Jennifer Rubin Forces Washington Post to Explain Its Own Bogus Poll
The media were all atwitter Monday over a new Washington Post-ABC News poll finding President Obama eleven points ahead of Republican presidential nominee Mitt Romney in key swing states.
Within 90 minutes of the Post's Jennifer Rubin exposing that the margin of error in the poll was - wait for it! - an astonishing eight points, the paper felt the need to publish a new piece explaining the whole thing.
The Post's Jon Cohen and Dan Balz wrote Sunday:
On the eve of their first presidential debate, President Obama leads or is at parity with Mitt Romney on virtually every major issue and attribute in what remains a competitive general election, according to a new Washington Post-ABC News poll.
The new survey also highlights an emerging dynamic: the disparity between the state of the race nationally and in battleground states, where campaigning and advertising by the two candidates have been most intense and where the election will be decided.
Nationally, the race is unmoved from early September, with 49 percent of likely voters saying they would vote for Obama if the election were held today and 47 percent saying they would vote for Romney. Among all registered voters, Obama is up by a slim five percentage points, nearly identical to his margin in a poll two weeks ago.
But 52 percent of likely voters across swing states side with Obama and 41 percent with Romney in the new national poll, paralleling Obama’s advantages in recent Washington Post polls in Florida, Ohio and Virginia.
An eleven point lead in swing states sounds pretty ominous, doesn't it? Maybe not as Rubin wrote at 2:37 PM Monday:
As I learned from Post pollster Jon Cohen, that finding is based on the responses of a total of 160 people, and it has a margin of error of 8 percentage points. So yes, there may be a difference between swing-state and national numbers, but the gap might be very small or it might be big.
Moreover, the swing-state votes aren’t aggregated. You need to look at each one separately. The presidential contest is a race to 270 electoral votes. So the true battleground states (sorry, that doesn’t mean Michigan or Pennsylvania) are where things will matter.
The Obama team and its dutiful media spinners were emphatic for a week or so that the race had been blown open. Now they will point to the gap between national and swing-state polling. But really how large is that. Look at the Virginia polling. Is the margin there two points or eight points? In Ohio, is the margin one, nine or four points? Take your pick.
This must have caught the eye of someone at the Post, for 87 minutes later, "The WaPo-ABC ‘Swing State’ Poll Numbers, Explained" miraculously appeared (emphasis added):
Recent state polls from other public pollsters in Colorado, Iowa, New Hampshire, Nevada and Wisconsin mainly show the same thing: advantage Obama.
Pulling out respondents in these eight states — all of which Obama won in 2008 — from the new national poll, shows Obama at 52 percent and Romney at 41 among likely voters. We report these numbers to help connect the dots between the clear Obama leads in the collection of states and the continued closer parity nationally, not to suggest pinpoint precision on what’s happening now in any particular state.
Out of 929 registered voters in the new poll, 161 live in one of these eight states, with a margin of sampling error of eight points. The likely voter sample in these states is about nine points, making the 11-point gap an apparent edge. That margin is significant at the 80 percent confidence level, not a standard, conservative 95 percent threshold, which we take as added evidence of the state-of-play in state polling, and the crucial link between those and the national numbers (49 percent for Obama, 47 percent for Romney). No such dissection of a national poll — no matter how many interviews — could be anything more than that. They are simply not designed that way.
I'm sure as the television news networks report on this poll the rest of the day, they will let viewers know about the statistical problems with it.
Or maybe not.
More importantly, why didn't Cohen and Balz report the methodology in their first piece so that readers would have known the margin of error?
Maybe that would have been telling.