Florida: Gravis Marketing - Romney 48% Obama 47%

[Vosem]

This poll has a lot of things wrong with it, though it's actual result -- Romney+1 -- seems like the 'right' answer, though I'd wait for more polling from PPP/Rasmussen/Quinnipiac/someone else well-known.

[Kalimantan]

One more point: I don't have it on me, but in grad school I did a short paper on party ID and two-party vote share. It's lost to history, my old college email server, and a fried laptop, but the basic results: (and you can run the numbers yourself using Roper's exit poll data, which dates back to 1976)

There is virtually no relationship between party ID and that party's presidential vote share. In fact, for the elections 1976-2004 there is actually a very slight negative correlation for both parties. A hell of a lot more people claimed to be Democratic than actually voted for Carter in both 1976 and 1980. If you remove those two elections and run only 1984-2004 (the post Solid South era), you get a positive correlation, but one that's so weak (below .25) that it's basically equivocal.

The correlation for the differences between party IDs and presidential vote shares is a little better - but still weak. (And it gets demolished if you add 1980 and 1976, of course).

None of this includes 2008, because that election hadn't occurred when I ran the numbers. But the bottom line is that the relationship basically doesn't exist - the 1984, 1988, and 1992 electorates were all 38D-35R, for example.

"Self-professed party ID is very volatile, and a dependent variable of voter intention."

lol no and certainly not today.
 
"In other words, if you think that topline results are due to party ID, you've got it backwards."

Yes I think and polls for now confirm that.

Guy, I'm on this forum since 2003 and I can say to you that the party id is the base of the polling. Each day it's confirmed. Look at the last washington post poll with a D +9 sample, look at the ppp polls each week: there is a very strong correlation between the result of obama and the difference between democrats and republicans: greater the gap is, greater the obama result is.

And if Ras gives better results to Romney, it's due to the fact that the numbers of republicans is more important than in others polls.

And I can continue again and again... it's quite logical: Obama will do better in a sample with 40 % of democrats than 30 %...

But what you say backs up what Craigo said, that Self-professed party ID is a dependent variable of voter intention. i.e. if a person wants to vote Romney they self-identify as republican, and vice-versa. You've just got the causal factors the wrong way round

So if more people support Obama, you'll have more self-identified democrats in the sample than if the same people preferred Romney

[DrScholl]

For reference, here's the Gravis poll of the PA senate race, where they have Tom Smith ahead of Bill Casey, 47-28:

http://www.scribd.com/doc/105306657/PA-Report-1

That speaks volumes about how reliable their polling is.

[Eraserhead]

One more point: I don't have it on me, but in grad school I did a short paper on party ID and two-party vote share. It's lost to history, my old college email server, and a fried laptop, but the basic results: (and you can run the numbers yourself using Roper's exit poll data, which dates back to 1976)

There is virtually no relationship between party ID and that party's presidential vote share. In fact, for the elections 1976-2004 there is actually a very slight negative correlation for both parties. A hell of a lot more people claimed to be Democratic than actually voted for Carter in both 1976 and 1980. If you remove those two elections and run only 1984-2004 (the post Solid South era), you get a positive correlation, but one that's so weak (below .25) that it's basically equivocal.

The correlation for the differences between party IDs and presidential vote shares is a little better - but still weak. (And it gets demolished if you add 1980 and 1976, of course).

None of this includes 2008, because that election hadn't occurred when I ran the numbers. But the bottom line is that the relationship basically doesn't exist - the 1984, 1988, and 1992 electorates were all 38D-35R, for example.

"Self-professed party ID is very volatile, and a dependent variable of voter intention."

lol no and certainly not today.
 
"In other words, if you think that topline results are due to party ID, you've got it backwards."

Yes I think and polls for now confirm that.

Guy, I'm on this forum since 2003 and I can say to you that the party id is the base of the polling. Each day it's confirmed. Look at the last washington post poll with a D +9 sample, look at the ppp polls each week: there is a very strong correlation between the result of obama and the difference between democrats and republicans: greater the gap is, greater the obama result is.

And if Ras gives better results to Romney, it's due to the fact that the numbers of republicans is more important than in others polls.

And I can continue again and again... it's quite logical: Obama will do better in a sample with 40 % of democrats than 30 %...

So your response to his detailed, evidence based argument is essentially "lol no, I've been on this forum for a long time, so I know how polls work". Wow, very compelling.

Way too much is being made of the whole Party ID thing by both sides. Of course, Republicans are making a bigger stink out of it this time because, well, their guy is losing...