The First Read gang alludes to a theory that many Republican pollsters are worried about:
"[R]ight now, pollsters will tell you that with older voters leaning McCain these days, any likely voter model is going to favor McCain for now. If Obama moves younger voters as well as many observers assume come October, the likely voter numbers could change."
There's a lot packed into here. Based on data, studies and experience, pollsters assume that older voters tend to reach information saturation earliest, tune in the earliest, and pick their candidates the earliest. Likely voter models this far out don't oversample older voters per se -- they oversample voters who have made up their mind and aren't likely, even if they say they're likely, to change their minds. John McCain's leading among older voters, but not by much. So when younger voters -- younger than 65 -- begin to make up their minds in the fall, likely voter models will move back into equilibrium and Obama's lead among registered voters should begin to match his lead among likely voters.
This theory neatly takes care of the so-called Bradley effect, where white voters will consistently overstate their preferences for a black candidate because they don't want to admit feelings of racial prejudice to pollsters. If one assumes that the younger an American is, the less likely he or she is to harbor actionable prejudices, the Bradley effect is operating now, and will diminish as time goes on.
Some pollsters have a different view. As the election approaches and the voter perceives the choice more acutely, he or she will fall back on the familiar, the safe, the less risky especially if there is a major difference between the known and unknown. Experimental psychologists have accounted for this effect in the lab, but I don't know whether it applies, or ought to apply, to politics.