Tuesday, September 28, 2004

Something wierd at Gallup
When most polling companies poll people, they ask their opinion questions, then ask some demographic questions, and finally correct the opinion results according to demographics. This is called weighting the results. If they poll too many men they will multiply the women’s results by a known fraction to make the percentages reflect the real division of the voting population. These assumptions about who will vote are one of the places where error can creep into the results of a poll. They might assume that only a small number of young people will vote (which is normal) and find out on election-day that some issue like the draft motivated them to turn out in far larger numbers than usual.

Each company has its own formula for weighting. Gallup is one of the companies that weights for party identification. Others, like Zogby, do not. Those that do assume that the electorate is x% Republican, y% Democratic, and z% independent or other and make the same kind of correction as they do for sex, race, region, or age. Party identification is tricky. Because states have different rules on registration, asking someone how they are registered and what party they consider themselves to be can bring two different results. For example, the latter produces more independents than the former.

The recent polls by Gallup have jacked up the number for Republican likely voters, thus giving Bush a larger lead than most other polls. The most recent poll seems to have jacked the number up again. No one is sure why. Bloggers have been talking about this for a few weeks now. MoveOn has an ad up about it. It seems the mainstream press is finally on the story.

No comments: