New CNN poll - more manufactured Obama support

There was a new CNN/ORC poll released yesterday afternoon, and it has President Obama up by 6 points over Governor Romney, 52-46.  But there are some curious numbers in it.  Such as how it's possible that Governor Romney could be trailing by six points in a survey that finds him with a 14-point lead among independents.

Since they don't include most the raw numbers, and every respondent did not necessarily answer every question, there's some guesswork required, but there's a lot of information, and it doesn't all add up.

To start with, they give us the following information about their sample:

Sample information

1022	Adults
875	Registered voters
709	Likely voters

In 2008, according to the census bureau, about 64% percent of the voting-age population voted, as opposed to the 69% of this survey tabbed as "likely."  Other sources suggest that 61% is closer to correct.  Given that those who are not registered and not at all likely to vote are also less likely to take the time to respond to the survey, their likely voter screen, whatever it is, is producing a  number of "likelys" that might be a little high but isn't too unreasonable.

So they ask the question:  Obama or Romney?  And they tell us, in question 1, that the voters responses are:

Ballot test

	Likely	Registered
Obama/Biden	52	53
Romney/Ryan	41	45
Neither	2	2
Other	*	*
No opinion	1	1

So far, so good.  No, that doesn't add up to 100, but there are rounding errors, so, close enough. And, as one would expect, Obama's lead is bigger among registered voters than among likely voters (though not as much bigger as one might expect.)

Next, they ask Obama voters whether they are voting for Obama or against Romney, and then ask the converse of Romney voters.  Here's where it starts to get interesting. In the preface to the questions (pg 3), they reveal how many people chose each candidate in question 1.

Respondents

	Likely	Registered
Obama voters	351	441
Romney voters	340	409
Total	691	850

Here's the first problem.  A serious problem.  The 691 voters for Obama and Romney combined is about 97.5% of their advertised 709 likely voters, which is what we'd expect.  But they're touting a 6-point Obama lead; when they give the number of respondents who chose each of the candidates, what we see is an 11-voter lead for Obama, (351-340) out of 691.  11/691 = ~1.5%, not 6%. 

Did their sample require that much manipulation?  On page 5, we see that they found 441 registered Democrats and 397 registered Republicans.  That breaks down to a 52.6-47.4 advantage for the Democrats in the two-party sample.  That's smaller than the 55-45 advantage in the two-party electorate that they had in 2008 (39-32), and maybe the pollsters are trying to match that.  On the other hand, that was a best-case result for Democrats (2004 was 50-50 [37-37]), and there's no reason to expect them to match it in 2012.  Obviously, there's a science, and an art, to coming up with the best analysis of the raw numbers, but this looks like a gratuitous thumb on the scale somewhere.



I mentioned the other anomaly at the top.  On page 21, they've broken the vote down by demographics.

By Party Affiliation

	Total	Democrat	Independent	Republican
Obama/Biden	52	97	40	2
Romney/Ryan	46	3	54	96
Other	*	*	*	*
Neither	2	*	4	2
No_opinion	1	*	2	*

Obama has a commanding lead among Democrats. To be expected. Romney has a commanding lead among Republicans. To be expected.

But what about those unaffiliated with either party? Where much of the Presidential contest is actually "fought"? Romney has a very large lead there, too, 14 points, 54-40. If that's the case, how can he be trailing by six?

Well, 441 registered Democrats plus 397 registered Republicans equals 838 register voters out of their  875 registered voter survey.  838/875 = 96%.  Given that Republicans and Democrats have made up 70-75% of the voting electorate in the past few Presidential elections, and that they make up 96% of the registered voters in this sample, it seems that they have undersampled independents.  There just are not a significant number of independents in this sample.

Which answers the question of how Romney could be trailing by 6 points in a poll that shows him leading independents by 14 points.


All in all, this looks like a poll result designed to be used to depress Republicans and promote the "inevitable reelection of the President" storyline.  This looks like news-making, rather than news-reporting.