As I have stated in the past, the art of polling has fallen into the same disrepute as statistics. And I have some bones to pick with the two polls that have been released recently.
#1. If you do not release your methodology, it is hard to take your poll seriously. There is absolutely no reason to do this other than trying to hide something. If you are trying to hide something, there has to be bias in your poll. It is as simple as that. If your job as a pollster is to ferret out what is happening in the electorate from an objective sample, you will want to release your methodology to stamp your reliability. Neither of these polls have done that.
Both third party polls taken during the primaries DID do this.
#2. Obviously the Greenburgh-Quinlin-Rosner poll was sloppy and can almost be tossed aside in terms of value. I would bet a dozen Krispy Cremes (and those are hard to get now) that this poll by far oversampled Democrats. We do not know why this poll was commissioned, or what feedback was being targeted by the Giffords campaign that led to these results. It could have been Pima County residents for all we know. Again without the methodology, the results are almost as useful as an online poll.
Also, look at the favorablility ratings for Graf. 82% of respondents know who he is, only 32% think favorably of him, and yet 35% will vote for him? Who are these people that have no idea who Graf is, or actually have an unfavorable opinion of Graf, but will vote for him anyway? Also notice that there are 11% of respondents who know Randy Graf but have no opinion on him. This seems very unlikely. It appears to me that people were being pushed to answer certain questions, perhaps not in a specific way, but nevertheless it diminishes the sampling credibility of the poll. An "I don't know" is perfectly valid as a choice.
And finally, anyone looking at this poll with an analytical bone in their body would immediately know that Gabby is not 19 points up. That was just silly.
#3- I believe the Star poll to be more sound fundamentally, but definitely not beyond question. Again, no breakdown is listed. What I can tell you right off the bat is that women were oversampled in comparison with men. It is also telling that in the Star poll released during the primaries, they DID give the breakdown and defined "likely voter."
Here is the thing that also makes me question the poll. If you buy that 45.8% of the voters view that Border control is the number 1 issue for this race, and that Gabby beats Graf on this issue, that means that Graf has to be beating Gabby on other issues such as the war and health care in order to make the numbers stand up. This is very counter to the conventional wisdom going both directions which could happen, but is unlikely. I also do not buy that Gabby is outdistancing Randy in the outlying counties, unless the sample size of these counties is statistically insignificant (which would explain a lot.)
The over sampling of women and the probable bad sampling of the outlying counties would suggest to me a rushed, inexact poll that the Star knew was flawed, but released anyway minus the internals. "600 likely voters" is probably the truth, but you also need to account for party registration, location, et al. I doubt that this was done.
So while this poll is certainly better than the Giffords poll, it does not, nor cannot tell the entire story.
I would argue that the last valid, verifiable, accurate poll that was taken was the Star poll just before the primary. And that poll is getting moldy.
I am sure that there are others who disagree. But take notice that whenever the poll methodology tightens up, so does Gifford's lead.