Why is Democracy Corps oversampling Republicans?

I suppose I can vaguely expect that from non-partisan or Republican pollsters, but shouldn't a Democratic polling organization be taking more care to obtain a quality sample?

Their sample

Among the 92% of people in their poll (90% in the battleground states: given by Democracy Corps as Colorado, Florida, Indiana, Iowa, Michigan, Minnesota, Missouri, Montana, Nevada, New Hampshire, New Mexico, North Carolina, North Dakota, Ohio, Pennsylvania, Virginia and Wisconsin) who claimed to have voted in 2004, they have 52% having gone for Bush, 44% having gone for Kerry (53% for Bush, 42% for Kerry in the battleground states) despite the fact that Bush won 50.73%-48.27% (+2.5) nationwide and 51.44%-47.73% (+3.7%) in 2004.

Given that exit polls showed those over 65 (who are the most likely to have died and thus no longer be likely voters) voting no differently than the nation did, it's pretty unlikely that those voters in 2004 who are likely to show up again this year were 5 points (8 in the battleground states) more likely to go for Bush than those voting as a whole.

Again, let's measure based on facts.