We have our first state-level tracking poll, this one from the Morning Call in Pennsylvania. The first edition showed Obama leading John McCain by 4 points.
Technically speaking, the model treats a tracking poll the same as any other poll, except that for purposes of calculating its weighting, we divide the sample by the number of days in the tracker. For instance, a 900-person tracking poll that spans a 3-day period is treated as having a 300-person sample size … that is, the number of new interviews that the tracking poll conducts each day. (If a tracking poll breaks out its daily samples, as the Research 2000 national tracker does, then we’ll simply use those results directly.)
As with other polls, we do not completely throw out the old results when the tracking poll publishes a new result; instead, have we have a formula for discounting the weighting for past polling from the same firm.
This does lead to a question about how you guys want me to list tracking polls in our data table. Do you want all the individual results listed — which is more comprehensive but will create more clutter? Or do you want me to combine all the results from the tracking poll into one line in the data table?