For a better browsing experience, please upgrade your browser.

FiveThirtyEight

Politics

I don’t quite know what to make of the new Rasmussen poll showing Rand Paul 25 (!) points ahead of fellow primary winner Jack Conway in Kentucky.

Paul has had a bad 24 hours in the Beltway media environment. Do I think that has any relevance whatsoever for Kentucky, a state which is culturally about as far removed from the Beltway as you can get? No, not really; it may even be a contrarian indicator. Nor is it unusual to see candidates get bounces after primary victories, although they often prove to be fleeting.

But. Rasmussen had shown a particularly large house effect in this race. Whereas the Pollster.com trendline of all non-Rasmussen polls had shown Paul ahead by just 1 point, Rasmussen’s trendline had him up by 15, even prior to this poll being released.

There are two reasons why this might be the case. First, Rasmussen relies pretty heavily on weighting by party identification. It’s a necessary evil if you poll like Rasmussen does, which means flash polls that are literally in the field for just 3-4 hours over the course of a single evening, with no callbacks, no cellphones, and no respondent selection. That probably works fine in many states, where party identification does a reliable job of predicting the vote. But it might not work so well in Kentucky, where you have a lot of registered Democrats who often vote Republican in federal elections. If you’re getting a non-representative sample of Democrats in your poll for any reason, you could get pretty tripped up.

Second, one of the candidates in this race, Rand Paul, generates unusual enthusiasm from certain types of voters, and it may be that these are the sorts of voters who are more inclined to answer a pollster’s phone calls. We’ve discussed before how Rasmussen’s polling seems to be especially favorable to tea-party-type candidates. The makeup of the Tea Party is somewhat amorphous and therefore it may be hard to correct for self-selection bias by means of ordinary demographic weighting.

Finally, if either or both of these problems are impacting Rasmussen’s polling in the state, they’re liable to be compounded in a poll conducted less than 24 hours after a primary that received an incredible amount of national media attention. If the Rand Paul people are somewhat more fired up (i.e. self-selecting) than usual on a normal day, they’re going to have been really fired up yesterday.

We’ve criticized Rasmussen quite a bit this year. To be fair to those guys, they poll so dang often that they necessarily make for a convenient target; release 20 polls a week, and you’re necessarily going to have one or two outliers.

At the same time, there are things that Rasmussen could do to quell the criticism from here and other quarters. For one thing, they could engage in a more earnest dialog about their methodology. For instance, they’ve said that their use of a likely voter model is the reason for their house effect, but a more careful examination reveals that it doesn’t suffice to explain the discrepancy.

And, as Markos Moulitsas pointed out yesterday, Rasmussen has engaged in a strange disappearing act this year when it was actually time to put their necks on the line and poll races close to election day. Last week, they gave us polls on thrilling contests such as the Kansas gubernatorial race and the Idaho senate race, and revealed to us that 51 percent of Americans think that the United States is the “last, best hope for mankind”. But the only poll we got of an actual, flesh-and-blood election was in Pennsylvania, and it was nearly two weeks out of date by the time that Tuesday came around. They ignored the primaries in Arkansas and Kentucky. They also ignored the special election in PA-12, although they’ve never done much polling of individual Congressional Districts before, probably because there’s no reliable way to establish party weighting targets when you go that micro.

None of which tells us anything about Kentucky per se. But it’s always good to get a second (and ideally at least a third) opinion about a race before coming to any conclusion about it. That’s probably more true when Rasmussen is the first man in than with most other pollsters.

Filed under , , , ,

comments Add Comment

Powered by WordPress.com VIP