A few days ago, Twitter users @Phrontiersman and @dmc0603 had an interesting conversation about taking the concept behind "nth Best Season" graphs and applying it to a pitcher's single season (all in the context of comparing the pitchers involved in AL Cy Young race). That is, using a metric that measures the effectiveness of a given start by a pitcher, create a graph in which the x-axis sorts each start best to worst, left to right. I found this to be an interesting concept and a perfect opportunity to steal their idea (but really guys, thanks for the inspiration). While the two aforementioned Twitter users couldn't decide on a statistic to evaluate each individual start, they did agree that something along the lines of WPA (win probability added) in a run-support neutral environment would be a good fit. After a considering a few different routes I could have taken, I decided on using Tom Tango's FIP-based version of Game Score.
The concept of Game Score is simple. Originally developed by Bill James, it is an attempt to measure a pitcher's effectiveness for a given start on a 0-100 scale. Tom Tango recently outlined a few more effective versions of Game Score at FanGraphs that improve on the limitations of the original formula. Among these new versions was a FIP-based Game Score, which I ultimately decided on using for a two main reasons: it's (pretty much) defense-independent and it is scaled to measure win probably (a game score of 75 means that you will win the game 75% of the time, etc). The formula is:
FIP-based Game Score = 2.5*IP + (2*K - 3*BB - 13*HR) + 40
Now on to the the practical application. I made two graphs - one for the pitchers involved in the often talked-about AL Cy Young Race and one for the (not so often talked-about) NL race. Again, I did this simply to get a new perspective on how each pitcher has performed in relation to one another, perhaps shedding some new light on a conversation that hears the same things over and over. Let's start with the AL race.
If anything, this graph confirms that this Cy Young race is incredibly close through 25-26 starts. According to this measurement, Weaver and Sabathia have very similar graphs: one bum start, lots of starts in the 60-70 range, and a few dominant starts. More than anything, that spells consistency. Verlander, on the other hand, has had a slightly more polarizing season with more starts that rank at both extremes. This is demonstrated in Verlander's standard deviation of 12.36 compared to Sabathia's 10.83 and Weaver's 11.3 (omitting their outlier starts). (Interesting little note: Verlander's no hitter ranks as his 8th best start of the season according to this metric with a score of 67.5).
Choosing pitchers for an NL "race" was much more difficult as there is not nearly as much racing going on as we see in the AL. Through 24 starts, Roy Halladay has put up great numbers in every category and has amassed 6.1 fWAR with Cole Hamels at a distant second at 5.0. Hamels is having a career year and is certainly in the discussion, as is Clayton Kershaw who has been a strikeout machine, ringing up nearly 10 batters per nine innings pitched. Just like Sabathia and Weaver, Hamels and Kershaw look nearly identical, with Kershaw perhaps edging out Hamels on the upper echelon of starts. Yet the obvious leader here is Doc, whose red line is on top for the vast majority of the graph. I would imagine it would take an injury/a handful of poor starts to prevent Halladay from notching his 3rd Cy Young this season.
Do these graphs really change how we look at the Cy Young races this season? Not really. But I feel it's a fresh way of visually looking at a pitcher's season. I anticipate utilizing these types of graphs in the future for perhaps a more enticing, bigger project.