Home Store In Memoriam Deborah Newsletter Forum Topics Blogfeed Blogroll Facebook MySpace Contact Us About

Faulty Polls

Reported by Eleanor - September 5, 2004 -

Many of you have shown an interest in discussing the obvious flaws in the polls that are being used by the media to show the status of the two political campaigns. It seems that if we are going to use these polls to demonstrate the general opinion of the American people at a point in time, the people who are polled should be a fair, representative sample of the American people. If the sample is not representative, it serves only one purpose, and that is to present a biased view intended to sway opinion with the use of faulty data.

The Newsweek Poll: Post-Republican Convention, Princeton Survey Research Associates International is the poll we have been discussing. It puts Bush over Kerry by 54-43 percent. See Newsweek Poll.

Final Topline Results

N = 1,008 Registered Voters, 18 and over
Margin of error: plus or minus 4
Interviewing dates: September 2-3, 2004

1,188 Total adults (plus or minus 3)

1,008 Registered voters (plus or minus 4) this is the number actually polled

505 Thursday interviews (plus or minus 5) half Thursday after Zell-Cheney
503 Friday interviews (plus or minus 5) half Friday after Bush
Therefore, the data has varying political influences

374 Republicans (plus or minus 6) this appears about right

303 Democrats (plus or minus 6) this appears to be understated
300 Independents (plus or minus 6) this appears to be overstated

476 Men (plus or minus 5)
532 Women (plus or minus 5)

112 18-29 (plus or minus 10)
349 30-49 (plus or minus 6)
516 50+ (plus or minus 5) this number seems way too high

283 Southern White (plus or minus 7)
559 Non-Southern White (plus or minus 5)
144 Non-White (plus or minus 8) this number seems too low

417 Military households (plus or minus 6) this number is grossly disproportionate
578 Non-military households (plus or minus 5)

310 Republican states (plus or minus 6)
409 Swing states (plus or minus 6)
289 Democratic states (plus or minus 7)

541 Bush/Cheney supporters (plus or minus 5)
403 Kerry/Edward supporters (plus or minus 6)


187 Republicans (plus or minus 8) correct
139 Democrats (plus or minus 9) grossly understated
165 Independents (plus or minus 8) grossly overstated

A national poll must be taken from a representative sample if it is to be used. The whole point of polling is to design a poll that accurately reflects the electorate. What the Newsweek poll did is create a sample of 1008 registered voters which bears little resemblance to the voter demographic for the entire country.

Polling organs go to great care to design samples that are accurate reflections. There are many areas in the US that are considered bellwether areas because they signal views similar to the nation as a whole. We don't know which ones they are, but the people who poll do.

Online polls are not often cited in media since they are known to be unreliable because they are not representative samples. If you called a thousand people at random, the result would all depend on where and when you called.

Pollsters are supposed to pick areas that reflect national standards of political affiliation, racial distribution, and age group, among others. Any poll in which over half the respondents are over 50 had better be designed to find out about the specific needs of seniors, NOT a national election.

Are any of you, or do any of you know, a statistician who might like to work on this? This poll is wrong, but we can't prove it the way a stat person can. Someone who actually does polls can prove it is wrong, someone who can speak with some authority. We need a professor. Sound like anyone you know? You can smell it and tell it is bad, but we need a microbiologist to identify the bacteria.

For comparative analysis, here is a look at the Zogby poll. Notice the care taken to assure a sample that is representative.
The Zogby poll for the same Aug 30 - Sept 2nd time frame showed Bush up 48-46% over Kerry.
Zogby International

Zogby is a superior polling group, and part of the reason is their willingness to share their methodology to show the soundness of their methodology.