Bob Somerby, responding to CNN's exit polls:
It's hard to believe that those data are accurate. Did four percent of last week's voters really vote for someone other than Bush or Kerry in 2004? And what would explain that nine-point gap between Bush and Kerry voters? In theory, Democrats were enthusiastic about last week’s election, Republicans somewhat less so. Can it really be that 46 percent of last week’s voters voted for Bush in 2004 -- versus only 37 percent who voted for Kerry?
Bob isn't considering one crucial possibility here: misreporting. Political scientists have found a "retrospective bandwagon" effect in which some people will remember having voted for the winner even if they didn't. One example, as this paper reminds us, is that after his razor-thin victory about 65 percent of respondents claimed to have voted for JFK. Admittedly, Bush's extreme unpopularity should lessen these effects, but then this a pretty small retrospective bounce.
It is, of course, true that exit poll data should be treated carefully. But there's nothing about the 2004 election question that would suggest that CNN's sampling was bad; it's about the result you would expect.
--Scott Lemieux