Weighting It Out: Party Identification and Election 2004 By Larry Hugick and Stacy DiAngelo

Using party ID figures from the 2000 national exit poll (39 percent Democratic, 35 percent Republican, and 26 percent Independent), they even went as far as taking poll results they didn't like and reweighting them to their party ID assumptions to produce "corrected" horserace figures that were more favorable for their candidate! The main target of these Democratic bloggers was Gallup, whose survey results were often more favorable for Bush than those of their competitors. But at various points these bloggers also went after other national media polls, including the Newsweek and Time polls, the CBS News/New York Times poll, and the ABC News/Washington Post poll.

Gazing at poll results through partisan lenses, these bloggers either didn't know or didn't care that the consensus among experienced nonpartisan media pollsters is that party affiliation is changeable, subject to both short-term and long-term variation, and should not be treated as a fixed characteristic like birth year or gender. What's more, because party affiliation is so closely related with candidate preference, weighting by party ID is just one step away from weighting poll results to presumed "correct" horserace figures.

Had the furor over party ID been limited to the insular echo-chamber of party activists and their websites, it might have been safely ignored by established media pollsters like me. However, the practice of using party ID results to discredit poll results crept into the mainstream media. John Zogby, one of the few nationally known pollsters who embraces the idea of routinely weighting survey results by party ID, wrote a column for The Financial Times questioning the accuracy of post-GOP convention polls by other organizations showing Bush with a double-digit lead, singling out the Newsweek poll as flawed on the basis of its party ID distribution.

Following Zogby's lead, the well-respected British newsweekly, The Economist, repeated his charge against the Newsweek poll:

Having conducted the poll for Newsweek for many years, it was disheartening to see a news organization I respected publish misinformation about our poll in the lead story in the news section of their U.S. edition. Beyond the mistaken notion of static party ID, the figures they cited for the party ID distribution were incorrect. The Economist appears to have appropriated Zogby's erroneous figures, most likely produced by taking the unweighted sample sizes for Republicans, Democrats, and Independents reported in the Newsweek poll topline to calculate a party ID breakdown for the poll. No one at The Economist bothered to call us to confirm the accuracy of these figures. (The correct weighted registered voter figures show a less Republican-heavy distribution of 35 percent Republican, 32 percent Democrat, and 30 percent Independent.)

In addition, The Economist 's reference to current party registration figures for the United States was, at the very least, confusing. Party identification and registration can be quite different, and there is no such thing as national party registration figures. Since many states do not have formal party registration, such information is only available at the state level.

Regrettably, it appears to me that The Economist decided to call another news organization's polling methodology into question for parochial reasons. Their own Economist/YouGov poll, which used an online panel, showed virtually no change in the presidential horserace after the GOP convention, while the Newsweek and Time convention-week polls showed President Bush making significant gains. Like Zogby, The Economist appeared to be using a party ID distribution to raise questions about another poll in order to justify their own poll results.