In a crush of work yesterday, I somehow missed a terrific article by Sharon Begley in the morning’s Wall Street Journal, “Public-Opinion Polls Diverge Because They Are Still Partly an Art.” Sound familiar? Attention reporters: Begley’s piece is absolutely, positively a must read. If you have time for only one polling methodology article, this is the one.
It’s tempting to quote it all – and I’ll certainly quote more in the coming weeks – but for now, here’s how Begley disposes of the party ID weighting debate:
Gallup does not adjust for party self-identification, and neither do many other major polls. Zogby International, however, treats party affiliation, as given by voters in exit polls in 2000 and other recent elections, much like age or sex, increasing the weight of whichever party is undersampled.
But every scientist I asked has grave qualms about that. Party affiliation can change in four years, or even overnight, as Prof. [Cliff] Zukin [professor of public policy at Rutgers University] found in a 2003 study: When people lean toward, say, a Republican, they then tell poll takers they are Republican. If more self-identified Republicans make the cut of “likely voters,” then that reflects that more of the former are likely to vote. [emphasis added]
The Journal has kindly made this article available free of charge. As they say, read it all.
Encouraging News, Finally
I love the fact that more people than ever are registering to vote. The NYTimes had an article today about the Get Out the Votes effort by America Coming Together and Project Vote and others. It reported that much higher
Maybe tracking for party ID doesn’t make sense. However, it’s obviously a problem if, due to chance or some other factor, you happen to survey too many Republicans or Democrats. The current theory is that groups of people that are more likely to vote Democratic (young, hip, or poor) are also more like to ONLY have a cell phone, or to not answer thier main line (caller ID).
HOWEVER, there is an easy, fool-proof way to see this bias. Ask who they voted for four years ago. The Time poll asked this question, and apparently Bush beat Gore by 12 points in 2000 (instead of what really happened-Gore beating Bush by half a point). Not coincidentially, Bush was leading Kerry this time in that survey by about the same margin in that survey, and that survey also showed the “too many Republicans” in the party ID data.
So, if you want to weigh by anything, weigh by that (of course new voters wouldn’t be counted, but they wouldn’t be “Likely Voters” anyways). Since that and party ID show similiar numbers, weighing by party ID might be a good idea.