Likely voter metric marred by poor polling methodologies

Names in "Comments" field. Erase your name when you're done reading.

The producers of the Gallup Poll are unlikely to read this editorial. At least, one can make that claim using a poorly designed and untenable metric. The producers of the Gallup Poll have been doing just that: for years they have been misrepresenting the facts by in reporting their polls based on a very poorly defined set of ?likely voters.?
Polls themselves can affect elections, and none more than the quadrennial Presidential election poll. In doing so, they must reflect the actual attitudes of voting Americans. Gallup has undertaken so-called ?polls on polls,? wherein they show the importance of public confidence not only in the polling numbers but also in the polling process; they make their predictions based on a randomly selected group of only 1000 adult citizens and claim that since halving the margin of error would mean quadrupling the sample size, larger samples are prohibitively expensive. But they may interview that many anyway, because before they select an interviewee for inclusion in those 1000, Gallup must mark a person as a likely voter.
In this election cycle, however, Gallup?s formula for selecting likely voters has come under attack for waxing conservative and skewing results in favor of the President. Gallup has been predicting better performance by George W. Bush than most other polls, and an examination of its results shows why: Approximately 40 percent of Gallup?s ?likely voters? are Republican, and about 10 percent fewer identify themselves as Democrats.
Gallup determines its set of ?likely voters? based not on party percentages or 2000 election exit polls, but on a questionnaire. Questions such as whether a person has voted in previous elections and interest in politics decide who constitutes the group. Several of these questions ? combined with polling only household telephones ? remove younger, and likely more liberal, Americans from the Gallup polls. Younger voters may not have been voters in the 2000 election ? obviously not those 18?21 ? and they are far more likely to live in a non-land-line, mobile-phone-only household. Other problems arise from questionable interest in politics and the election. The most egregious disparities between registered and likely voters came right after the Republican convention, when interest among Bush supporters surged and their numbers overwhelmed likely voter metrics.
Zogby International has been one of the biggest critics of the Gallup organization in the past several months of the 2004 election cycle. Founder and CEO John Zogby claims that the disparity between the ?likely voter? results and the ?registered voter? results is due to an outdated ?likely voter? model. Zogby is a proponent of weighting results by anticipated party ratios and overall likeliness party to vote. Gallup claims that this cannot be known from census data and refuses to consider using these data in any capacity for their polling, but a compromise between the two may exist.
Gallup can weight its ?registered voters? result, not using census data, but by averaging likely voter questionnaires from months before their actual poll takes place. It would eliminate temporally local spikes while still considering likelihood for any individual actually to enter the booth on November 2.
The Gallup Organization should change their polling practices to reflect reality. Not doing so is likely to harm voter education efforts and unduly influence the election ? naturally, within a margin of error.