The “game score” is a value invented by Bill James that attempts to evaluate the quality of a pitcher’s start. It basically represents an attempt to quantify the quality of a pitched game without regard for factors such as game outcome, game conditions, park factors, lineup factors, time of day, or factors which would risk increasing inherent bias in the game score itself. Consider the calculation method:
Start with 50 points, because 50 points is a perfectly arbitrary number, and there is nothing quite like quantification built upon some arbitrary number.
Record an out? Add 1 point. Finish an inning after the 4th inning, and you add 2 points. Add 1 point for each strikeout, and subtract 2 points for each hit allowed. Take away 4 points for yielding an earned run and just 2 points for each unearned run. Subtract 1 point for any walks issued.
Philip Humber pitched a “perfect game” for the White Sox this past weekend, and his game score in that perfect game was a healthy 96. That’s 50 points plus 27 points for outs recorded and an extra 10 points for finishing the 5th-9th innings. That’s a total of 87 points plus 9 points for strikeouts which gives you the final total of 96 for the game score.
Oddly enough, Humber’s 96 tied Matt Cain‘s gem that he tossed against Pittsburgh on April 13th. Cain went 9 innings, giving up a single hit ans striking out 11.
Does this mean that Cain’s one-hitter was as good as Humber’s perfect game? Was it as difficult? Maybe. Maybe not. That’s not the point of the game score. The game score includes minimal bias in exchange for a very no nonsense approach to evaluating a pitching outcome which does not equate to a game outcome.
The limitations imposed by the game score concept make it difficult to quantify anything more than the cumulative effect of pitched ball outcomes for a game, and that leaves me wanting more. How difficult can it be to develop a slightly more complex evaluation that takes a lot more factors into account?
- Start with a base score of 100.
- Add 1 point for each out recorded the first time through the lineup.
- For subsequent times facing the hitters, use a small multiplier to alter the value added as a function of times through a lineup. The basic premise here is simply that it really is difficult to get the same guy out 3 or even 4 times in a game, and the game score should reflect that.
- Add 1 point for each strikeout and subtract 1 point for each walk.
- Take away 2 points for each hit, 4 points for each earned run, and 2 points for each unearned run.
- Add 2 points for each base runner resulting from an error that does not result in an unearned run.
- Add 2 points for being the visiting team’s pitcher, and add 1 point for being the home team’s pitcher.
- Include a fancy park factor that involves the park factor ranking subtracted from 30 and normalized to a scale between 0 and 3 such that the most hitter friendly park adds 3.0 points to the pitcher’s “Game Score Remix”, and the most pitcher friendly park adds only 0.1.
- Include another fancy factor that is a function of lineup factors. Take the combined OPS+ of the starting lineup and rank it among the 30 teams in baseball. Use the same normalization method as in step 8 so that facing the lineup with the best OPS+ in baseball earns the pitcher 3 extra points while facing the weakest results in a 0.1 bonus.
- One last thing to keep in mind. Wins and losses do matter. If the pitcher’s team wins the game, the starting pitcher gets 1 bonus point. If the pitcher’s team losses the game, the pitcher loses 0. It is mostly a symbolic gesture, but it shows that wins matter to some people.
Perhaps the world is not ready for a “GSR” or “Game Score Remix” concept yet, but I just want to get the discussion moving in the right direction. I just have issue with the idea that the showdown on April 18th between Cliff Lee (game score 85) and Matt Cain (86) resulted in the 2 pitchers being separated by a single game score point. Maybe the strikeout needs to be weighted as a function of situational leverage or something, because not all strikeouts are equal.