Since the creation of my pitch quality model, aStuff+, earlier this season, I have written numerous posts discussing how pitch quality models can be utilized to evaluate pitching performance, as well as to identify potential shortcomings that these models might possess. For example, a notable shortcoming of aStuff+ is its lack of adjustment for altitude and environmental effects, which causes pitches to be graded higher in controlled environments like Tropicana Field and lower in high-altitude ballparks such as Coors Field.
Over the past couple of weeks, I have noticed some discussion on social media regarding whether pitch models are biased towards either right or left-handed pitchers. Since there are more right-handed pitchers in the league than left-handed pitchers, some observers believe that models are skewed to favor right-handed pitchers. Other observers have cited anecdotal examples of left-handed pitchers being “underrated” by pitch quality models. In this article, I will take a look at whether aStuff+ favors a specific handedness or is handedness-neutral in its evaluation of pitch quality.
First, a simple way to analyze whether aStuff+ favors a specific handedness is to evaluate the average aStuff+ for both right and left-handed pitchers. Similar to the analysis I conducted in my article introducing my pitch quality model, I calculated the average aStuff+ for each pitch type as well as the average aStuff+ among all pitch types, grouped by pitcher handedness, for all pitches thrown during the 2021, 2022, and 2023 seasons.
As shown by the table above, right-handed pitchers experience a slight advantage in the aStuff+, with the average pitch being graded as a 101, while the average pitch thrown by a left-handed pitcher grades as a 98. On the individual pitch type level, pitches by right-handed pitchers tend to be graded one or two points higher than pitches thrown by left-handed pitchers. I would not draw too many conclusions from the differences present with slurves, splitters, and sweepers, since splitters and sweepers are less commonly thrown by left-handed pitchers, and true slurves are rarely thrown by pitchers of both handedness.
Does this mean aStuff+ is biased against left-handed pitchers? In my opinion, not necessarily. While pitches thrown by left-handed pitchers do average a lower aStuff+, a higher proportion of left-handed pitchers are starter pitchers, potentially contributing to this phenomenon.
To conduct this analysis, I analyzed all pitchers who have thrown at least 250 pitches during the 2024 season. To qualify as a starting pitcher, the pitcher must have started in at least half of their appearances this season. As shown by the table above, a higher proportion of left-handed pitchers are starting pitchers compared to right-handed pitchers.
Starting pitchers tend to have lower aStuff+ grades than relief pitchers for a variety of reasons. First, command is an important attribute to have as a starting pitcher which lowers the bar of minimum “stuff” required to be utilized as a starter. Second, starting pitchers are more likely to have a diverse pitch arsenal, also lowering the bar of minimum “stuff” required to be utilized as a starter due to the deception to the hitter a wide arsenal creates. Third, relief pitchers by nature are more “max effort” with their pitches, since they typically only pitch one inning at a time, typically resulting in higher velocities than if they were used as a starting pitcher. Since a larger percentage of left-handed pitchers are utilized as starters compared to right-handed pitchers, I believe that it can be reasonably expected that pitches thrown by left-handed pitchers would on average be graded lower by aStuff+
In addition to being utilized to evaluate the effectiveness of specific pitches, aStuff+ and other pitch models are valuable to use as a predictive metric to project a pitcher’s performance. As I discussed in the model’s introductory article, aStuff+ is an effective metric to use to predict a pitcher’s future ERA, FIP, and K-BB%. To further evaluate whether aStuff+ is biased towards a specific handedness, it is imperative to evaluate these correlations, this time grouped by whether the pitcher is right or left-handed.
As shown by the tables above, aStuff+ continues to perform well at predicting future FIP and K-BB% when broken down by pitcher handedness. Notably, the model largely performs better at predicting the future performance of left-handed pitchers than right-handed pitchers, contrary to the narrative that pitch quality models undervalue southpaws.
Frankly, I was a bit surprised to discover that the model performs better as a predictive metric for left-handed pitchers given that I went into this analysis with the expectation that right-handed pitchers would perform better. Furthermore, it is a bit confusing that left-handed pitchers grade lower by aStuff+, but their pitch grades are more predictively powerful. Perhaps this difference can be attributed to the fact that there is a smaller sample size of left-handed pitchers in Major League Baseball, causing there to be a stronger correlation between aStuff+ and future FIP and K-BB%. Also, ERA and FIP are commonly used to evaluate starting pitchers, not relievers, and it is interesting to note that there is a larger difference in correlation coefficient between right and left-handed pitchers when it comes to predicting FIP than K-BB, which could tie back to the fact that a larger proportion of left-handed pitchers are starting pitchers. Regardless of interpretation, the data indicates that aStuff+ largely performs better at predicting the future performance of left-handed pitchers than right-handed pitchers.
After conducting this analysis, it is difficult to state that aStuff+ favors a particular handedness. Right-handed pitchers grade slightly better on a pitch-by-pitch basis, but the model is more powerful at predicting the future performance of left-handed pitchers. I do think this analysis opens up opportunities for future research regarding which pitch specifications are more effective for right or left-handed pitchers, and I encourage everyone to utilize my aStuff+ Blob App to explore the model themselves. While location is not a feature in my pitch quality model, I am curious if location matters more to left-handed pitchers which could cause them to perform better than pitch models indicate. Perhaps left-handed pitchers are inherently deceptive to opposing hitters, causing them to outperform the grades assigned to them by pitch models. All of these topics will be interesting avenues for future research. While pitches thrown by left-handed pitchers are graded slightly lower by my pitch quality model, my initial conclusion from this research is that there is not a significant handedness bias present in the aStuff+ model.
Thanks for reading!
Follow @MLBDailyStats_ on X and Adam Salorio on Substack for more in-depth MLB analysis. Photo credits to Getty Images.