News Archive
10:23 am
Wed May 28, 2008

Interpreting Political Polls

Lexington, KY – How to sort out conflicting polls from Rasmussen Reports and Sen. Mitch McConnell's campaign? Remember error margin and methodology.

Rasmussen's result was Lunsford 49, McConnell 44. Applying the 4.5-point error margin, Lunsford's number could be as low as 44.5 and McConnell's could be as high as 48.5.

Van Lohuizen's result was McConnell 50, Lunsford 39. Applying its 4-point margin, McConnell's number could be as low as 46 and Lunsford's as high as 43, pretty close to his low in the Rasmussen poll.

And remember that error margins are based on a 95 percent confidence level. In other words, 1 out of 20 polls will be wrong outside the margin. One of these could be.

The principal cause of the difference is probably methodology. Rasmussen polls use recordings, not live interviewers, used by traditional pollsters such as van Lohuizen. For many years, I generally avoided reporting such robo polls because the methodology was not proven, but Survey USA, which uses recorded questions, has racked up a good performance record in this state and others. However, Rasmussen is less familiar to us, and many political reporters in Washington lack confidence in his polls. One, Stu Rothenberg, was quoted by the Herald-Leader on its Pol Watchers blog. However, as Joe Gerth reported on The Arena blog for The Courier-Journal, Slate magazine and the Wall Street Journal have said that Rasmussen has been among the most accurate national polls in recent elections.

There are other considerations. I think robo polls and their bang-bang approach tend to elicit more gut reactions, probably more so two days after a high-turnout primary. The questionnaires I have seen from Jan van Lohuizen tend to ease respondents into the substantive questions, making their responses more reflective, which I think would be to McConnell's advantage. I have asked the McConnell campaign for the text of its questionnaire, up to and including the last question for which a result is provided. That would also reveal how the poll determined likely voters, which could also account for differences in the two polls. I have asked Rasmussen for his full questionnaire. But in the meantime, I stand by my original conclusion, that his poll shows the race to be competitive.