to be statistical noise. The 50% confidence interval, I guess? Not sure when that's reached. Wish there were a calculator for that.
I was just pointing out being concretely sure it isn't statistical noise when it's within the 95th confidence interval is a bit dangerous. Methodology/sampling errors + statistical noise = a lot of uncertainty, maybe 1-in-7 polls or something like that. But, yeah, better, def.
If we assume a normal distribution, then the reported MoE for a 95% confidence is stdev*1.96, whereas the MoE for a 50% confidence is stdev*0.675. That means a move of (MoE=4%)*.675/1.96=1.38% is more likely than not to be non-noise legitimate movement. Of course, that is for sampling error. There is also systematic error, such as the wording of questions (I still can't get over the fact that most pollsters ask how you feel about George Bush immediately before asking about the presidential race).
In any case, since McCain just visited that tiny empty state, the echo chamber of local press can be expected to have given him a transient bump that a good pollster should be able to detect while it is still fresh.