>This is amazing. So the "credibility" of US polls rests on the
>proposition that some proportion of their sample consists
>of liars about voting behavior who nevertheless are telling
>the truth about voting intentions?
Yeah, and why is that so incredible? There's no bandwagon effect when the winner-to-be is unknown. (People also exaggerate their voting behavior; Census turnout figures are often 10 or more percentage points higher than the actual.) And the polls do a very good job in projecting election results, with little deterioration in recent years. See <http://www.gallup.com/content/?ci=1258> and <http://people-press.org/reports/display.php3?ReportID=211>.
Doug