The return of the Gallup anomaly

In 2000 Gallup polls of likely voters were significantly more favorable to Bush than the average poll. I was convinced that this had something to do with their likely voter filter. I was (really honestly) planning to post on this to reasure Dems dismayed by the Gallup poll in April which showed Bush 5 points ahead.

Now there is proof and Atrios (of course) found it first.

"And, Schneider let slip that Kerry's ahead by 6 points among registered voters - it's only using their likely voter formula that Bush is up 48-47."

Bush-Kerry increases 7% with the likely voter filter which reduces sample size by less than 50% ! How the hell does it work ? Gallup can't decide that registered voters are more likely to vote if they say they support the Republican.

They wouldn't dare use race would they ?