Bush's lead GONE; Kerry leads 49-46 now

[Ramo]

Since Zogby's probably the best pollster out there (and no, he's not a Democratic outlier), his numbers are probably the most accurate - Kerry and Bush are essentially tied, possibly with Kerry behind a little, but with tons of momentum. As I've been saying, the real story is the internals of all of these polls, which show a great deal of momentum for Kerry.

[Jon Miller]

you do use averaging to make random uncertainties be less significant

used in any real science (chemistry, physics, biology, ...)

and of course averaging does nothing to reduce statistical error, but I don't think they claim it does

JM

[Shi Huangdi]

Polling that generates a random sample of a sufficent size should generally be accurate. Any one poll could be off, but by averaging them all of these polls together we can dilute that.

If all of the polls were horribly wrong, then of course averaging wouldn't do you any good, but then you wouldn't get anything from looking at a specific poll either. If what they are doing is mostly right however, averaging should take care of anything weird happening or some incorrect methodology in any one poll.

[Edan]

Originally posted by Shi Huangdi
Polling that generates a random sample of a sufficent size should generally be accurate. Any one poll could be off, but by averaging them all of these polls together we can dilute that.

It dilutes them, yeah, but it assures that there's some error and bias introduced by those polls that are off due to statistical error or bias. You also need all the internals to make sure that the data can be combined, and hope that the changes in methodology and timing aren't important enough to matter (averaging a poll taken right before and after the debate would clearly be a bad idea)

Personally, I'd rather look at the half a dozen or so polls individually and take them for what they are.

[Drachasor]

Originally posted by Jon Miller
you do use averaging to make random uncertainties be less significant

used in any real science (chemistry, physics, biology, ...)

and of course averaging does nothing to reduce statistical error, but I don't think they claim it does

JM

You average results from the same experiment, and that reduces random error, not bias.

Averaging random polls together of unknown biases, various sizes, various polling methods/questions, done at different times is not the same thing by any means. Besides, the very use of statistic themselves take into account random error, and indicates it with the standard deviation and chance of error.

This averaging of polls garbage is pure pseudo-science.

-Drachasor

[Jon Miller]

if biases are randomly distributed (which is not the caste, but could be a first order model of them) than averaging polls of unknown biases will give a better response than not taking that into account

same with other unknowns (polling methods/questions, etc)

Jon Miller

[Bosh]

Well as far as Zogby being a democratic outlier, he definately was in 2000. Virtually every poll conducted in the last days of the campaign predicted a Bush popular vote victory, very very few predicted a tie or a Gore popular vote victory and Zogby was one of the correct handful (forget if he predicted a popular vote tie or Gore victory). So last time there was what seems like a systematic pro-Bush bias across the board pretty much. Is it happening this time? Who the hell knows? Its hard as hell for pollsters to get a random sample of the voting population and I'm sure there's huge methodological flaws and difficulties and its hard to hell to tell which way these systematic biases lean, but I'm guessing that reality is a bit more pro-Kerry than the polls, simply because the Dems are doing more voter registration and reality was more pro-Gore than the polls last time around.

[Ramo]

Yep. The Dem field operation is pretty much unprecedented. Also, undecideds very rarely break for the incumbent. Basically, if Bush gets less than 49% in the polls, he's gonna lose.

[Lefty Scaevola]

47-47 in CBS-NYT poll
51(bush)-46 in ABC-WP poll

[Shi Huangdi]

Originally posted by Ramo
Since Zogby's probably the best pollster out there (and no, he's not a Democratic outlier), his numbers are probably the most accurate - Kerry and Bush are essentially tied, possibly with Kerry behind a little, but with tons of momentum. As I've been saying, the real story is the internals of all of these polls, which show a great deal of momentum for Kerry.

http://cellar.org/archive/index.php/t-2599.html

and http://www.salon.com/news/feature/20.../index_np.html

Has more on the problems with Zogby.

You keep on saying the internals were the real story. You made big noise about how Kerry had a great DNC bounce based on the internals and that it would soon materialized into a big Kerry lead. It never happened. Bush still nonetheless maintains a significant lead on terrorism and Iraq, what I suspect may have happened in large part is people who were already voting for Kerry are now more confident in his ability on terrorism and Iraq. That it was debates typically do in any case- mobilize and make confident the base in their own candidate.

[Ramo]

Was looking at the first link and noticed two things:

It measures the accuracy of the poll only through the spread, as opposed to the final numbers. For instance, in the Arkansas race, the final Zogby poll had Pryor at 56% and Hutchinson at 43%, and the final Gallup poll had Pryor at 51% and hutchinson at 43%. While the final results have Pryor at 54% and Hutchinson at 46% (http://www.cnn.com/ELECTION/2002/pag...ate/index.html). So, while Gallup got the more accurate spread, Zogby's Pryor number was actually closer to Gallup's Pryor number (and they have the same Hutchinson number).

Furthermore, if you look at Zogby's numbers for the same time period as Gallup's numbers, they had 55% for Pryor and 44% for Hutchinson, which was much closer than Gallup's numbers in the actual distribution.

You keep on saying the internals were the real story. You made big noise about how Kerry had a great DNC bounce based on the internals and that it would soon materialized into a big Kerry lead. It never happened. Bush still nonetheless maintains a significant lead on terrorism and Iraq, what I suspect may have happened in large part is people who were already voting for Kerry are now more confident in his ability on terrorism and Iraq. That it was debates typically do in any case- mobilize and make confident the base in their own candidate.

That's because the internals show the instantaneous trend of the race. Circumstances can derail these trends. In particular, the SBVT very, very effectively blunted the gains in leadership, national security, and foreign policy Kerry acquired from the convention. So, what I'm saying WRT internals isn't a sure thing: if Kerry ****s up the next debate, for instance, the trend could be derailed. However, the race is currently trending very heavily towards Kerry.

[Ramo]

Read the Salon article, it was mentioning Zogby's Battleground Poll, which uses the internet, so it could get young people who often don't have land lines. This isn't the Battleground Poll, rather it uses Zogby's traditionally very accurate methodology.

[MikeH]

These polls seem to be using terribly few people as a sample. Our election polls are generally a few thousand minimum.

Anyway, percentage vote is irrelevant aren't they? It's what states they are in.

[Edan]

Originally posted by Edan


In Ohio and Florida, at least, according to a sampling by the NY Times.

Looks like Nevada, too:

June Registration Figures:

Democrats: 354,950
Republicans: 363,463

August Registration Figures:

Democrats: 383,651
Republicans: 382,630