My Polling Methodology

TPT reveals how his polling projections are produced

1. I only look at polls of likely voters.

2. I exclude polls released/leaked by campaigns (I include polls comissioned by The Daily Kos in this catagory).

3. I treat a poll over multiple days as occuring on its end date.

4. If there are multiple polls for a single data I sum the sample sizes and the weight the results by the sample sizes.

5. Overlapping tracking polls are discarded.

6. I filter the data through Samplemiser using the default setting.

7. The processed data is then used to produce daily polls, even for days that did not have polls in the original data.

For example if I was looking at three polls:

A. Doe Polling (July 1-6, 500 LV) McCain 50 Obama 42
B. Smith Analytics (July 6, 1000 LV) Obama 48 McCain 44
C. Obama Internal Polling (July 5-6, 2000 LV) Obama 47 McCain 46
D. Green Research (July 4-6, 500 RV) Obama 50 McCain 40

I would ignore polls C & D and count this as :

(July 6, 1500 LV) McCain 46 Obama 46