Reading between the (poll) lines

Why Rasmussen's numbers are different

This article originally appeared on CommonWealthMagazine.org.

Over the last month or so, we have seen a steady stream of new polling data in the Massachusetts gubernatorial election. In reading these polls, there seem to be two different assessments of the election. Rasmussen shows Patrick with a 47-42 lead over Baker, with Cahill drawing a scant 6 percent, his support having fallen recently. Other pollsters show Patrick and Baker each holding just about steady in the mid to high thirties, with Cahill stable in the low double digits (see graphics). What is going on here?

All pollsters except Rasmussen

Rasmussen Polls Only

The short answer is that Rasmussen includes a group that he calls "leaners" in his current horserace calculation. I add the quotation marks because his definition of leaners is unconventional. While most, if not all, pollsters include leaners in their polling data, the usual definition is different from Rasmussen’s. To clarify these differences, I obtained confirmation from Scott Rasmussen himself that my understanding of his methodology is correct.

A conventional definition of “leaner” can be found on NORC's website.

"A survey respondent who does not make a choice among alternatives in an initial question, but makes a choice once asked if he or she leans toward one of the alternatives."

In Rasmussen's case, he includes another group in his definition of leaners. This extra group of leaners includes those who initially say they are voting for someone other than the Democrat or Republican, but who are uncertain of their vote. This includes both those who say they are voting for “some other candidate” as well as those who say they are planning to vote for Tim Cahill. So, for example, if a respondent says she is voting for Tim Cahill in the initial horserace question, but that she is uncertain of her choice, she will get a followup “leaner” question.

Respondents will not, however, get this same "leaners" question if they say they are supporting Patrick or Baker in the initial horserace question. For this reason, when compared to the basic horserace, Cahill's numbers drop when "leaners" are included and Patrick’s and Baker’s numbers rise. This calculation accounts for why both Baker and Patrick appear to be drawing more support overall when compared to other pollsters.

Cahill's recent downward trend in Rasmussen’s data is also an artifact of another aspect of this methodology. Rasmussen reports the race without "leaners" before Labor Day and "with leaners" after Labor Day. When this change was made last month, Cahill’s support appeared to drop suddenly, whereas other pollsters are showing Cahill’s support having leveled out recently.

It may turn out that Rasmussen is correct in pegging the final support that will go to Tim Cahill. As Rasmussen writes, their approach toward leaners "anticipates that support for third party candidates typically declines as Election Day nears and voters begin to gravitate toward one of the major party nominees." Perhaps he is right.

For now, however, this difference in methodology makes it difficult to compare Rasmussen's polls with other those of pollsters.