A Note on Polling

I'd never paid much attention to all of the various firms out there until I got into campaigning last summer; when you're alert, you notice that some are better than others, and more accurately predict elections. A regular reader over at The Fix posted a statistical breakdown of the various big firms' successes in 2004, and I thought it worth posting here, just to give you all an idea of how credible all of these numbers are:

In reviewing state by state polling by various firms in the 2004 Presidential race, I set up a quick scoring system to judge accuracy.

Taking the final poll done by each firm in a state and matching with actual results, I scored 1 pt for each correct call of the state, 3 pts for the exact margin of victory/1 pt if within margin of error (3%) and 3 pts for calling the candidate vote total by exacft percentage / 1 pt. within MoE.

I also tracked misses (states called incorrectly) and quality picks (where the state was called and the margin and candidate percentages were all within the MoE). No points were awarded for misses:

Results:

ARG (American Research Group)
44 states polled
131 pts. (avg. 2.997)
5 misses (11.36%)
10 quality picks (22.73%)

Gallup
10 states
19 pts. (avg. 1.9)
4 misses (40%)
2 quality picks (20%)

Mason Dixon
21 states
92 pts. (avg. 4.381)
1 miss (4.76%)
7 quality picks (33.33%)

Quinnipiac
5 states
11 pts. (avg. 2.2)
1 miss (20%)
0 quality picks (0%)

Rasmussen
30 states
101 pts. (avg. 3.367)
1 miss (3.33%)
13 quality picks (43.33%)

Research 2000
11 states
39 pts. (avg. 3.545)
0 misses (0.0%)
6 quality picks (54.55%)

Strategic Vision
11 states
24 pts. (avg. 2.182)
4 misses (36.36%)
3 quality picks (27.27%)

Survey USA
30 states
116 pts. (avg. 3.867)
1 miss (3.33%)
16 quality picks (53.33%)

Zogby
20 states
55 pts. (avg. 2.75)
3 misses (15%)
6 quality picks (30%)
1 perfect call- Missouri

The state by state polling data was compiled at: www.tripias.com/state/

The point system may be worthless as I just concocted it myself but it allows a comparison since all firms did not poll in all states.

As far as accuracy, Mason Dixon, Rasmussen and Survey USA came away with only 1 miss and had the most net quality picks minus misses

Thanks RMill - awesome as always.