2002 is the model, not 2000

The Kerry spot has it about right:

The Kerry Spot on National Review Online: "In 2002, pollsters used that same turnout model, expecting high turnout among Democrats and disappointing turnout among Republicans. And they got burned by that model."