Bill James developed a theory that run differential (squared) would long term approximate a MLB’s teams win %.   Appears to make sense – long term, if you score a lot more runs than you give up, you will win more games than you lose.

HOWEVER, there is a BIG EXCEPTION – strong offensive teams.  If you score a lot of runs, your run scoring becomes ‘clumpy’, you are a positively skewed outlier, and you will end up appearing ‘unlucky’ based on James’ calculation.  Specifically, you will consistently win LESS games than the James model predicts.  How many less games?  3 per year.   Looking at the data for 3 MLB seasons (2013-15), laid out below, the top 20% of teams in offence, measured by runs scored, average losing 3 MORE games per year than Bill James model predicts.   Its remarkably consistent as the top 6 offences each of the 3 years combined to lose exactly 18 MORE games than James model predicts each year.

Furthermore, each year there is a .3-.45 correlation between runs scored and being ‘unclutch’ (or winning less games than James model predicts).   This is further evidence that if you score a lot of runs, you positively skew your +/- causing your team to win less games than expected.

Finally, homerun hitting teams are ‘unclutch’ using James’s model.   Over the same 3 yrs, the 4 best homer hitting teams, the O’s, Jays, Mariners, and Astros combined to win 24 LESS games than Bill James model predicts.

Power hitting = unclutch based on Bill James model due to the positive skew of piling on runs.

2 Lessons to be Learned:

1 – for sabermetrics, or anyone obsessed with blindly following stats, always question the formuls.  There are sometimes major errors.

2 – for offensive teams, beware, you will consistently underperform expectations.  This is NOT due to bad luck, but the clumping nature of offense (scoring 12 runs doesn’t increase your chances of winning a game much more then 8 runs, but does help your +/-).   Come playoff time, 1 run is precious, building high powered offences may disappoint as the team isn’t built to execute.

Last modified: March 18, 2023