First, for this cycle as with previous cycles, PollTracker reports online polls but does not calculate them in our averages. The 'online' methodology is maturing rapidly -- and some seem quite solid. But we still don't consider the methodology as a whole sufficiently reliable for inclusion. So the Ipsos/Reuters tracking poll, which has been running over the last ten days, is not included in our averages at all.
The more important issue is how we compute tracking polls.
Tracking polls are not complete polls. Each day is a sliver of a complete poll which cycles on a rolling basis over a period of days -- 3, 5 or even 7. A tracking poll from today and yesterday include mostly the same data. So they're only included every complete cycle, not every day they appear. Because of that, even though Gallup and Rasmussen have been moving rapidly over the weekend, that data hasn't gone into the average yet.
Our system also, by design, 'down-weights' tracking polls like Gallup and Rasmussen. Both Gallup and Rasmussen have strong 'House Effects'. And because they release numbers so frequently their "House Effects" tend to dominate an average. But that's for a totally arbitrary reason -- the rapidity by which they're generating numbers. So Gallup and Rasmussen count for way more than ABC/WaPo, NBC/WSJ and NYT/CBS just because they're churning out so many numbers. And that doesn't give us a clear picture of what's happening.
When you put all those factors together, what you're left with is a dearth of 'real' polls over the last two weeks. Just CNN on September 3rd (tied) and the new ARG poll which was completed on the 6th and released today (Romney +3).
Put all those factors together and we're left with an odd case where the tracking polls show rapid movement over the weekend -- a point I dedicated a whole post to yesterday. And yet the trend estimate shows Romney just barely ahead. Yes, it's kind of weird. But the point of having a methodology is to find one that you think makes most sense in terms of averaging numbers and predicting trends and stick with it. Our methodology was developed in consultation with Professor Charles Franklin of the University of Wisconsin, one of the nation's top experts on the statistical analysis of public opinion and elections data. And we think it's the best one. So sticking with it, even when it generates the occasional weird result, is just what we're doing.
Expect a raft of new national polls over the next few days which I suspect will clarify where things are pretty quickly.