I am being selfish and reposting this so I can find it easily over the next several months.
As I understand it, a tracking poll is a moving sample taken daily, discarding the earliest third, let's say, and adding the most recent third. The size of the sample each day may be smaller than the typical (one-month apart) Gallup or WaPo Poll but the aggregate may be the same or bigger. It assumes that the sample on day three exactly is equivalent to the sample on day one. But what if it's repub skewed on day one compared to day three?
Here's an academic review:
Lau examined eight factors that might effect the accuracy of a poll to predict the vote for the U.S. President. Six were methodological factors: (1) sample size, (2) estimates of likely voters, (3) number of days that the interviewers were in the field, (4) whether the poll was a tracking poll 'where a small sample is drawn each day, but reported results sum across several days), (5) whether the poll was done on weekdays only, and (6) the strictness of the definition of a 'supporter' of a candidate. In addition, two contextual factors were examined: percent undecided and days to election. Lau examined these variables for a sample of 56 national surveys conducted between August 31 and November 2, 1992. The data for each of the methodological factors did show wide variations. For example, sample sizes varied from 575 to 2086.
Note that tracking polls do well for accuracy in the academic review (i.e. Rasmussen).
Also note:
Discussion from the News Hour and Andy Kohut (Pew, non-partisan) and Dotty Lynch, the senior political editor for CBS News.
TERENCE SMITH: Tracking polls, explain briefly a tracking poll.
ANDREW KOHUT: A tracking poll is a continuous survey where a new component of a sample is added each day to show the progress of public opinion. It's generally designed to show the ways in which advertising campaigns are being effective in markets or states. But applied to the national electorate, what are they tracking? They've been tracking public opinion for two months, and most of it, save around the time of the debates, has been noise. (...)
TERENCE SMITH: Dotty Lynch, what is the point and what is achieved by reporting a tracking poll, perhaps on a daily basis, and then immediately saying, as a caveat, that it's within the margin of error and it's therefore a dead heat? I mean, is anything served by that?
A venial sin?
DOTTY LYNCH: Well, I think that's one problem that I have. And I violate the rule myself, because I'll say, well, our poll shows that it's statistically tied. And then someone will say, but who is leading? And I'll say, well, Gore or Bush. And we know ourselves as pollsters that that's wrong, it's at least a venial sin to say, well, but Gore has a little bit of a lead, but it's human nature. And it's what people want to know. And everybody will ask you, who's going to win. So I think it serves a purpose of at least defining the race, yes, it's still close, but this guy may have had a little edge because of the debate. That one did something because of an ad, good or bad or whatever. It gives you a way to understand something that's uncertain. And I think myself, some of the poll bashing is just people are uncertain. They don't know what's going to happen. And it sort of, you know, kill the messenger, kill the pollster, try to blame somebody for the uncertainty.
The NewsHour discussion, while referring to 2000, is quite relevent today.