Steven M. Warshawsky looks into Polls specifically their "margin of error", sampling noise, ideology weighting, and voter intertia.
It's an interesting peice that essentially says that the polls are too close to call given the sampling errors.
Which is a perfectly fine way to look at the data. One example is the drastic shift in voter opinion and party affiliation that some polls have in day to day. Normally, voters don't change their opinions back and forth like that, but... a sample that has a lot of noise would show a fair bit of random jittering.
Of course, it's not much of a horserace if you look at the data and go. "Well, things may be going in Candidate A's favor but we can't really be sure."
It's much more marketable to cite numbers and points and make it look like a contest that's acutally being tracked.