It’s difficult to know which polls to trust.  We’ve seen wild swings in many of them, and the only point of agreement between any of them is that none of them show the same electorate.  However, when selecting a poll for its predictive value in 2008, one can probably disregard one that still gets 2004 wrong.  The CBS poll released yesterday showing an 11-point lead for Barack Obama also picks John Kerry as the winner in the last election, at least among early voters (page 2):

2004 vote:    Early            Likely
Bush……………40%…………..40%
Kerry…………..45……………..40

This shows two significant problems.  First, Bush beat Kerry by three points, which makes this a little less valuable as a predictive tool.  Second, how did CBS determine “likely voters”?  In their methodology, they show only 9% of their sample as new voters, but the combined total of Bush and Kerry voters only comes to 80%.  Where did the other 11% go?  Into the ether?  And why would Bush voters be less likely to vote in 2008 than in 2004 (the missing 3%)?

It’s not the only problem in this poll.  As it does in most polls, CBS far oversampled Democrats.  The party-identification split on Democrats to Republicans has a 13-point gap, 41% to 28% respectively, among likely voters.  With most people predicting a gap closer to 3-5%, CBS gives Democrats at least an extra 8% in the sample — which accounts for most of Obama’s lead, apart from the other problems in this survey.

Polls are always an educated guess at the best of times.  When a poll has trouble predicting the outcome of an election from four years earlier, it’s safe to ignore it. (via Reliapundit)