This is Part 2 of my post in trying to come up with a better way to show a pitcher's effort using Bill James' Game Score as a basis to a new improved system which I will term GS+. You can review my starting premise
here. The long and short of my research was to determine how Aaron Cook's 74 pitch masterpiece stood up against other pitching performances during the 2007 season and against other great pitching starts. The anemic 67 game score seems to me not a good way to show the importance of this start.
In trying to come up with a modified Game Score I came with three game pitching indicators which I felt was missing from the Game Score scoring system.
1) Innings pitched. With Game Score you do score additional points as you go past the 4th inning but a complete 9 inning game should mean the most to any pitching performance. Therefore there should be something penalizing not going the full 9.
2) Pitch Count. The second indicator I looked at is how close was the pitcher to the magical "81" pitches. If you recall a perfect game score is 114 points which is 27 strikeouts in a complete, perfect game. To do this a pitcher would have to throw 81 pitches. So for Cook's 74 pitch game performance, he was actually more economical and to me and that should be a factor in his performance and Game Score.
3) Pitch Normalization. My final indicator is a a way to normalize pitching performances. As mentioned earlier I think going 9 should mean more and yet with Game Score you do find high strikeout pitchers getting high game scores of 70s when only pitching 6 or 7 innings. Normalization was basically done to determine the amount of pitches from 81 in 9 innings. For instance, if a pitcher goes 7 innings and has 76 pitches we would normalize this to 9 innings and 98 pitches.
With these indicators as goals, I began trying to come up with values that would keep Game Score relatively intact but give a little extra for an economical type starting pitchers. Data sets that I used was ESPN's 2002 - 2007 best pitching performance
page, 2007 Rockies game score data from
baseball reference, and finally 1997 - 2001 STATs Inc Baseball Scoreboard. (Which by the way was one of the best year end reviews of baseball using stats ever!). This spread of data (about 330 pitching performances) is by no means complete and for further validation would probably require to look at all MLB pitching data over a period of time (ha, more work for me and probably some database experience!).
From my indicators, I developed two adjustment factors which I felt described what I needed and made them usable in numbers format. The first is pitches per out. This takes a starter's pitch count and divides it by the number of outs. I then take 3 which is a minimum number of pitches for an out in the "81" perfect game scenario and divide it by the above value. So in the perfect scenario a perfect game score pitched game would give a pitcher a ratio of 1.0. Graphing Game Score versus the ratio (pitches divided by outs) you get data tightly grouped with no real correlation between pitch count and Game Score.
Now if you take the Game Score and use the adjustment factor (Game Score multiplied by 3 divided by pitch count divided by outs) you get the following:
From this point the graph shows a better relationship to pitch count and Game Score. Throughout this study I used two landmark games to serve as a guide of whether this GS+ made sense. The two games were Red Bartlett's 58 pitch game (lowest for a 9 inning game which had a Game Score of 83) and Kerry Wood's highest 9 inning Game Score (105 with a pitch count of 122). The graph above currently shows just well pitched games not a team's yearly pitching performance. I will get into this analysis later but for any system it would have to make the games that Red Bartlett pitched (58 pitch complete game versus Kerry Wood's 105 all time best pitched game) look somewhat comparable. With the adjustment factor above Bartlett's game score becomes 116 and Wood's game becomes 70. Therefore although the method looks promising it adjusts the Game Score a bit too far out of whack to make it a logical extension of the Game Score framework.
The second adjustment factor that match the indicators I set out was something I called "Extra Outs" which basically takes the pitch count (normalizing it to 9 innings) and then dividing it by 3 (min for a strikeout...give pitcher's some credit for the pitches they aren't really throwing!). In this instance I graphed normalized pitch count versus Game Score to check the reality of this method. For Game Score you get the following:
If you look add the first adjustment factor and add it to the graph you get this:
Again you see a nice linear relationship between Game Score and pitches thrown. The outliers include Bartlett's 58 pitch game with a Game Score of 116 (on the low end) and Randy Johnson's 160 pitch 8 inning lost at Texas when he had 18 strikeouts and a 76 game score (top end of graph). Now adding in the second adjustment factor ("Extra Outs") you obtained the following:
The orange dots is the "Extra Outs" adjustment factor. The important factor in this adjustment factor is that Bartlett's new Game Score (GS+) is 91 and Wood's new GS+ is 91. What a coincidence! The key to this adjustment factor is that it awards pitchers who go 9 (no normalization required) and for those pitchers who do go 9 and have a pitch count lower then "81" and they get a so called "negative outs" and thus their Game Score is adjusted higher. So from 2002 - 2007 the best games were:
And the new GS+ top 10 ranks them this way:
Three of the original top 10 still make it. Problem with most high strikeout pitchers is that they throw a lot of pitches. So with the best pitched games the GS+ seems to be a good fit for pitching but going back to my original question how would Aaron Cook's 74 pitch complete game stack up (and how would one team's adjusted Game Scores look)? Also remember we gave each pitcher a break when normalizing their pitched games by using three pitches, what about normalizing to what their actual pitches per out? Stay tuned as Part 3 will look into this...