When it came to polling on "American Idol" last week, Rasmussen Reports was bold.
On Tuesday, it released a poll conducted over the previous weekend (Friday to Monday) in which adults who watch the televised talent show expressed a slight and not quite statistically significant preference for Crystal Bowersox over Lee DeWyze (32 percent to 29 percent).
The fact that viewers chose DeWyze 24 hours later is not the issue here (though a major disappointment in our household). The actual vote, like Rasmussen's result, was reportedly close, and the poll sampled no teenagers and made no effort to replicate the Fox show's vote-as-many-times-as-you-like system.
What is striking is why Rasmussen has not been quite so bold lately in conducting last-minute polls in closely watched political contests.
In 2008, Rasmussen routinely polled through the final weekend and sometimes on the eve of the election. But in the New Jersey and Virginia governor's races last November, in the Massachusetts special election in January and in hotly contested primary elections this year in Texas and Pennsylvania, the firm fielded its final polls at least five to seven days before votes were cast.
In other ways, however, Rasmussen has been far more prolific. When I tallied polls conducted so far this cycle in February, Rasmussen had fielded 45 polls on general election contests for Senate or governor, making up more than a quarter (28 percent) of all the public polls on such contests that we tracked at Pollster.com.
The contrast left some observers smelling something fishy. Liberal blogger Markos Moulitsas fumed that Rasmussen was mostly concerned with "setting the narrative that Democrats are doomed." He noted that two weeks ago Rasmussen released surveys on contests in Colorado, California and Idaho, yet fielded nothing on the Pennsylvania Senate primary or House special election.
"Why take the risk of getting an actual election result wrong, this early in the cycle?" Moulitsas asked. "There's a special election in PA-12? Who cares! Rasmussen is nowhere to be found. Just like in January, when Rasmussen -- who had polled the Massachusetts Senate special election twice earlier, decided to pull out of the race two weeks before the actual election."
So I put this question to Scott Rasmussen via e-mail: Just before the general election in November 2008, you polled on Sunday night in six crucial states -- Colorado, Florida, Missouri, North Carolina, Ohio and Virginia -- and your national tracking poll ran through Monday night. Yet you've stopped polling earlier in many recent contests. What gives?
"We have always and will again this year poll general election races right up to the end," he answered. "We are polling every Senate and governor's race in the nation and cover most about once a month." He added that the pace in the close races will quicken in the fall, and that they will continue to poll "up to the Sunday before the election" on those races generating strong interest, "just as we did in 2004, 2006, and 2008."
But primary elections? "Primaries have always been different for us, and we don't cover them as much. We never have, and I can't imagine we ever will."
The firm did not poll on the Lincoln-Halter Democratic primary in Arkansas, he said, "because the general election polling suggests whoever wins that race will have a tough time winning in the fall. That run-off may be of intense interest to some on the political left, but it is unlikely to have a larger impact."
He added that they covered 2010 primaries for governor in Texas and Senate in Pennsylvania "more than any other firm" and expressed pride that his firm's polling of both races "turned out to be quite accurate."
That puzzled me. By this time in the 2008 cycle, Rasmussen Reports had polled right up to Monday night before the New Hampshire and New Jersey presidential primaries, up until Sunday night in Ohio, Texas and Pennsylvania, and through Saturday night in California and Georgia. Yet for the Pennsylvania and Texas primaries this year, Rasmussen fielded its last poll more than a week before the election.
So I e-mailed again. His response: "In 2008, you're probably talking presidential primaries. People care about those a lot more than Senate or governor's primaries."
Rasmussen also rejects the suggestion that his firm has been excessively cautious of late: "We routinely put out information that challenges the conventional wisdom in Washington and then have it confirmed by other polls." Recounting early polls on the Arizona immigration law and the Massachusetts Senate race, he adds: "If we were being cautious, we would let others go first."
Count me as agnostic on the question of whether Rasmussen or other media pollsters should field surveys at the last moment. As a campaign pollster, I almost always finished polls for my clients on the Thursday night before the election, because after that no critical strategic decisions remained to be made.
It's understandable why media pollsters who are confident in their methods and intent on proving their accuracy would want to poll late. A pre-election forecast is one of the few opportunities pollsters have to check the validity of their results. But all things considered, I would rather see Rasmussen be more transparent about its methods than survey every contest up until the last moment.
Political junkies like me always want another fix. The tougher question is: Do media polls serve any larger purpose by polling late?