Poll: Specter Barely Leads Sestak In Dem Primary

This sounds closer than reality. Other polls have a much bigger spread. Also, Sestak's name recognition cannot be that high. I wish this were accurate, as I am a big Sestak supporter. However, I think this poll might better reflect Scott Rassmussen's assessment of who would be an easier opponent for Pat Toomey than the reality.

Ras is playing favorites here. I still think Sestak with machine backing would be unbeatable, but the Obama Administration, the DSCC and the Rendell Machine are all in for Specter.

I think the Dems might be making a huge mistake here and in Florida and the GOP Caucus could add Toomey and Rubio come 2010.

This is excellent news. Specter simply cannot be trusted---it's time to end him once and for all.

Oh come on TPM, you know better than to trumpet a single poll, a year otu from an election. And from Rasmussen? Gosh, could they have any agenda here?

How accurate were the polls in 2007? Nice how that Guiliani vs. Hillary election worked out.

I think there is a lot of HCR noise right now.I'd like to see numbers once this contentious debate settles down.

TPM should not publish a single Foxmussen poll, ever, without linking to a detailed disclaimer article showing the history of Foxmussen polling and its rather poor record of reflecting reality.

To do otherwise is wasting our time and wasting TPM resources in efforts which will inevitably mislead TPM readers—and serve Foxmussen's interests and the interests of its patrons.

Don't ever quote Rasmussen and act like its credible. Evidence of their bias is easy to find. For example, take a look at this:

http://www.realclearpolitics.com/epolls/other/president_obama_job_approval-1044.html

"The new Rasmussen poll of Pennsylvania finds Sen. Arlen Specter just barely ahead of his Democratic primary challenger, Rep. Joe Sestak.

The numbers: Specter 46%, Sestak 42%, with a ±3% margin of error."

Ummm, that means the new Rasmussen poll found them *statistically tied.*

The published MOE is the MOE **of each estimate**, not the MOE **of a difference**.

If the estimates are independent (which they're not, here), the MOE of the difference would be roughly (square root of 2)*(MOE of each estimate). But if they're strongly negatively correlated (i.e. when one goes up, the other goes down, which is the case here), then the MOE of the difference is nearly double the MOE of the estimate.

So we're looking at an MOE of the difference of ~5%. In the Rasmussen sample, Specter leads by 4%. So we don't know who's ahead in the population as a whole, based on this sample.

Why pollsters never explain this stuff, I have no idea.

Well, if he loses he can always pull a Lieberman and run as an Independent.