More on Exit Polls

[gstelmack]

http://www.cnn.com/2005/ALLPOLITICS/...lls/index.html

Good discussion about why they think the exit polls were so wrong. Turns out they really didn't have a random population sample after all. Not intentionally, but it turns out more Kerry supporters were willing to be surveyed than Bush supporters.

[Dutch]

http://story.news.yahoo.com/news?tmp...m/bush_poll_dc

And another Poll. Check out how the story is sold.

Global Poll Shows Negative Reaction to Bush Win

By Michael Holden

LONDON (Reuters)

"This is quite a grim picture for the U.S.," said Steven Kull, director of the Program on International Policy Attitudes at America's University of Maryland.

"Fully one in four British citizens say the Bush re-election has made them more opposed to sending troops to Iraq, resulting in a total of 63 per cent now opposed," said Doug Miller, president of GlobeScan which carried out the poll.

"Our research makes very clear that the re-election of President Bush (news - web sites) has further isolated America from the world."

The survey found that 47 percent of those questioned now see U.S. influence in the world as largely negative.

"Those saying the U.S. itself is having a clearly negative influence in the world still do not constitute a definitive world-wide majority, suggesting there may be some underlying openness to repairing relations with the U.S.," he said.

Steven Kull is anti-Bush in almost every reference found on the internet and routinely has pushed the anti-Bush doctine by implying the UN's failure to act in Iraq was an American mistake.

A quick search finds that GlobeScan is affiliated with UN Global Compact whose main objective is to better promote the goals of the UN over all else.

So much for biased polls parading around on the internet as unbiased polls.

[Glengoyne]

Originally Posted by Dutch

Steven Kull, director of the Program on International Policy Attitudes at America's University of Maryland..

These are the same people that released that poll that said that ridiculous numbers of Bush supporters believed that there were still WMDs in Iraq, that Saddam had been behind the September 11th attacks and such. It was so ridiculously out of line, that I never really considered anyone actually buying into it. But alas it became very popular in Kerry supporters circles. I don't know what kind of sampling this group uses, but I don't buy for a second the results of their polls now, because that one was so far out of the bounds of reality.

[Klinglerware]

Originally Posted by gstelmack

http://www.cnn.com/2005/ALLPOLITICS/...lls/index.html

Good discussion about why they think the exit polls were so wrong. Turns out they really didn't have a random population sample after all. Not intentionally, but it turns out more Kerry supporters were willing to be surveyed than Bush supporters.

It was quite the CYA article on the part of CNN. The disclaimer that CNN did not publish the "inaccurate results" is kind of weak, since none of the other Networks published it either (they're contractually obligated not to do so).

Self-selection bias is nothing new to polling. It's well known that democrats are more likely than republicans to take part in exit polls, and VNS/Edison/Mitofsky adjust for this. They do this by waiting until the actual precinct vote comes in for the sample precinct in question, and reweight the sample accordingly. That's why any early exit poll number that is leaked in the afternoon is likely to be misleading. I don't think that the exit-polls were wrong necessarily, it's just that raw numbers were leaked before they were releasable.

The precinct sample is selected randomly, respondents are selected to insure randomness: the exit poll workers are instructed to approach every 5th voter (or someother predetermined interval) that walks out of the polling place. The issue is that not all voters approached choose to fill out the survey questionnaire--and unfortunately, the "refusers" are not randomly distributed...

[gstelmack]

Originally Posted by Klinglerware

Self-selection bias is nothing new to polling. It's well known that democrats are more likely than republicans to take part in exit polls, and VNS/Edison/Mitofsky adjust for this. They do this by waiting until the actual precinct vote comes in for the sample precinct in question, and reweight the sample accordingly. That's why any early exit poll number that is leaked in the afternoon is likely to be misleading. I don't think that the exit-polls were wrong necessarily, it's just that raw numbers were leaked before they were releasable.

If you have to wait for the actual precinct numbers, then what's the point of the exit poll?

[flere-imsaho]

Harris Poll from 10/21/2004

• 63 percent believe that Iraq, under Saddam Hussein, was a serious threat to U.S. security.
• 62 percent believe that Saddam Hussein had strong links to Al Qaeda (a claim which Vice President Cheney has made more than President Bush).
• 41 percent believe that Saddam Hussein helped plan and support the hijackers who attacked the U.S. on September 11, 2001.
• 37 percent actually believe that several of the hijackers who attacked the U.S. on September 11 were Iraqis.

[Klinglerware]

Originally Posted by gstelmack

If you have to wait for the actual precinct numbers, then what's the point of the exit poll?

It's a misconception that exit polls are only there to predict the winner of an election. It is actually a very small part of the overall statistical model that they use for calling races: looking at a sample of actual precinct votes and the overall actual votes as they come in, actually matter more. Races are only called by the networks with primarily exit-poll data if the spread between the candidates is huge and conforms to prior history (for example, if the Wyoming exit poll numbers are showing Bush with 65% of the vote--that's completely expected).

More valuable is the wealth of demographic and public opinion data that can be gleaned from the survey. The national exit poll survey questionnaire is actually rather extensive and will ask a dizzying array of questions about your demographic background, political views, etc. The final exit poll offers data on why voters voted the way they did, how demographic blocs voted, etc. That information is ultimately more interesting to the researchers, politcal consultants, and party bosses.

[Dutch]

Races are only called by the networks with primarily exit-poll data if the spread between the candidates is huge and conforms to prior history

Like in Florida in 2000 after the Eastern Time Zone polls had closed but the Central Time Zone Panhandle that is roughly 70% Republican was still waiting in line to cast their votes?

Oh wait, that's right the left-leaning bias claimed victory in Florida for Gore and sent their "volunteers" running through the major voting lines to yell, "Gore Won! Gore Won!" to get the Republicans to leave the voting lines.

Networks should not be announcing any exit polling numbers until Hawaii closes it's doors. That's the only fair way to do it.

[Klinglerware]

Originally Posted by Dutch

Like in Florida in 2000 after the Eastern Time Zone polls had closed but the Central Time Zone Panhandle that is roughly 70% Republican was still waiting in line to cast their votes?

Oh wait, that's right the left-leaning bias claimed victory in Florida for Gore and sent their "volunteers" running through the major voting lines to yell, "Gore Won! Gore Won!" to get the Republicans to leave the voting lines.

Networks should not be announcing any exit polling numbers until Hawaii closes it's doors. That's the only fair way to do it.

As I mentioned earlier--how exit polls are actually used in the statistical projection of races is widely misunderstood by the general public (and this is the networks' fault for a variey of reasons). Exit polls are just one of the data streams in the model, and will rarely be used on it's own. The problem with Florida 2000 was that all three major streams had problems (including the tabulated vote total), the model is only as good as the data--normally, the model will "self-correct" as the day proceeds (good streams of data will over-ride bad streams), but when all data sources had problems, the model broke down.

[JonInMiddleGA]

Ah yes, good ol' Harris polls ... which I must have participated in at least 100 of in the past year. Nothing like the purely random nature of the internet I always say

I'm not saying that's their only methodology, I'm just pointing out the possibility that their sampling methods sometimes carry some inherent flaws.

[flere-imsaho]

I find that polls always contain inherent flaws when they disagree with my preconceived notions.

[gstelmack]

Originally Posted by flere-imsaho

I find that polls always contain inherent flaws when they disagree with my preconceived notions.

If you stopped after "inherent flaws", I'd agree with you. Fact is I don't trust them no matter what they try to spin. The results can be spun so many ways, their results are essentially useless. And it's usually too difficult to get information on how the polling was done (or you find out too late) to determine if there is even a shred of credibility in the results.

Heck, we can have everyone in the country vote on an issue and there is STILL disagreement over the final results of the poll...

[flere-imsaho]

Well, I was making a joke....

I agree, though. For me, polls are basically worthless, on a personal level. I don't give much credence to them.

[Klinglerware]

Originally Posted by gstelmack

Heck, we can have everyone in the country vote on an issue and there is STILL disagreement over the final results of the poll...

To that point, exit poll results can be useful as a check in elections where the vote counting mechanism is not to be trusted. Exit polling was at least partially responsible for Marcos' ouster in the 1980s, and the overturning of election results in the Ukraine a few months ago.

Survey research is often abused and the methodology behind them can be questionable depending on the poll. But so long as they are done right, and people recognize their limitations, they can serve as useful tools in a variety of fields (politics, marketing, etc.).