Swinging, Fast and Slow: Interpreting variation in baseball swing tracking metrics

Scott Powers, Ronald Yurko · 2025

Read Swinging, Fast and Slow: Interpreting variation in baseball swing tracking metrics by Scott Powers, Ronald Yurko (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-01238.txt.

Introduction

Scott Powers and Ronald Yurko build a framework in 2025 for interpreting new baseball bat tracking data. Major League Baseball released swing-by-swing bat speed and swing length in 2024. The point where each swing is measured depends on timing, which complicates reading the raw numbers.

They first estimate batter intent from count and pitch location with a Bayesian model. Batter-specific effects then feed a causal analysis of bat speed and swing length on contact and power. A Markov chain model weighs the contact-power trade-off. Batters could cut strikeouts by swinging slower with more strikes, but lost power roughly offsets the gain.

Catalog record

CREATORScott Powers, Ronald Yurko
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE52 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-01238

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Swinging, Fast and Slow: Interpreting variation in baseball swing tracking metrics Scott Powers, Ronald Yurko Abstract In 2024, Major League Baseball released new bat tracking data, reporting swing-by-swing bat speed and swing length measured at the point of contact. While exciting, the data present challenges for their interpretation. The timing of the batter’s swing relative to the pitch determines the point of measurement relative to the full swing path. The relationship between swing metrics and swing outcomes is confounded by the batter’s pitch recognition. We introduce a framework for interpreting bat tracking data in which we first estimate the batter’s intention conditional on ball-strike count and pitch location using a Bayesian hierarchical skew-normal model with random intercept and random slopes for batter. This yields batter-specific effects of count on swing metrics, which we leverage via instrumental variables regression to estimate causal effects of bat speed and swing length on contact and power outcomes. Finally, we valuate the tradeoff between contact and power due to bat speed by modeling a plate appearance as a Markov chain. We conclude that batters can reduce their strikeout rate by reducing bat speed as strikes increase, but the tradeoff in reduced power approximately counteracts the benefit to the average batter. * * footnotetext: Both authors contributed equally to this work. 1 Introduction Baseball has a long history as fertile ground for statistical analysis of rich datasets. For Major League Baseball (MLB), detailed records are publicly available at the play-by-play level going back to 1912 ( Retrosheet 2025a ). Beginning in 2007, MLB made optical pitch tracking data freely available online ( Fast 2010a ), enabling unprecedented insights into player performance ( Swartz et al. 2017a ). In 2016, MLB made waves again when they updated their ball tracking technology to incorporate Doppler radar and released batted ball tracking data to the public ( Arthur 2016a ). In May 2024, MLB released their most exciting new dataset since batted ball tracking in 2016 by making bat tracking data available at the pitch-by-pitch level ( Petriello 2024a ). The new bat tracking metrics include bat speed , fast-swing rate , squared-up rate , blasts and swing length . Of these, only bat speed and swing length represent new information: The other metrics (fast-swing rate, squared-up rate and blasts) are functions of bat speed and other ball tracking data that were already publicly available. Both bat speed and swing length are measured at the moment of bat-ball contact (or at the point nearest to contact, in the case of a swing-and-miss). Bat speed is the linear speed of the “sweet spot” of the bat, approximately six inches from the end; and swing length is the distance traveled by the end of the bat from the start of the swing until contact. Critically, bat speed and swing length are available for each swing, not just aggregated at the player level. This opens the door to pitch-level analysis within player, not just analysis of differences between player averages. An open research question in baseball is the effect of batter aggressiveness. Conventional coaching wisdom would suggest that the harder a player swings, the less likely they are to make solid contact, but the harder they are likely to hit the ball if they do make solid contact. What is the effect of bat speed on contact? Figure 1 illustrates a simple analysis, and the result is counterintuitive. We observe that swings which result in solid contact actually have slightly faster-than-average bat speeds (relative to the batter’s own average). This shows that bat speed is positively correlated with making solid contact. Taken at face value, it would seem to suggest that batters should always swing as hard as possible because it does not reduce contact quality and may even increase it. Figure 1: Distribution of bat speed relative to batter average by contact quality (squared up or not), across all swings in the dataset. The x-axis represents the difference between the bat speed and the batter’s average bat speed. Following MLB’s definition, a swing is considered “squared up” if the batted ball’s exit velocity is at least 80% of the theoretical maximum (given pitch speed and bat speed), a proxy for solid contact. Like the famous problem of the bat and ball whose total cost is $1.10 ( Kahneman 2011a ), this is a problem for which slower, more deliberate thinking is helpful. The dilemma is due to how the data are measured. The measurement occurs at the point of contact, the timing of which is closely related to the outcome of the swing. Figure 2 illustrates two hypothetical swings with identical swing mechanics, but because one swing is early and the other is late, the measurements are taken at different points in the swing. Thus, the timing of the swing affects the measured bat speed and swing length. From this perspective, bat speed and swing length are themselves outcomes of the swing, not entirely under the batter’s control. Figure 2: A diagram illustrating the effect of swing timing on swing length measurement due to the fact that swing length is measured at the point of contact. Imagining two swings with the exact same mechanics, the left image shows the swing length measurement if the swing is late, and the right image shows the swing length measurement if the swing is early. We hypothesize that the counterintuitive results shown in Figure 1 are explained by an unobserved confounder: the extent to which the batter correctly identifies the pitch type early in its trajectory (by lucky guess or by other means—see, e.g., Elmore & Matthews 2022a ). For example, when the batter prepares for a high fastball and this guess is correct, they are likely to complete their swing with full intent. When the batter preparing for a high fastball is met with a curveball, they need to make a mid-swing adjustment, reducing their bat speed in the process. The latter scenario is less likely to result in contact than the former. Our approach to this problem is to model the intent of the batter rather than taking observed bat speed and swing length at face value. We control for contact quality, and we model the distribution of swing metrics for each batter when they are successful to better identify what they are trying to do. We estimate separate Bayesian hierarchical skew-normal models for intended bat speed and intended swing length, and we demonstrate heterogeneity between batters in how they modulate their intended swing metrics as the number of strikes increases. We perform instrumental variables regression to estimate the causal effect of these intended swing metrics on swing outcomes. Finally, we model a plate appearance as a Markov chain to evaluate the tradeoff between contact and power when modulating bat speed and swing length. 1.1 Related Work Shortly after MLB released the new bat tracking data, there were several analyses published in online outlets. Clemens 2024a explored correlations on the player level between the new bat tracking metrics and existing performance metrics, finding that the new metrics did not explain much variation in overall performance. This was followed by more careful analyses at the swing level. Woodward 2024a estimated a generalized additive model

Read the complete text