State-by-State Polling 2010 - An Analysis

[tpfkaw]

Not sure if predicting the final margin is the best way to rank polls - for example, that Hays Research poll in AK had McAdams in second by a substantial margin, IIRC.

It would probably be better to compare them based on the average differential from the final performance of each candidate than the final margin.

For example, if the final margin were:

R: 53
D: 46

(R+7)

And you had one poll:

R: 48
D: 42

(R+6)

And another:

R: 54
D: 43

(R+11)

The second poll is further off from the final margin, but:

First poll:

|53-48| = 5
|46-42| = 4

(5+4)/2 = avg. candidate error of 4.5

Second poll:

|53-54| = 1
|46-43| = 3

(1+3)/2 = avg. candidate error of 2

While the first poll predicted the margin better (and thus has a greater level of accuracy), the second poll still has a greater level of precision, more closely tracking the actual results of the candidates.  I personally think precision is more important than accuracy when it comes to polling, although some combination of the two would also be a good measure.