There have, naturally, been several attempts at reading the tea leaves from the
daily results from the tracking polls. Some of the dailies are published up-front (like Suffolk?), others are only available through "reverse engineering". It is important to remember that all sorts of corrections are done in order to increase the validity, of any serious poll, at least. So: the reverse engineering is often not as straight-forward as it seems.
Corrections mean that the numbers from the responses are weighed against responses to questions like "what did you vote last general election, last statewide election, last congressional race, ballot measure", and when (in 99 of 100 cases) those results are off, compared to what they should be (based on the actual results from previous results) the poll results are weighed. There is often weighing based on sub-samples too, like gender, age, rural/urban, congressional district, where the facts are known. The day-one vs. day-two subsets are (normally) irrelevant for corrections, because the one-day samples are too small.
And, to make the matter worse, people often don't tell the pollsters the truth about how they voted last election. Generally, people tend to claim they voted for the winner, except if the winner has come to be seen as unpopular...
The important point: The correction math is done from scratch every day, based on the new two-day / three-day sample. Because of this, "reverse engineering" of two-day or three-day polls into "one day results" will often not work. Or, to be precise: they will often be inaccurate.
Also: Each polling institute uses its own blend of correction formulae (independents vs. democrats, what is a likely primary voter this year, what's the conversion factor for leaners? How many undecideds are likely voters?). That fact may account for most of the differences between the polls. Because of this, polls are much better at predicting ranking than actual percentage. All the tracking polls are more in agreement of rank, and of trends, than on results, in terms of numbers.
Current rank, as I understand it: Kerry, Dean, Edwards/Clark (tie), Lieberman. All the polls are pretty much in agreement on this.
