I received an email from John Perricone over at Only Baseball Matters, and he raised a very good point. He states that an inconsistent offense could cause a team's actual won-loss record not to reflect its expected. If you follow the link, you can read about it at his site.
I had touched on this subject as it concerns the Red Sox-Yankees race on August 5 ("Underachieving Red Sox"). I had noticed that 5 of the 7 game differential between their expected and actual standings could be accounted for by isolating 8 games that the Red Sox had been involved in, in which the margin of victory was 10 or more. I had left it out, perhaps disingenuously, so as not to complicate the issue. Thanks to John for keeping me honest.
I proposed to John that there is probably some standard deviation threshold for runs scored and/or allowed. Once a team surpasses it, the formula no longer holds. Perhaps some analysis can be done on the ratio of standard deviations in the runs for, runs against, and margin of victory to the differential between expected and actual record. Maybe a modified formula could be employed in these cases. One issue: standard deviations have limitations especially when dealing with populations of different sizes (this is a topic for another day). As Retrosheet makes more game scores available, these sorts of analyses will be possible.