The indispensable Mark Blumenthal responds to my query about whether polls of registered voters might understate support for (and, possibly, the bounce of) a candidate like Obama, whose supporters are especially enthusiastic. Blumenthal's bottom line: maybe, but there isn't a whole lot you can do about it, since it's incredibly tough to identify a likely voter several months out from an election. (Which is why ABC doesn't seem to want to report its own likely voter numbers.)
Intriguingly, though, Mark says Gallup has one of the more sophisticated likely-voter models in the business, and that Obama consistently does better against McCain among likely voters than registered voters there. So there's some support for the theory. It'll be interesting to see how big a bounce Obama gets in both versions of the Gallup poll whenever the first post-June 7 edition is released. (June 7 being the day Hillary officially conceded...)
For the moment, here's the chart Mark made to catalogue the difference between likely-voter and registered-voter results in the Gallup poll:
P.S. Mark notes that Kerry did slightly worse among likely voters than registered voters in Gallup's 2004 polling. But, in retrospect, that could be because his voters were less enthusiastic than we thought (and less enthusiastic than Bush's). Bush is generally assumed to have won the turnout battle, after all.
Come to think of it, the difference between Gallup's likely and registered voter numbers is probably a decent indication of who's going to have the turnout advantage in November. Advantage Obama so far. Not that that's surprising.