"Unskewed" Gallup
A while back I accepted that Gallup has about a 3-4% error in their base sampling—not a likely voter error, but a registered voter error.
Their "all adults" polling seems reasonable. Perhaps even a little Obama-friendly. But their registered voters don't seem to be everybody else's registered voters. And likely voter errors are derived from the collection of registerd voters.
So I read today's Gallup as being
Obama Approval 53%-42%. (What they report)
Registered voters —Obama 52% - 43% (What they report with +4 adjustment for Obama)
Likely Voters —Obama 50%-Romney 47% (What they report with +3 adjustment for Obama)
I do not present this "unskewing" as fact. I have little respect for "unskewing" in general. But Gallup has a solid method so any built-in deviation is likely to poll "true." That whatever the error is, in absolute numbers, that they are a valid trend indicator, once we realize that they are by far the slowest tracking poll to reflect reality. (Being a 7-day average, versus 3 or 4 day)
So if I had to extrapolate reality from Gallup that is my sense of how it would look.
