Saturday, September 5, 2009

Greg Dworkin (who posts as DemFromCT at Daily Kos) was kind enough to send me an e-mail, notifying me that he mentioned the Health Care Polls blog during a panel on polling at the recent Netroots Nation conference and providing a link to a video of the session. The panel consisted of a real all-star line-up of survey experts from the academic, polling, statistical, and political-writing fields. I watched the video last night and found all the speakers to be lively and informative.

In particular, I would recommend the presentation by Pollster.com's Mark Blumenthal late in the 88-minute session. Regular visitors to Health Care Polls know that I sometimes dwell on how Rasmussen Reports sometimes generates findings that are out of step with those of most other pollsters. Using the methodologically "old school" CBS/New York Times poll as a comparison, Blumenthal reviews Rasmussen's procedures at several key steps of the polling process.

From the information Blumenthal presents, a reader might draw the conclusion that Rasmussen's polling methods are aimed at quickness and expediency (e.g., the CBS/NYT poll makes up to four call-back attempts to reach a household where no one answers the phone, whereas Rasmussen makes none). Rasmussen, it should be acknowledged, does not let sources of potential unrepresentativeness in his samples (as from the no call-back policy missing highly mobile individuals) go completely unaddressed; he weights his samples at the end to bring them into sync with known demographic parameters (e.g., from the U.S. Census). Other pollsters weight (or post-stratify) at the end, as well, but it appears that they take more direct action during the stages of data collection to maintain population representativeness than does Rasmussen.

No comments: