You can clearly see that as the election boils down it is breaking for Webb.

you can see it looks like George Allen and his campaign laid out cold, face down,  after a war hero Marine "knocked Allen's soft whiny teeth down HIS throat".

Channel 7 in Roanoke is reporting a Roanoke College poll showing Allen up 45-42. The same poll also shows the 'marriage' amendment up 53-36.  Full story, including crosstabs on Allen-Webb is at http://www.wdbj7.com...

But how good and how recent is it?
Compare it to other recent (and actually more recent) polls.

http://www.realclear...

(You may want to go to this URL to see the formatted data).

RCP Average 10/20 - 10/29 - 46.0 47.2 5.6 Webb +1.2

Rasmussen 10/29 - 10/29 500 LV 46 51 3 Webb +5.0
CNN 10/26 - 10/29 597 LV 46 50 4 Webb +4.0
Roanoke College 10/22 - 10/29 453 LV 45 42 12 Allen +3.0
SurveyUSA 10/22 - 10/24 613 LV 49 46 3 Allen +3.0
LA Times/Bloomberg 10/20 - 10/23 385 LV 44 47 6 Webb +3.0

Real Clear Politics places it after the CNN poll in time.  Note that it took them 7 or 8 days to do the poll, and it started early, on October 22.  Sample was on the small size.  And look at the high level of undecideds -- 12.  The other polls surrounding it have undecideds at 3, 4 and 3.  This sounds like a "class project" poll.

If it was just SWVA thats not not unexpected, if its local to Roanoke then the sample size REALLY counts... If it includes NoVa and Tidewater, then there is something strange here..... It doesn't matter, we have our work cut out to do, to GoTV all weekend through POLL CLOSING Tuesday....

...then we previously realized. Allen needs huge margins in SWVA (and almost everywhere else) to counter what is likely to be a monumental drubbing in NOVA. If Allen is only up 3% in SWVA then he is dead where he stands...

Taking a close look at the crosstabs, I see some big problems with the poll's sample. If I understand the numbers correctly, only 20% of the sample consisted of NOVA. Clearly, NOVA will supply more than 20% of the votes on Election Day (NOVA accounted for nearly 28% of the total vote in the 2005 election, probably higher at 30% this time), while at the same time the % for central Virginia is the same 20%. That will not be the turnout pattern on Election Day. There is no way that "central Virginia" presumably Richmond-Petersburg and surroundings will have an equal number of voters as Fairfax, Arlington, Alexandria, Loudon, and Prince William. Another example of the sample's underlying errors is the under-representation of African-American voters, they have it as 7.7% of the sample; total bullsh#t! African American voters will be anywhere from 13-16% of the total electorate who will vote this time. Another error is in party identification.

The numbers are correct, but their underlying assumptions about who will vote on Election Day are BS. This might have been correct when Felix started out, back in the day when the "Byrd Machine" still had some power, but these assumptions no longer reflect the reality of rapidly growing NOVA.

Good assumptions for the "Byrd era" not worth much in 2006. As they say, "garbage in, garbage out".

Low sample size, over a long period of time, the problems you raise above.  I'm not giving it a minute's thought.

I picked up those two problems too, but I figured I had already made the point that it is "garbage in, garbage out". . I guess they gave it the "old college try". 

The negative publicity Allen generated from his Philistine novel-gate attack really just came in the last few days.