Today, the Gallup polling firm was for Barack Obama's lead over John McCain before they were against it.
If that sounds complicated, it's only the beginning. Early Monday, Gallup released the latest of its daily tracking polls, which showed Obama holding an eight point lead over John McCain 48-40. Then, at the close of business, Gallup revealed results of its co-effort with USA Today, in which -- gasp! -- John McCain was shown to have a four-point lead over Obama.
In the latter instance, the metric being evaluated was one near and dear to the hearts of pollsters, the "likely voter." In the earlier poll that showed Obama ahead, Gallup merely surveyed registered voters.
Obama partisans would perhaps point out that the Illinois Democrat's entire campaign is based on drawing new voters -- or "unlikely voters" in the parlance of pollsters -- into in the political process. Many observers have taken the record-breaking turnout from the Democrats' primary season as empirical evidence of an unusual enthusiasm among rank and file voters on that side of the partisan divide.
Which makes investigating the Gallup/USA Today "likely voter" statistics all the more odd. Besides its rare finding of a McCain lead, almost all of the voters deemed "unlikely" to turn out just so happened to be Obama voters.
Emory Univeristy political scientist Alan Abramowitz broke it down for the Huffington Post. Noting that out of the 900 voter sample surveyed by Gallup/USA Today, the pollsters deemed 791 of those individuals to be "likely" ones, and it is their responses which make up the 49-45 figure that immediately got coverage on MSNBC's Hardball.
By contrast, the full 900 person sample of registered voters polled by USA Today showed Obama with a 47-44 lead. So what about those 109 unlikely voters? According to Abramowitz, "among your 109 unlikely voters, according to Gallup, Obama leads McCain by a whopping 61 percent to 7 percent. Putting it another way, according to Gallup 16 percent of registered Obama supporters are unlikely to vote compared with only 2 percent of registered McCain supporters."
Meanwhile, Gallup's independent tracking poll is conducted with an entirely different -- and larger -- sample of 3,000 voters.
And Abramowitz notes that this isn't the first time Gallup has courted controversy in calculating "likely" voters. "Eight years ago the Gallup organization got in hot water for using a likely voter screen several weeks before Election Day that produced wild fluctuations in candidate preference. At one point, the Gallup tracking poll went from an 8 point Gore lead to an 11 point Bush lead in three days. Of course, this was nonsense. The wild swings in the tracking poll were almost entirely caused by the likely voter screen. Those results were not to be believed. And neither are these."
Jeff Jones of the Gallup Poll pushes back on Abramowitz's critique of their joint USA Today poll by noting that the voter model "assumes a 60 percent voter turnout of national adults. The likely voter sample is weighted to match this assumption, so the weighted sample size of likely [voters] is 604."
Therefore, Jones says Obama's lead among "unlikely voters" is merely 51 percent to 21 percent.
Still, that's a huge margin. And it begs the question: How useful is it to try to estimate, 100 days out from the election, which registered voters are likely to vote -- especially when they favor one candidate so dramatically?