Do undecided voters really break for the challenger?

Writing at The Hill, Mark Mellman casts doubt on the idea that undecided voters choose the challenger over the incumbent by large margins and points out that the phenomenon was anything but ironclad in the 2008 and 2010 Senate cycles.

Of course, the big question for our focus is how voters behave in presidential elections, so I checked into a few, quick stats.

There wasn't an incumbent president in 2008, so let's travel back to the last time an incumbent ran -- 2004.

In 2004, John Kerry won 53% of voters who made their decision within the final three days of voting, while George W. Bush took just 44% of that group.

Meanwhile, Bush won 52% of those who decided earlier than three days, while Kerry won 47%.

Thus, in the 2004 election, the law of the challenger played out.

Now... here's one of the more impressive validations of the idea that undecided voters break for the challenge in presidential elections.

In 1996, voters who made their decision in the last three days actually broke for Bob Dole over Bill Clinton by 3%, and voters who decided in the last week backed Dole by 12%!

Clinton, though, led by double digits among those who made their decision in the last month or before that, and won the general election by nearly 9%.

So based on the two, most recent presidential elections with incumbents, the law of the challengers looks like it holds in recent presidential history.

But here's my question -- will it hold more for some states than others, and is that based on whether they're red or blue states?

In other words, would late deciders in Florida be more likely to back Romney than late deciders in Wisconsin, considering that the former is, generally, a more conservative state than the latter?