Atlas 3 Poll Aggregate Map v. Actual Results

[King]

I took the numbers off the Presidential 3 Poll Aggregate for each state in the polling section and compared the margin of victory in the actual results to the margin predicted in the polling section and then mapped it out.

The polls did fairly good.  Most of the off states were safe ones that were too sparsely polled to be very accurate anyway.  Overall, there were 31 states that were only 3 points off or better and 9 states that were dead on.

The most Obama's best poll over performances came in Vermont and Idaho where he did 12 points better than the polling spread.  McCain's best was Alaska where he did 13 points better.

Overall, there was no pattern found on this map though there were some neighboring states that posted similar results. Notably there were two strings off connected "dead on" states.

[Gustaf]

So, the only state called wrong by the Atlas was Indiana? How does that compare to RCP and 538?

[Alcon]

So, the only state called wrong by the Atlas was Indiana? How does that compare to RCP and 538?

538 missed Indiana (had McCain +1.5) and NE-2 (McCain +6, although it's ridiculously hard to model NE-2).

RCP mis-called North Carolina (M+0.4) and Indiana (M+1.4).  No attempt was made at NE-2 but they would have been wrong.

Thought I don't think we should be punishing sites more for the difference between, say, McCain +0.1 and Obama +0.1 and McCain +0.1 and McCain +7.6.

[Gustaf]

So, the only state called wrong by the Atlas was Indiana? How does that compare to RCP and 538?

538 missed Indiana (had McCain +1.5) and NE-2 (McCain +6, although it's ridiculously hard to model NE-2).

RCP mis-called North Carolina (M+0.4) and Indiana (M+1.4).

Thought I don't think we should be punishing sites more for the difference between, say, McCain +0.1 and Obama +0.1 and McCain +0.1 and McCain +7.6.

I do. Be proud of the poll map attached to a forum of which you are a member!

[kevinatcausa]

It looks like if we forced the Atlas polls to choose on tossup states, it would also have missed IN and ND and may or may not have missed NC (I'm not sure how the Polling average handles things when multiple polls are released on the same day...is it just arbitrary which one counts as the most recent?)., as well as completely abstaining on Missouri.