August 5, 2012 Leave a comment
Very inside baseball, but for the political junkie poll-watching types (hmmm, maybe sound familiar to anybody who reads or writes this blog?) I thought this was a very interesting post by TNR’s Nate Cohn about how the changing composition of the Real Clear Politics polling average is likely driving apparent changes in relative poll standing more than actual changes in voter opinion:
Take the RCP average, which only considers the last two or three weeks of polling. As a result, only a handful of polls are usually represented. The decision to privilege timeliness over comprehensiveness is entirely justifiable, since a shorter window will pick up shifts in the race more quickly than a less sensitive long-term average. But so far in 2012, there haven’t been genuine shifts in the race—so poll-watchers haven’t yet reaped the benefits of responsive, narrow time-frame polling averages (although they will after the conventions). Instead, the polling averages are shifting with the changing composition of the pollsters included in the average—and their house effects.
If all polls were released at similar intervals, this wouldn’t create systemic bias. But Gallup and Rasmussen release polls every day, so they’re always in the RCP average. So, even though Gallup and Rasmussen are just two of 20 or so national pollsters, they usually represent anywhere from one-third to one-sixth of the RCP average. In contrast, Pew Research, which consistently shows Obama leading by 4-to-7 points, generally releases polls once a month. And for good measure, they produce huge sample sizes, accrued over long periods, so they’re kicked out of the polling average sooner than they would if they surveyed fewer individuals over one weekend…
This is a relatively minor issue, but a 1-to-1.5 point difference in the RCP average affects perceptions of the election. Analysts often give the impression of a deadlocked race, but while the race is close, Obama actually has a clear and narrow lead.
On a quasi-related note, Nate Cohn has bee producing lots of top-notch analysis of the election yet I never see any of the usual suspects link to him. They should. (Also just discovered his latest post is noticing this exact problem in the latest pollster.com averages).