Pollster.com

When is the last time a true third party candidate had a seat in the U.S. House? I am not counting "Independent" candidates who have no party affiliation or local parties, such as MN's DFL.

Hmmm, i wonder if this is truly accurate. CFG sponsored the poll and they endorsed the candidate that the poll shows leading. Never heard of Basswood Research either. High margin of error too.

Yeah, this poll is statistically invalid based on reason previously stated.

Also, what is the poll's confidence percentage? Anything under 95% is invalid.

I find the result unlikely. Compare to the Siena and Research 2000 polls from the 13th and the 21st respectively, which can be found here:
http://www.pollster.com/polls/ny/09-ny-23-ge.php

Both of those polls found Owens at 33-35, Scozzafava at 29-30 and Hoffman at 23.

Obviously some time has elapsed but not a huge amount, just 4 days between the R2k and this poll and yet this indicates significant movement with Owens -8, Scoz -10, and Hoff +8. Even if we accept that 8 points of Scoz support changed to Hoffman after some of the conservative endorsements recently why would 8 points of Owens support evaporate? As the election approaches the amount of don't knows should either decrease or hold steady, not increase.

Given the large suspicious difference from two previous polls, the partial source, and the small sample size I suspect the CfG commission several small polls and chose to publicize the one that said what they liked.

Basswood research is the house pollster for Club for Growth. They get the numbers their masters ask for and they are trying desperately to create a narrative that allows for a Hoffman victory.

Appears to be a Hoffman poll. Sample too small to take seriously in any case. See 539's piece on his site that shows that comparing her to other NY Republicans she actually is slightly more conservative than average. Of course this would make her a "Socialist" for many Republicans elsewhere.

I expect to see Hoffman overtake Scozz mainly because over the weekend, Palin came out to stump for Hoffman drawing huge crowds and helping him raise over $120k. It will be interesting to see the results on the 3rd.

CLUB FOR GROWTH POLL? You have to be kidding me. I wait for another one

On further look the issue looks more suspicious. teh cross tabs for the poll are here:
http://www.clubforgrowth.org/media/uploads/NY-23-top-lines.pdf

The big thing that pops out is their sample set had 48% republican and 36% democratic. According to the Cook PVI NY-23 is only R+1 (the PVI is measured by averaing the previous two presidential election results). I'm trying to find a good party ID breakdown and havign trouble but previously I think I've seen D+2 (I may be misremembering so don't take as fact).

In addition the largest age group was 65 or older at 34% of the sample. This was a LV survey and seniors tend to vote more often than other ages but unless NY-23 is a big retirement area that much of a skew towards seniors seems odd. Overall NY demographics have just 13% of the population aged 65 or older:
http://en.wikipedia.org/wiki/Demographics_of_New_York

Again more suggestions that this survey was cherry picked. Unless we get confirmation of such a swing from an independent source I think we can safely ignore this poll.

Gopherguy: If you're referring to statistical confidence levels, 95% is standard for the margin of error reported in polling, although that is left out of most polling reports. So while margin of error varies by sample size, the confidence level is held constant.

Hope that answers your question!

Ms. Swanson,

Yes, I'm wondering what the confidence level of the poll is. I have a feeling the level is 90% or lower. If my suspicion is correct then this poll is highly dubious.

Keep in mind that the confidence level is relevant to calculating the margin of error, but is NOT a characteristic of the poll itself. Margin of error could theoretically be calculated using any confidence level.

For example, the margin of error reported for a sample of 1,000 is about 3% if calculated using the standard 95% confidence level, but if it's calculated using a 99% confidence level the margin of error will be over 4%, while at the 90% confidence level it will be about 2.6%.

In this case, a sample of 300 does in fact have a 6% (5.66%, actually) margin of error using a 95% margin of error. Anytime you see margin of error reported with no confidence interval, you can assume that it was calculated using the standard 95% confidence level.

Correct, but it does go to the reliability of the poll. I would rather take a drug that had a 95% confidence level of not harming me than a drug that had an 80% confidence level of not harming me.

If confidence level is too low we cannot rely on the poll's accuracy for its snapshot in time.

For example, 95% means the results should be accurate 19 out of 20 times. 90% would mean that the results should be accurate only 9 out or 10 times.

Thanks Emily, but it seems to me that failing to report the confidence level is poor practice by the pollster.

Having faith, especially from a biased polling group, that a poll has a 95% confidence level makes no sense.

We should strive for evidence based conclusions, not blind faith.

After all, blind faith got the public to believe WMD existed in Iraq, which gave the Bush Administration public support to wage war. We all know how that has turned out.

@ Gopherguy

Most polls state their confidence intervals at a 95% level, and this one is no different. Gallup, Ras, PPP, whatever, they all typically imply a 95% confidence interval when they report their findings. As Emily said, for a samples size of 300 the 95% confidence interval is roughly 6%, which is huge for a poll (hence the reason that most polls don't stop at 300 respondents).

There are much better reasons to doubt this poll: the incredibly small sample size, the partisan sponsorship, the huge margin of error (a result of the small sample size), the questionable party ID breakdown.

http://www.fivethirtyeight.com/2009/10/reality-check-ny-23-poll-may-seek-to.html