Polls

[ATFFL]

If nobody minds me going back to somethign eariler, a poll of 500k peopel would pass the law of diminishing returns on MoE, but would really boost the confidence level in the poll.  A problem would be that it would be very hard to do a sister poll or two  to confirm the confidence.  It would be much more cost effective to sponsor 3 polls of 10,000 or even 5,000 voters and get a similar MoE and have checks to confirm your confidence in the poll.

A lot of polsl you see released with 1200 respondents are 3 polls of 400 merged.  The the 400 polls are taken and if they are all in MoE to each other, they are released as one poll.  If 2 are similar and one not, you can eithe rget anothe rsample of 400 or release it as 800 and lower your MoE and confidence a bit.

[ATFFL]

The process you are describing will skew in odd ways some polling data.  The correct process would be always to merge three four hundred person polls into a 1200 person poll, on an overlapping basis if it is a time series, even when the sets differed,  recalling that part of the time the data sets are supposed to be outside the margin of error of each other.  Rasmussen does this.  If the data sets are always within the MOE of each other, and enough sets have been taken, the data is very likely to be wrong.

If nobody minds me going back to somethign eariler, a poll of 500k peopel would pass the law of diminishing returns on MoE, but would really boost the confidence level in the poll.  A problem would be that it would be very hard to do a sister poll or two  to confirm the confidence.  It would be much more cost effective to sponsor 3 polls of 10,000 or even 5,000 voters and get a similar MoE and have checks to confirm your confidence in the poll.

A lot of polsl you see released with 1200 respondents are 3 polls of 400 merged.  The the 400 polls are taken and if they are all in MoE to each other, they are released as one poll.  If 2 are similar and one not, you can eithe rget anothe rsample of 400 or release it as 800 and lower your MoE and confidence a bit.

You are describing a tracking poll, I am describing an internal check method.

[ATFFL]

Zogby's Internet polls are a bad joke and including them would invalidate the entire point of the polls page.  I'm pretty sure the only Zogby polls included are his traditional polls.

SLight requires a 2-5% lead AND to have won all 3 of the last polls.  The sates you cite fail the last requirement.

[ATFFL]

One more note on the polls page:  If you win 2 of 3 polls but still have an average of 5% or greater, it will show the candidate with a slight lead.

[ATFFL]

Do you know the rules of the polling secetion for sure or you explaining your observations of them.

As best I can figure out.  Only Dave knows exactly how they work, Vorlon has said he is pretty sure it uses only the three most recent, and I number crunched the states when questions came up.

[ATFFL]

The Rasmussen tracking poll should be updated weekly, with his weekly numbers.

[ATFFL]

Strategic Vision is one of the better pollsters that are aligned with one party.  Their independent polls (the ones they do on their own and not for a candidate) are usually decent.

Look at some of their recent polls:

Wisonsin: Kerry +3
Michigan: Kerry +2
Florida: Bush +2
Georgia: Bush +11

None of them are way off.

[ATFFL]

Why is it that Pennsylvania has been moved to tossup because of a tied poll, while Colorado has a tied poll in one of the last three but remains 'Leans Bush'?

Colorado still averages over 5%.  If the average is over 5% it will show lean despite ties, or even the candidate losing in one poll.