The 2000 Florida recount, which has generated renewed interest in light of the HBO dramatization to debut tonight, seemed at the time like a once-in-a-lifetime fluke of probability, the electoral equivalent of being struck by lightning. But then just four years later, had John Kerry pulled a few thousand votes closer to George W. Bush, we might well have had an automatic recount in Ohio and a replay of the whole scenario, with Ken Blackwell playing the role of Katherine Harris.
So just how (un)likely are these recount things anyway? To evaluate the problem, I looked for all occurrences in my simulation when the following happened: there was some state, or some combination of states, that the candidate who lost the electoral college had lost by no more than 0.5 points. But if the candidate had won some or all of these states, he would have won the elecotral college.
For example, in simulation run #132 of 10,000, Barack Obama initially picked up 250 electoral votes. But that doesn't count Pennsylvania, which he lost by just 0.04 points. That's a very obvious recount scenario.
Or, in simulation run #336, Barack Obama won the election with 290 electoral votes -- but had carried Indiana and Missouri by just 0.4 points and 0.5 points, respectively, close enough to trigger recounts. If McCain had won both recounts, he would have gained 22 electoral votes, enough to give him a 270-268 electoral win.
Overall, the recount scenario was triggered 703 times out of 10,000 simulations -- slightly more than 7 percent of the time.
By one definition, that does qualify as a "once-in-a-lifetime" occurrence: 7 percent means once out of about every 14 elections, and the average American adult will have the opportunity to vote in roughly 14 presidential elections before they die. But that's still quite a bit more likely than I would have thought.