By Adam Barone, Senior Editor (Contract)
Coaches have always described great athletes with words that resisted measurement: instincts; field vision; explosiveness; anticipation. Today, advances in measurement technology and statistical modeling are transforming those once-subjective observations into measurable quantities, allowing researchers, teams, and leagues to invent new datasets and quantify once-invisible factors of the game.
“These are quantities they didn’t think about in the Moneyball era,” says Ron Yurko, assistant teaching professor in the department of statistics and data science at Carnegie Mellon University and director of the Carnegie Mellon Sports Analytics Center. “The Moneyball era was like, ‘Oh, this is a stat that people are overlooking. It’s a good measurement.’ Now, we’re able to collect and analyze data that coaches and scouts had always analyzed visually. We can now directly measure those factors based on new athlete tracking data that’s being collected and put hard numbers to athletic traits demonstrated in real time while actually playing in the game.”
It’s the kind of data Yurko could have only dreamed about when he first entered the sports analytics world as an intern with the Pittsburgh Pirates in 2014, where a big part of his role involved documenting in-game datapoints. He watched every game from the video booth, identifying pitch types and logging precise locations of all defenders as pitches were released, balls entered play, and runners crossed the plate. This painstaking, manual process produced spatial datasets the Pirates used to refine defensive positioning during baseball’s early embrace of infield shifting.
“It was a little horrifying in a way because I realized I was actually creating the data,” says Yurko. “And then the following morning, I would go through the recording of the game from a bird’s-eye angle.”
But the latest advances are rendering obsolete the need to manually track such data.
In Major League Baseball, the intern in the video booth has been replaced by a system of networked cameras and computer vision called Hawk-Eye. It records every player’s movement automatically, hundreds of times each second. Originally developed in the early 2000s to reconstruct ball trajectories in cricket, the platform quickly gained worldwide attention after professional tennis adopted it as a line-calling aid in 2006. MLB adopted it in 2020, and now, multiple high-speed, synchronized cameras mounted throughout MLB stadiums continuously track player position, ball, bat, and, more recently, full-body biomechanics through kinematic tracking, which tracks the movements of athletes’ individual joints.
Kinematic tracking allows analysts to model the biomechanics underlying performance—from how efficiently a batter transfers energy through the swing to how a pitcher’s body mechanics influence velocity, command, and injury risk. This data creates new opportunities for statistical modeling, player evaluation, and player development.
“Kinematic tracking data is one of the really cool new datasets that’s becoming available in lots of different leagues,” says Scott Powers, assistant professor of sport analytics and statistics at Rice University and director of the Hutchinson Leadership Initiative in Sport Analytics. “They’re not just tracking the player’s center of mass over time—not just the location of the player—but they’re doing pose estimation on the player. They’ve got data points on all the joints, and they’re able to reconstruct the pose—essentially the skeleton—of the players. Now you’re able to track not just where the player is, but how they’re moving their arms and legs.”
Another example of this new kind of data at work is STRAIN, a metric created to evaluate pass rushing in the NFL. It was developed by Yurko and fellow ASA members Quang Nguyen (a PhD student at Carnegie Mellon), and Gregory Matthews, associate professor of statistics at Loyola University, Chicago.
Traditional pass-rush metrics—sacks, quarterback hits, and hurries—only measure outcomes at the end of a play and can be influenced by factors beyond an individual defender’s performance. In contrast, STRAIN uses continuous player-tracking data to measure defensive pressure throughout the play by combining two directly observed quantities: the distance between the pass rusher and the quarterback and the rate at which that distance is shrinking. By evaluating pressure continuously rather than waiting for a binary outcome, like a sack, the metric provides a more granular and consistent measure of pass-rushing performance.
In their 2024 paper in The American Statistician, “Here Comes the STRAIN: Analyzing Defensive Pass Rush in American Football with Player Tracking Data,” Yurko and team discuss the impact of this new level of data: “In recent years, tracking data has replaced traditional box score statistics and play-by-play data as the state of the art in sports analytics. Numerous sports are collecting and releasing data on player and ball locations on the playing surface over the course of a game. This multiresolution spatiotemporal source of data has provided exceptional opportunities for researchers to perform advanced studies at a more granular level to deepen our understanding of different sports.”
STRAIN would have been impossible using conventional play-by-play and box score data because the underlying measurements didn’t exist. The metric depends on the NFL’s continuous player-tracking within its Next Gen Stats platform, which records the position of every player throughout every play with RFID sensors embedded in players’ shoulder pads and in the football itself. The tech enables analysts to calculate player locations, movement trajectories, speed, acceleration, and relative distances over time.
Interestingly—perhaps to prevent the NFL from becoming an arms race in analytics software—league rules require teams to only use NFL-issued, specially configured computers and tablets that do not permit internet access or installation of non-approved software during games. To ensure compliance, the league collects all these devices immediately after games. And teams don’t have live access to the Next Gen Stats Decision Guide, a quantitative modeling tool that estimates a team’s probability of winning under different strategic options. The decision guide was created specifically to allow fans and broadcasters access to the same types of analytics used by teams to prepare for games.
However, the NFL does permit the team’s analyst in the coaches’ booth to use a league-approved Excel dashboard that consolidates game data, statistical summaries, historical trends, and approved analytics into a single interface. The dashboard is also enabled with Microsoft Copilot, which lets coaches query the information in natural language.
“What’s really cool about these advances is they actually close the gap between the people with on-field experience and those with more technical backgrounds,” says Powers. “Instead of only talking about the outcome, we’re now able to talk about how a player achieves that outcome.”
This flood of new data is also creating new statistical challenges. As modern tracking systems produce increasingly detailed observations of athletic performance, extracting meaningful insight depends not only on what can be measured, but also on understanding what those measurements mean in the context of gameplay.
We’re able to collect and analyze data that coaches and scouts had always analyzed visually. We can now directly measure those factors based on new athlete tracking data.
Statistical vs. Situational Awareness
In their 2026 paper, “Swinging, Fast and Slow: Interpreting Variation in Baseball Swing Tracking Metrics” in The American Statistician, Yurko and Powers collaborated to look at MLB’s newly released bat-tracking data captured by Hawk-Eye. Rather than interpreting these new metrics at face value, Yurko and Powers showed that they must be understood within the context of the batter’s intent, pitch location, ball-strike count, and swing timing. By combining kinematic tracking data with Bayesian hierarchical modeling and causal inference methods, they demonstrated how statisticians are moving beyond simply collecting richer measurements to developing new frameworks for interpreting them. Statistical models can identify patterns that remain invisible to the naked eye, but those insights must ultimately be translated into real-time situational judgments during games.
ASA Fellow Hal Stern—provost, executive vice chancellor, and distinguished professor of statistics at the University of California, Irvine—has been thinking about this situational-statistical tension for decades. “The statistics that are easiest to do, and the first to do, are at the aggregate level. It’s easy to show in baseball that teams were sacrificing—bunting—too often. It’s a bad play. It generally lowers the number of runs you’re going to score. But that doesn’t eliminate the possibility of a specific situation happening where you should do it,” Stern says.
Indeed, the impacts of these latest advances in sports analytics are most noticeable to fans when coaches begin making in-game decisions that challenge conventional wisdom—as with football’s fourth-down revolution. Statistical models can now estimate the value of punting, attempting a field goal, or trying to extend the drive by combining conversion probabilities with the effect of each possible outcome on a team’s chances of winning. This new analytical framework has helped pushed coaches away from traditional risk aversion. In situations where Next Gen Stats favored going for it by at least two percentage points in expected win probability, teams kept their offenses on the field only 31% of the time in 2017. By 2020, that rate had risen to 53%.
Yet even these sophisticated tracking systems measure only what athletes do—not necessarily how they decide what to do. But that is changing. Sports analytics researchers are now increasingly asking whether the cognitive processes that precede movement can also be measured.
The Next Frontier in Quantifying the Invisible Game
Today’s tracking systems know where every athlete is on the field and how each limb moves throughout a play, but the next frontier may be determining what athletes actually see. Elite athletes routinely anticipate events before they occur, recognizing subtle cues in an opponent’s posture, movement, or positioning. Coaches have long described these abilities as instincts or “feel for the game.” Although stadium-scale systems like Hawk-Eye do not yet perform true eye tracking, advances in computer vision, facial landmark detection, and gaze estimation are rapidly narrowing that gap. Increasingly, researchers are beginning to study these factors as measurable phenomena.
A 2019 review in Psychology of Sport and Exercise, “Anticipation in Sport: Fifty Years On, What Have We Learned and What Research Still Needs To Be Undertaken?” by A. Mark Williams and Robin Jackson, concluded that athlete anticipation in sports is supported by several interacting perceptual-cognitive skills that include recognizing postural cues, identifying recurring patterns of play, integrating contextual information, and detecting deceptive intent. Drawing on five decades of research, the review found that expert performers consistently anticipate more accurately than novices because they extract predictive information earlier and more effectively, allowing them to forecast an opponent’s actions before they fully develop.
“I’ve seen research on gaze tracking that seeks to connect it with athlete decision-making,” says Yurko. “Researchers are evaluating the extent to which gaze tracking can predict decision-making: Do athletes who acquire more information with their eyes make better decisions? If so, can we train athletes to collect more information with their eyes, and does that improve their decision-making? Collecting this data using markerless motion capture in the context of real competition would significantly increase the scale of the data at our disposal to address these research questions.”
One 2023 eye-tracking study, “Visual Search Strategies of Performance Monitoring Used in Action Anticipation of Basketball Players” published in Brain and Behavior, showed elite basketball players employ markedly different visual search strategies than less experienced players. Rather than following the ball, elite players quickly identify and repeatedly return their gaze to an opponent’s body—the source of the most informative movement cues. The study noted that expert players anticipated shot outcomes significantly more accurately than novice players because they “instantly searched and identified important cues,” concentrating their visual attention on the shooter’s body instead of the ball. The authors concluded that improving anticipation may depend less on tracking the ball’s flight than on learning where and when to look.
“I do think it’s probably a while before we see this type of technology carried out at scale though,” adds Yurko. “We are only just in the start of the skeletal tracking era, with leagues and teams still figuring out how to handle this data. This notion of trying to infer where athletes are looking is something that teams have tried to figure out—e.g., using the orientation of the shoulder pads to infer where the QB is intending to throw—but no-look passes completely disrupt this.”
Throughout the history of sports analytics, the biggest breakthroughs have come not from asking better questions of existing data, but from inventing entirely new kinds of data. Box scores gave way to player tracking. Player tracking expanded into biomechanics. Tomorrow’s advances may come from quantifying perception, anticipation, and decision-making themselves. As technology continues to reveal previously invisible aspects of athletic performance, statisticians will once again be tasked with answering a familiar question: Now that we can measure it, what can we learn from it?
Adam Barone
ASA Senior Editor (Contract)
Leave a Reply