Electoral Vote Projections

[Kepler]

You would think the trend would be towards hardening of positions, but the latest FOX poll shows a fairly significant increase in undecideds: http://www.pollingreport.com/wh04gen.htm

If true, I think this might be (proto)buyer's remorse on both sides. While historically speaking I would think that was a good sign for the challenger, this time around I dunno -- Kerry has yet to close the deal on the people he needs to win in OH and FL.

[Rimbaud]

You would think the trend would be towards hardening of positions, but the latest FOX poll shows a fairly significant increase in undecideds: http://www.pollingreport.com/wh04gen.htm

If true, I think this might be (proto)buyer's remorse on both sides. While historically speaking I would think that was a good sign for the challenger, this time around I dunno -- Kerry has yet to close the deal on the people he needs to win in OH and FL.
I can guarantee that Fox News won't report on their numbers again because they don't show Bush way ahead. But that's alright, I've come to expect it.

The last two nights I've been called by one polling place, but I just haven't had time to do it with my parents in town, I really want to too, but it just stinks that I can't.

[Mark Laliberte]

I can guarantee that Fox News won't report on their numbers again because ... (some liberal blather)I think it was Fox that put out the poll that showed that race being statisically even in early-September while other polls had Bush up 11-12 percentage points (unless you think that Fox did that to give Kerry too much confidence so he would let his guard down? :confused: )

[Rimbaud]

I think it was Fox that put out the poll that showed that race being statisically even in early-September while other polls had Bush up 11-12 percentage points (unless you think that Fox did that to give Kerry too much confidence so he would let his guard down? :confused: )

This is true, but then when I would see their commentators talk about polls, they would only talk about the Gallup, Times, or Newsweek polls that had Bush up by 10 or more. That's just what I saw a few times and thought it sneaky of them not to use their own poll so as to paint the President in a better light.

[Kepler]

The new Time poll is out, and Kerry is just one point outside the MoE in polls of both registered and likely voters. This may be significant, since Time was the big outlier among the previous polls, having Bush with a huge post-convention lead, and was severely criticized by Zogby and Rasmussen for poor sampling methodology.

http://www.pollingreport.com/wh04gen.htm

[dude]

Here is a link that shows Bush ahead in a couple of different electoral projections.

http://www.dalythoughts.com/ecb.htm

[NCAA watcher]

Interesting that moveon.org has taken the expense to challenge Gallup's poll methodology and motivations:

http://www.moveon.org/content/pdfs/Final-Gallup-Ad.pdf

This ran in today's New York Times.

[dude]

Interesting that moveon.org has taken the expense to challenge Gallup's poll methodology and motivations:

http://www.moveon.org/content/pdfs/Final-Gallup-Ad.pdf

This ran in today's New York Times.

Moveon.org should do just that...MOVE ON! What a bunch of morons. The left has lost it. As soon as they see something they don't like than they have to blame someone, and with this case it's the religious right? Come on, get real. :rolleyes:

[Red Cloud]

The new Time poll is out, and Kerry is just one point outside the MoE in polls of both registered and likely voters. This may be significant, since Time was the big outlier among the previous polls, having Bush with a huge post-convention lead, and was severely criticized by Zogby and Rasmussen for poor sampling methodology.
Wait... so the big Bush lead was not significant and was poor sampling, but Kerry drawing close is the exact opposite?

[ScoobyDoo]

Wait... so the big Bush lead was not significant and was poor sampling, but Kerry drawing close is the exact opposite?
No, they all blow.

Whoever gets the vote out wins.

[Red Cloud]

No, they all blow.

Whoever gets the vote out wins.
w00t.

[Todd]

Wait... so the big Bush lead was not significant and was poor sampling, but Kerry drawing close is the exact opposite?If they've straightened out their faulty methodology, yes. If not, no.

[Red Cloud]

If they've straightened out their faulty methodology, yes. If not, no.What leads us to believe that they have - or for that matter, that it was faulty in the first place?

[Jon]

w00t.


Boo yah!

(Welcome back) :D

[NCAA watcher]

What leads us to believe that they have - or for that matter, that it was faulty in the first place?

It's a basic principle of polling that you're trying to eliminate potential bais in the sample by replicating the population that you are sampling. If your sampling method generates a sample population that is not similar to the general population, then there is a potential bias.

That is where the art, rather than the science comes in to play. By eliminating certain people who you call, you are making a selection of the population. That elimination comes from screening for only "likely voters." The polling org's decision of what question to ask to determine which voters are "likely voters" should be a good proxy for ascertaining those likely to vote. What this ad suggests is that Gallup's method of screening for likely voters has not been too accurate, because it allows more Republicans to be polled than are likely to actually vote (proportinately). I'd be interested to see the exact questions they use to screen the audience, to see how this potential bias is created.

[Milmo]

What this ad suggests is that Gallup's method of screening for likely voters has not been too accurate, because it allows more Republicans to be polled than are likely to actually vote (proportinately). I'd be interested to see the exact questions they use to screen the audience, to see how this potential bias is created.

Not necessary to inspect the entrails of the questions (for other than academic interest) to determine whether the moveon.org ad has validity.

Simply look at Gallup past performances (at http://www.gallup.com/poll/content/default.aspx?ci=1258&pg=1).

Gallup has correctly predicted the winner, with the exceptions of Truman/Dewey '48 and Carter/Ford '76, since 1936.

In recent elections they've been particularly accurate on the % voting for the Republican (off .1%, .3% and .7% in 2000, 1996 and 1992 respectively). So if they have Bush at 52%, history says that's pretty close to reality.

[Jason A]

Most polling organizations do not reveal the exact questions they use to establish who is a likely voter. However, the questions involve past voting behavior (Did you vote for presdient in 2000? Did you vote in the 2002 midterms?), intention to vote in the election (How likely are you to vote: very unlikely, somewhat unlikely, somewhat likely, or very likely?), interest in the election (How closely have you followed the presidential campaign: not very closely, somewhat closely, or very closely?), and knowledge of specific issues or details (Do you know where your polling place is located?).

The answers to these questions are then put into a formula -- for example, voting in 2000 is worth one point, voting in 2002 is worth one point, following this campaign very closely is worth two points, knowing where the polling place is located is worth another two points, and so on -- and if the respondent hits the required number, he or she is considered a likely voter.

The bottom line is that, like Watcher said, this part of political polling is not an exact science. That's why, when you look back on polls after an election, sometimes the RV numbers were more accurate and sometimes the LV numbers were more accurate.

[Patman]

It's a basic principle of polling that you're trying to eliminate potential bais in the sample by replicating the population that you are sampling. If your sampling method generates a sample population that is not similar to the general population, then there is a potential bias.

Define not similar? Do they go out of their way to sample the general population faction by faction or do they target people at random? Do they go after everybody equally or get a set number from each precieved significant subgrouping of people? If either is the case you aren't concerned about bias but rather variation. Bias is if I just decided to get people walking by my office on a pleasant fall afternoon.

[NCAA watcher]

Define not similar? Do they go out of their way to sample the general population faction by faction or do they target people at random? Do they go after everybody equally or get a set number from each precieved significant subgrouping of people? If either is the case you aren't concerned about bias but rather variation. Bias is if I just decided to get people walking by my office on a pleasant fall afternoon.


Well, every process of sampling contains a "potential bias." For example, do you call people? If so, then you can sample only those people with phones, so you do not get people in lower socio-economic areas. Do you call only during the day? If so, you don't get people who work, so you do not get upper socio-economic people. Do you use Mall Interceps?

Organizations can try and control for this potential bias by weighting the results of any survey - i.e., if they sample and get 8% african-americans, but know that the population that votes is actually 13%, then you can "weight" the results that you receive from those 8% accordingly.

So, if you know that you are getting 60% republican responses to your survey, but that republicans only vote constitute 51% of the population, one method of controlling this potential bias resulting from the non-representative nature of your sample is to weight the results.

Gallup does not weight results, as I understand it.

As Jason A points out, the method of asking and "guessing" who are likely voters appears to be the bigger issue. Just guessing, but say, for example, that youngsters 18-22 are more likely to vote democratically. Well, if you base your definition of "likely voter" on those who voted in the last election, you're going to exclude the responses from a group that is more likely to be in favor of Kerry.

While we can debate all of these points, the ad suggests that Gallup regularly misses in its "predictions" of Repub vs. Democratic voting trends. If so, then their refusal to revisit the questions that they use to define a likely voter casts some doubt on the results that they generate.

[Kepler]

If they've straightened out their faulty methodology, yes. If not, no.
Correct.

Of course, "faulty" is in the eyes of the beholder (and in the eyes of an organization that sells its polling data, it is *never* faulty :D ).

As I understand it, organizations control for sampling discrepancies by normalizing results against the relative proportions observed in the real population. A simple example: say the population is registered 40% GOP 40% Dem and 20% Independent. When I sample, I record each respondant's registration, and I wind up with 1 Dem, 2 Republicans, and 2 Independents. When I extrapolate from my sample results to the population, I weight the responses of each respondant as follows:

GOP: 100% weight.
Dem: 40% pop / 20% sample = 200% weight.
Indy: 20% pop / 40% sample = 50% weight.

Polling orgs slice and dice their samples by all sorts of indicators, so it gets convoluted quickly. Also, every time they subsample, they should adjust their MoE to reflect that adjustment (i..e, it should go up). Do they? I doubt it.