Introducing aBatting+
Quantifying offensive production utilizing swing quality, contact ability, and swing decisions.
The introduction of bat tracking metrics, such as bat speed, into the sphere of public analysis is poised to become a major advancement in how the public is able to analyze the offensive production of Major League Baseball players. A player’s average bat speed becomes reliable very quickly (in a sample size of fewer than ten swings), which, combined with its predictive power, makes it a powerful tool to utilize when attempting to analyze the “true talent level” of a given player in a small sample size. Last week, I introduced my swing quality metric, aSwing+, which combines bat speed with various other bat tracking metrics (such as attack angle, swing path tilt, swing length, etc.), which outperforms simply utilizing a player’s average bat speed to analyze their “true talent” raw power ability.
As mentioned in my aSwing+ release article, hitting at the Major League level is a lot more complex than simply possessing a powerful swing, and while having a high aSwing+ is a straightforward path to generating above-average production at the Major League level, this raw power ability will not be able to translate into in-game production without making quality, “squared-up” contact. In addition, a player must be able to display good swing decision ability at the plate, as they may be vulnerable to swing-and-miss by swinging at pitches out of the strike zone, or they may be too passive in the zone, which will lead to unfavorable count states for the batter. Evaluating an offensive player’s production can be complex, but it largely amounts to these three skills: making quality swings conducive for offensive production (as measured by aSwing+), displaying good swing decision ability, and generating efficient “sweet spot” contact.
In addition to aSwing+, I have created two new metrics, aDecision+ and aContact+, which measure the quality of each player’s swing decisions and each player’s ability to generate efficient “sweet spot” contact, respectively. Utilizing the features that were included in the aSwing+, aDecision+, and aContact+ models, I have created a fourth model, aBatting+, which quantifies a player’s offensive production based on these three components of hitting and outperforms existing metrics, such as xwOBA, in terms of predictive ability. In this article, I will discuss the methodology behind constructing each model, evaluate which players are the best offensive players in Major League Baseball according to the model, and discuss the limitations and promise of utilizing aBatting+ moving forward.
aDecision+
Over the past couple of years, I have been enthralled in identifying a more effective means for evaluating a hitter’s swing decision ability. As mentioned in an article I wrote earlier this season, existing metrics such as Chase% and Z-O Swing% provide value in evaluating certain areas of a player’s swing decision ability but display a few limitations, such as treating all count states as equal and treating pitch location as a binary outcome (in-zone/out-zone, as opposed to specific locations). Swing decision models, such as Drew Haugen’s SwRV and Robert Orr’s SEAGER, have attempted to account for these factors, and I attempted to create a swing decision model last year, with the release of my SOTO metric.
aDecision+ is similar to SOTO in that it attempts to measure the quality of a player’s swing decisions by predicting the expected run value of each swing or take, based on the location of the pitch, the vertical and horizontal approach angle of the pitch, and the count state. On takes, aDecision+ arrives at an expected run value by calculating the probability that the pitch would result in a ball, called strike, or hit-by-pitch; while on swings, aDecision+ arrives at an expected run value by calculating the probability that the pitch would result in various batted ball classifications, a foul ball, or swing-and-miss. An expected run value, grouped by count state, is mapped back onto each probability to quantify the swing decision value of a given pitch. The model, a CatBoost classifier model, is trained on all pitches that contain bat tracking data from the 2023 and 2024 seasons, with predictions made on all pitches with the requisite data present from the 2023, 2024, and 2025 seasons.
In essence, aDecision+ attempts to take the best qualities of Chase% and Z-O Swing% and combines them into one comprehensive swing decision metric. The tables above depict the descriptive correlations between aDecision+ and wOBA, BB%, and ISO, among players who faced at least 500 and 1,000 pitches during the 2025 Major League Baseball season, as well as the predictive correlations between aDecision+ and wOBA, BB%, and ISO among players with at least 500 and 1,000 pitches faced in each of the 2024 and 2025 Major League Baseball seasons.
As shown by the table above, Chase% performs slightly better at describing BB%, and Z-O Swing% performs slightly better at describing ISO; however, aDecision+ performs well at describing both metrics, resulting in aDecision+ displaying better performance at describing wOBA among the swing decision metrics. aDecision+ displays the same predictive performance as Chase% as it pertains to predicting BB%, while Z-O Swing% displays superior performance at predicting ISO, however; aDecision+ displays the strongest predictive ability at predicting wOBA.
The table above depicts the players with the top 10 highest and bottom 10 lowest aDecision+ grades during the 2025 Major League Baseball season, minimum 500 pitches faced. aDecision+ is scaled on a 100 mean / 10 standard deviation scale, relative to the average player’s swing decision quality. For example, a player with a 120 aDecision+ possesses a swing decision ability that is two standard deviations better than the average Major League player (in terms of scouting grades, this would be 70-grade plate discipline). Kyle Tucker has displayed the best swing decision ability in Major League Baseball this season, with a 126 aDecision+, with Max Muncy a close runner-up with a 125 aDecision+ this season. Ronald Acuña Jr.’s mix of low Chase% and high Z-Swing% results in him being awarded a 122 aDecision+, while Juan Soto also generated a 122 aDecision+. To little surprise, Javier Báez displays the worst swing decision ability in Major League Baseball, with a 58 aDecision+, with Jhonkensy Noel displaying the second-worst swing decision ability at 61 aDecision+.
Evaluating a player’s swing decision ability can be particularly complex, since very few hitters generate positive run values on their swings, underscoring the importance of making swing decisions to improve the likelihood that a given swing will result in a positive run value. As shown by the graph above, aDecision+ values players who display lower swing rates, regardless of location, further displaying the importance of a patient approach at the plate. As with the other offensive metrics I have released this week, there is some level of selection bias present with this relationship, as there is an undefined threshold at some point where a hitter is simply too passive at the plate, which would result in them getting into too many unfavorable count states, elevating their strikeout totals and diminishing their on-base ability, however, in general, a player with a lower swing rate will display a better swing decision ability than a player with a higher swing rate.
Swing decision modeling has, personally, become a fascinating topic in recent seasons, because the ability to make good swing decisions affects many aspects of hitting. Not only does good swing decision ability increase a player’s on-base ability, but most damage on batted balls occurs against pitches that are located over the heart of the plate, and hitters who are either not patient enough to get into favorable count states, or are too passive and not swing at ideal pitches to generate damage, are likely to underperform the level of offensive production that is expected by their swing quality. Swing decision models, such as aDecision+, allow analysts to evaluate which players make the best swing decisions, as well as to review which pitches a hitter should have swung at during a given at-bat. There will be a lot more to dig into regarding swing decisions on this blog; however, at the present moment, the primary conclusion of this model is that it is critical for a hitter to display good swing decision ability to generate above-average offensive production at the Major League level.
aContact+
As mentioned earlier, displaying high bat speed and an ideal swing path is a very important element of generating offensive production at the Major League level. In addition, analyzing a player’s batted ball quality (exit velocity and launch angle) has been a critical means of assessing how much value a player is expected to produce on their batted balls. While these metrics are essential components of offensive evaluation, they are irrelevant if a player is unable to make contact with incoming pitches during the game.
Since the use of peripheral metrics became widespread, statistics such as Contact% and Z-Contact% have been commonly used to evaluate a player’s bat-to-ball ability. While these metrics make intuitive sense, they miss one crucial component of the bat-to-ball collision, which is where contact was made on the bat. Contact that is made on the “sweet spot” of the bat is more valuable than contact that is made on the “handle” of the bat, since contact made closer to the bat’s center of percussion (“sweet spot”) induces less vibration of the bat, resulting in more energy being transferred into exit velocity (For more information, please check out this interview that Keenan Long of LongBall Labs did with Eno Sarris on the Rates and Barrels podcast in April 2025). In addition, contact that is made away from the “sweet spot” of the bat has different values in different count states (for example, a ball hit off “the end” of the bat when the hitter is ahead in the count is a wasted opportunity for damage, while it avoids a strikeout in a two-strike count), and the traditional metrics do not account for these factors. Therefore, I decided to create aContact+, which takes into account collision efficiency (“smash factor”), count state, and pitch location to evaluate a hitter’s contact ability.
In essence, aContact+ is a more comprehensive version of Driveline’s Smash Factor metric, with the additional context of the count state of a given pitch, as well as the incoming pitch’s location and approach angles. The model, a CatBoost classifier model, is trained on all batted balls that contain bat tracking data from the 2023 and 2024 seasons, with predictions made on all swings with the requisite data present from the 2023, 2024, and 2025 seasons. aContact+ arrives at an expected run value for each swing by calculating the probability that each batted ball would result in various batted ball classifications, then maps an expected run value, grouped by count state, back onto each probability to quantify the contact value of a given swing. For swing-and-misses and foul balls, the observed run value of these events is utilized as the swings’s aContact+ value.
The table above depicts the descriptive correlations between aContact+ and wOBA among players who have faced at least 500 pitches during the 2025 Major League Baseball season, as well as the predictive correlations between these metrics among players who have faced at least 500 pitches during the 2024 and 2025 seasons. In terms of descriptive performance, aContact+ has a stronger relationship with wOBA than Contact%, and notably, aContact+ displays a positive relationship with wOBA, compared to the negative relationship displayed between Contact% and wOBA. While both metrics are relatively weak in terms of predictive power, Contact% displays a slight edge over aContact in predicting a player’s Y2 wOBA, as well as in year-to-year stability.
The table above depicts the players with the top 10 highest and bottom 10 lowest aContact+ grades during the 2025 Major League Baseball season, minimum 500 pitches faced. aContact+ is scaled on a 100 mean / 10 standard deviation scale, relative to the average player’s contact ability. For example, a player with a 120 aContact+ possesses a contact ability that is two standard deviations better than the average Major League player (in terms of scouting grades, this would likely be a 70-grade hit tool).
Luis Arraez has displayed the best contact ability in Major League Baseball this season, with an outstanding 140 aContact+, well above Steven Kwan, who displays the second-best contact ability with a 131 aContact+. DJ LeMahieu is the only other player with an aContact+ greater than or equal to 130, while Jacob Wilson possesses a 126 aContact+. The bottom of the leaderboard is filled with players who display fast swings but frequently swing and miss, such as Alexander Canario, Jhonkensy Noel, and Giancarlo Stanton.
In addition to taking into account the collision efficiency of each batted ball, aContact+ differentiates itself from other contact metrics due to its ability to take into account the count state of each contact event. As shown by the table above, while “squared up” contact events are always the most valuable, the value of a “non-squared up” batted ball event relative to a swinging strike varies depending on the count. Early in the count, as well as when the batter is ahead, it is more preferable for a hitter to swing and miss at a given offering, rather than make weak contact. This difference in value exists because, for example, in a 1-0 count, the batter will have other opportunities within the at-bat to make productive contact, while a “non-squared up” batted ball has a higher probability of turning into an out. In a two-strike count, a swing-and-miss will directly lead to an unfavorable outcome for the batter (an out, or possibly a dropped third strike), compared to a “non-squared up” batted ball, which at least has a chance of resulting as a hit. While this nuance may perhaps contribute to diminished predictive value relative to Contact%, I believe that this added context is valuable to understand a hitter’s “true” contact ability, as evidenced by the metrics’ descriptive performance.
aBatting+
Now that metrics have been introduced to quantify the three most important aspects of a player’s offensive profile: swing quality (aSwing+), swing decisions (aDecision+), and contact ability (aContact+), it is time to combine these three factors into a single comprehensive metric to quantify a player’s offensive production, aBatting+. aBatting+ is a fourth model that takes into account all the features that are present in the existing three models to create a comprehensive metric for offensive production that displays better predictive performance than existing expected production metrics such as xwOBA and Process+.
Similar to the aDecision+ and aContact+ models, aBatting+ is a CatBoost classifier model, trained on all pitches that contain bat tracking data from the 2023 and 2024 seasons, with predictions made on all pitches with the requisite data present from the 2023, 2024, and 2025 seasons. On takes, aBatting+ uses the same features as aDecision+ to arrive at an expected run value by calculating the probability that the pitch would result in a ball, called strike, or hit-by-pitch. On swings, the model directly observes whether the swing resulted in a ball-in-play, foul ball, or swing-and-miss; then, on balls in play, uses the bat tracking metrics, as well as the collision efficiency of the batted ball, to calculate the probability that the batted ball would result in the various batted ball classifications. Then, an expected run value, grouped by count state, is mapped back onto each probability to quantify the expected offensive production of a given pitch.
The tables above depict the descriptive correlations between aBatting+ and wOBA among players who have faced at least 500 and 1,000 pitches during the 2025 Major League Baseball season, as well as the predictive correlations between aBatting+ and wOBA among players who have faced at least 500 and 1,000 pitches in both the 2024 and 2025 Major League Baseball seasons.
As shown by the table above, while aBatting+ does a solid job at describing a player’s offensive production during the season, xwOBA displays superior descriptive performance, while Process+ sits between these two models in terms of performance. Regarding predictive performance, aBatting+ displays superior performance at both predicting itself and a player’s wOBA year-to-year, displaying higher correlation coefficients than both xwOBA and Process+.
The table above depicts the players with the top 10 highest and bottom 10 lowest aBatting+ grades during the 2025 Major League Baseball season, minimum 500 pitches faced. aBatting+ is scaled on a 100 mean / 10 standard deviation scale, relative to the average player’s predicted offensive production. In addition, I have also created aBat wOBA, which scales aBatting+ on a wOBA scale for easier interpretation. Aaron Judge is the best offensive player in all of Major League Baseball with a 133 aBatting+, with Shohei Ohtani and Yordan Álvarez possessing a 129 aBatting+. Perhaps two surprising names that are present on the aBatting+ leaderboard are Ben Rice (with a 124 aBatting+ and .402 aBat wOBA) and Josh Bell (with a 122 aBatting+ and .394 aBat wOBA). Players towards the bottom of the leaderboard include Martín Maldonado, Hyeseong Kim, and Nick Allen.
As mentioned earlier, the aBatting+ predicts the probability of whether or not each batted ball is a barrel using the bat tracking, swing path, and collision efficiency metrics. The table above depicts the leaders in expected barrel rate (xBarrel%) during the 2025 Major League Baseball season (minimum 500 pitches faced), according to the aBatting+ model. Aaron Judge leads the league with a 24.9% xBarrel, while the remainder of the leaderboard contains the league’s best power hitters, such as Shohei Ohtani, Kyle Schwarber, and Cal Raleigh. xBarrel% displays a strong correlation with observed Barrel%, with xBarrel% explaining ~82% of the variance in a player’s Barrel%, among batters who faced at least 500 pitches during the 2025 Major League Baseball season.
The table above depicts the predictive correlations between Barrel%, xBarrel%, and various offensive metrics among players who have faced at least 500 pitches in both the 2024 and 2025 Major League Baseball seasons. As shown by these correlations, xBarrel% displays better predictive performance with wOBAcon and xwOBAcon than Barrel%, and is superior at predicting itself year-to-year, while Barrel% displays an advantage at predicting next year’s ISO. Both Barrel% and xBarrel% are equally powerful at predicting next year’s wOBA and Barrel%. Given how close these metrics are in terms of predictive ability, either can be effectively used as a predictive metric for batted ball quality; however, given its slightly better performance at predicting wOBAcon and xwOBAcon, I believe that xBarrel% is a more effective predictive metric than Barrel%.
In summary, here are what I believe are the main takeaways from this suite of offensive metrics:
Swing the bat fast and on an upwards plane (aSwing+).
Place an emphasis on swinging at pitches that the batter can produce damage against (aDecision+).
Making contact on the “sweet spot” of the bat is critical in non-two strike counts (aContact+).
Combine these three elements to generate above-average offensive production (aBatting+).
Reliability Analysis
While these metrics are effective tools to utilize when analyzing how a player generates their offensive production, it is important to conduct a reliability analysis to understand how large a sample size is necessary before aDecision+, aContact+, and aBatting+ can be used to reliably understand a player’s “true talent level”. As discussed in my previous article, one of the outstanding features of aSwing+ is its ability to become reliable in small sample sizes, similar to how Stuff+ can be utilized for pitchers. Unlike aSwing+, the three new offensive metrics take larger sample sizes to become reliable, similar to how Pitching+ becomes reliable after a handful of starts for pitchers.
As shown by the table above, it takes aBatting+ and aDecision+ a sample size of ~1,000 pitches, and aContact+ a sample size of ~350 swings to reach an .80 level of reliability. Assuming a player faces 15-16 pitches per game as a member of the starting lineup, it takes just over 60 games or around 250 plate appearances for aBatting+ and aDecision+ to reach a high level of reliability, with around 45 games necessary for aContact+ to become reliable, depending on the player’s swing rate. Surprisingly, I haven’t found much recent research regarding the reliability points for commonly-used offensive metrics, with this article by Russell Carleton on the WebArchive from 2007 (!) being the most informative article on the topic (this research desperately needs an update, and perhaps this will be a topic for a future blog post). While BB% likely becomes reliable in a smaller sample size than aDecision+, it appears likely that aBatting+ becomes reliable in a smaller sample size than wOBA, further displaying aBatting+’s value as a predictive metric.
Moving Forward and Next Steps
With the release of detailed bat tracking and swing path data from Statcast, it truly feels like a new era in analyzing offensive players has begun on the public side, and I am looking forward to seeing what other tools and models will be created to further our understanding in the coming months. Creating aBatting+ and the associated suite of models has been an incredibly rewarding experience, and I hope that these models not only provide the public with a better understanding of the inputs that contribute to the generation of offensive production but can serve as an inspiration for others to build upon this research. While I am impressed by the results of this model, there are a couple of areas of research that I am looking to expand upon in the future.
First, I realize that I was a bit ambiguous about why aBatting+ outperforms existing models, such as xwOBA and Process+, at predicting itself and Y2 wOBA, outside of the simple explanation that aBatting+ uses bat tracking metrics, while the others use batted ball metrics (exit velocity and launch angle). As I demonstrated in the aSwing+ release article, it can be difficult to extract exact interpretations out of machine learning models due to their “black box” nature; however, I have a couple of hypotheses as to why the model outperforms existing metrics. One, aBatting+ undervalues line drives hit at lower exit velocities compared to xwOBA, which, while contributing to its weaker descriptive performance, aligns with existing research that line drive rate is less sticky year-to-year than other batted ball profiles. Two, aBatting+ places a higher value on batted balls hit at extreme launch angles at high exit velocities, since the ability to hit the ball at a top-end exit velocity (ex. hitting the ball 115+ MPH) is more unique and predictive than the ability to hit the ball at an extreme launch angle (ex. 50 degrees, which nearly every player can do). These are two reasons why aBatting+ outperforms xwOBA that I have found when conducting preliminary research on the topic, and I am looking forward to expanding upon this phenomenon with its own blog post in the future.
Second, while I have been fascinated with swing decision modeling for some time now, there is a minor flaw in the public approach to evaluating swing decisions in that we only take into account the final location of the pitch, as opposed to the location of the pitch when the player actually makes their swing decision (which is somewhere between the mound and home plate). As one might imagine, this is a very challenging problem to solve, as it would be very difficult to determine the exact time each player makes their swing decision, which likely varies count-to-count and pitcher-to-pitcher (perhaps biomechanical variables could be used to determine when a player begins their “swing motion”). Stephen Sutton-Brown of Baseball Prospectus has conducted some excellent research regarding whether hitters anticipated the correct pitch utilizing Statcast’s bat tracking data, and I believe the creation of a “Recognition+” metric would be a useful tool to utilize alongside existing metrics, such as aDecision+, to evaluate a player’s swing decision ability.
After many seasons of pitchers being the recipients of valuable insights provided by tracking systems such as Hawk-Eye, it appears that the hitters might soon have the balance of power tilt back in their favor with the release of Statcast’s bat tracking metrics and the continued development of tools that utilize this valuable data. The development of swing quality models, such as aSwing+ and Jake DeBoever’s Swing+, and exciting initial steps in utilizing these bat tracking metrics to garner new insights, and it appears that aBatting+ is a predictive improvement over existing metrics that attempt to quantify overall offensive production. As we venture into the future, the integration and continued evolution of these bat tracking metrics stand poised to redefine how we analyze offensive production and further our understanding of the science of hitting.
Thanks for reading!
View the current aBatting+ leaderboard at the following link: 2025 aBatting+ Leaderboard
Follow @MLBDailyStats_ on X and Adam Salorio on Substack for more in-depth MLB analysis. Photo credits to Taylor Taormina/USA Today, Chris Arjoon/Icon Sportswire, Sergio Estrada/Getty Images, Kirby Lee/Imagn Images.