Jonathan Chait discusses the “Rasmussen problem”, which basically boils down to the fact that Rasmussen’s polls have become rather extreme outliers, and they all seem to serve a narrative about the country taking a sharp pro-conservative, anti-Obama turn. As Nate Silver has pointed out, Rasmussen’s “house effect” can’t be explained by a likely voter model alone; moreover, the magnitude of Rasmussen’s house effect didn’t begin to show up until after the 2008 elections, which Rasmussen did a pretty good job of pegging. In addition, as Kos noted, Rasmussen of late has largely avoided races as they are being decided. They had nothing in the field for the PA-12 special election, for which there was a widespread perception that the Republican candidate would win, and their last poll of the Coakley/Brown race in MA was a week out. (Chait discussed these items in a followup post.) In other words, their track record in 2008 is of limited use for assessing their current results, because there aren’t many general election results to provide an objective comparison for their apparently new model. (I’ll stipulate they did a good job calling the Texas GOP gubernatorial primary, but polling a primary is not the same as polling a general election. And their overall performance in recent years is mixed.)
You can see the result of their narrative-setting right here in Texas, where their most recent horse race poll instantly became the conventional wisdom about the state of that race. I lost count of how many mainstream media folks cited the poll, but I saw very few attempts to come up with explanations for how a “four point race” (which too was suspicious) suddenly became a “13 point race”. It just was, because that’s what Rasmussen said. And to reaffirm that point, Rasmussen is out there with more polls, telling us that Texans don’t want tax increases to be part of the budget solution and want to see the Affordable Care Act repealed, and if you wait a week there’ll be more where that came from. Rasmussen is often the only outfit polling these questions, yet their track record on “issues polls” is much less reliable than its horse race polling, so once again, what they say gets into the discourse and stays there. I think it’s time for there to be a little more skepticism of what they’re saying.
To put it another way, let me address Paul Burka:
Either this poll or the most recent Rasmussen poll is an outlier. On May 20, Rasmussen had Paul with 59% of the vote and Jack Conway, his Democratic challenger, with 34%. Research 2000 has the race at 44% for Paul, 40% for Conway. Somebody is really wrong here.
Rasmussen now has Rand Paul with a 49-41 lead over Jack Conway, which is not only much more in line with Research 2000 and SurveyUSA, it’s also consistent with its own prior polling of this race. Just scroll down to the bottom of the screen on that Rasmussen link above and see for yourself. The lesson is simple: If you’re wondering who the outlier is in a given set of poll results, the odds are good it’s Rasmussen.
UPDATE: Kos has another example of a Rasmussen outlier that was intended to set a narrative, then later “corrected” once it was shown to be ludicrous. This one was in Connecticut. Thanks to Ed Sills for the catch.
Pingback: One more thing about polling – Off the Kuff
Pingback: maxomai.org » Blog Archive » 2010 Senate Race Handicapping — Part Four (New Hampshire through Ohio)
Pingback: Republicans for White – Off the Kuff