Firstly, 17 extra-innings games seems like a lot, and it is a bit more than normal. Over the last 50 years (since the schedule went to 162 in 1962), the average number of extra-innings games played by a team is 14.4. We can model the number of extra-innings games played using a binomial distribution with a best-fit probability of about 9%. So, it’s not remarkably many; the Nats have 19 this year, and every season, we can expect about one team to have at least 20 extra-innings games.

However, a 15-2 record is very remarkable! We can well model the historical data with the naive assumption that every extra-innings win is a coin-toss: a binomial distribution with probability 50%. (I tried allowing some deviation from this assumption and found that the data will not support a fit allowing more than 1% standard deviation around 50%.) So for a team to win 15 of 17 is a probability of 1 in 964. Given that it’s somewhat rare for a team to play 17 games in the first place, we expect that this would almost never happen. In fact, under the assumptions of this model, we would expect to have to play MLB seasons for 574 years in order to find one that contains a 15-2 team at this point in the season (149 games in).

Conclusion: The Orioles 2012 extra-innings win-loss record is *not* well modelled by the same coin-toss distribution as the rest of the historical data. The Orioles (somehow) are good, not lucky.

But if you think something six sigma just happened, that’s often a sign that your model is wrong. (Again, unless you’re talking about something that’s literally about one person out of everyone in the USA, or similar things.)

]]>