If the same percentage of 2000 Gore voters and 2000 Bush voters had returned to the polls in 2004, we'd see equal percentages of them in the exit polls on 2000 voter preference. But instead the exit poll breaks this down as:

Did not Vote in 2000:	    17%
Voted for Gore in 2000:     37%
Voted for Bush in 2000:     43%
Voted for Other in 2000:     3%

(see the CNN exit poll summary, about 4/5 of the way down for the actual data.)

According to the exit polls, Kerry won among the new voters 54/45, and among the "Other" voters 71/21. Since Kerry and Bush each took about the same fraction of their own 2000 supporters (90% and 91%, respectively), it's clear that the 2000 Gore voters who stayed home are what cost the election -- there were 7M of them.

43% of the roughly 116M votes cast = 49.9M
37% of the roughly 116M votes cast = 42.9M

What's going on here? Are the exit polls screwed up? Did people lie about who they voted for in 2000?