Now Cross-Posted on Dailykos
For the past few months, I have been frustrated by the difficulty of determining rolling averages for national Democratic primary preference polls. There are three basic causes for this difficulty, but at long last I have developed system to deal with those three difficulties that I personally find satisfactory. Here are the three difficulties, along with my solution to each:
- Not all polls include the same candidates in their questions. Polls that include Gore and Kerry artificially deflate the numbers for all other Democrats, while polls with only Hilary, Edwards and Obama artificially inflate the numbers for those three candidates. Thus, for my rolling poll average, I have decided to only include polls that include all announced candidates, and do not include either Gore or Kerry. This way, the numbers I am using in the averages are more compatible with one another.
- Different polls sample different universes of voters. However, during my behind the scenes work on the Inflated Clinton Poll Theory, I now feel that the universes samples among the polls that meet criteria #1 are currently not different enough to pose a serious problem during averaging. While that includes Rasmussen and Gallup, both of whom are outliers in this regard, to a certain extent cancel each other out, both because they are outliers in opposite directions and because they are the two most prolific pollsters. I still believe a tighter sample of Democrats needs to be the universe all pollsters sample, and I will keep pushing for it behind the scenes, but this will do for now.
- Different polls push undecideds to different degrees. This is still a problem, but name recognition among the three leading Democratic candidates that pushing undecideds no longer heavily favors Hillary Clinton. Obama and Edwards both hover around 90% name recognition among Democrats. By restricting myself to the post announcement period (mid-February and later), while this is still a problem, it is not as much of a problem.
Now, with all of this in mind, without further ado here are the national Democratic primary polling average from February 23rd until April 24th, presented in three and four day intervals:
| Date |
Clinton |
Obama |
Edwards |
Others / Unsure |
| Apr 24 |
35.7% |
28.4% |
17.6% |
18.3% |
| Apr 20 |
35.7% |
28.0% |
17.2% |
19.1% |
| Apr 16 |
37.0% |
26.3% |
17.0% |
19.7% |
| Apr 13 |
37.0% |
24.5% |
17.0% |
21.5% |
| Apr 10 |
37.8% |
22.8% |
17.3% |
22.1% |
| Apr 06 |
37.7% |
23.3% |
17.0% |
22.0% |
| Apr 02 |
37.3% |
23.3% |
17.2% |
22.2% |
| Mar 30 |
36.6% |
24.6% |
16.8% |
22.0% |
| Mar 27 |
37.0% |
25.4% |
15.6% |
22.0% |
| Mar 23 |
37.7% |
26.0% |
14.0% |
22.3% |
| Mar 20 |
39.0% |
26.3% |
13.3% |
21.4% |
| Mar 16 |
39.2% |
27.5% |
13.3% |
20.0% |
| Mar 12 |
39.0% |
26.8% |
14.0% |
20.2% |
| Mar 09 |
37.4% |
27.1% |
13.7% |
21.8% |
| Mar 06 |
37.9% |
27.1% |
13.4% |
21.6% |
| Mar 02 |
37.8% |
24.8% |
14.0% |
23.4% |
| Feb 26 |
38.8% |
24.5% |
13.8% |
22.9% |
| Feb 23 |
42.3% |
23.0% |
14.3% |
20.4% |
The excel spreadsheet I use to gather this data
can be found here. Note that polls are added to the average the day after they are completed, and removed from the average fifteen days after they are completed. Only candidates averaging over 5.0% in national polls are included in the chart. Source polls can be found at
Polling Report and
Rasmussen. I will update this chart every three or four days throughout the entire primary / caucus season.
For now, I will leave analysis of this data to you. One thing I will note is how Edwards and Obama were stuck at a combined 38.3%--41.6% for a long time. However, over the last ten days, they have quickly risen to a combined 46.0%, mainly at the expense of others / unsure, but also, to a lesser extent, at the expense of Clinton. That strikes me as a particularly important trend in the campaign, one that is an entirely new development and that impacts every single announced candidate.
