Basketball Shooting as a Model System for Understanding Skill Learning and Motor Variability

Abstract

The motor system has the incredibly ability to learn complex movements which require fine control and coordinated timing of numerous joints and limbs. The study of motor learning and motor control has traditionally studied simpler movements in order to control the complexity well enough to have resolution on particular subsystems and components of interest. Virtual tasks and virtual systems have been incredibly helpful by allowing precise measurement of movement and precise control over movement and feedback variables. However, little is known about how well the knowledge gained of the components of motor control generalizes to complex, real physical tasks. The first part of this work deals with designing a system capable of precisely measuring movement and performance in real basketball free-throw shooting. Free-throw shooting is particularly well-positioned as a model system in which to study complex motor skills because of its naturally controlled nature, its widespread popularity, and as a complex throwing motion. In particular, the system is focused towards understanding the role of longer feedback latencies in motor skill learning and performance, which are hypothesized to impair components of motor learning crucial to complex motor skills. Overall, the system is capable of predicting shot outcomes with high accuracy and providing shooters with predictive feedback about their shot’s future outcome in just 37 ms, reducing the feedback delay in basketball shooting by a factor 25x. The second part of this work investigates the use of this system as a training tool for shooters. Motor learning studies on arm-reaching suggest that delays in feedback specifically impair implicit learning, which is thought to be crucial for complex skill learning. However, these studies have mostly focused on shorter latencies of hundreds of milliseconds. Throwing tasks naturally contain longer delays on the order of seconds, and it is unclear whether these tasks are subject to similar decrements in performance from their inherent latencies. For example, for a free-throw shot it takes the ball ~1 second after it leaves hand to reach the rim. Interestingly, data indicates that even among professionals free-throw performance is mediocre, suggesting that there may be factors limiting performance in the task. We trained shooters in two groups, one which received predictive feedback of their shot outcome just 37 ms after releasing the ball, and another which received the same predictive feedback but delivered concurrent with the ball arriving at the rim. Shooters were trained over 6 days of shooting with 3200 total shots. Neither group displayed detectable changes to overall performance over, though upon dissection of the ball’s landing location we did see an improvement in alignment of mean shot landing location with an optimal success region for shooters receiving advanced feedback, suggesting the potential for time-advanced feedback to improve motor skill performance in improved experimental designs. In the last section of this work we use the basketball tracking and prediction system’s capability for precision measurement of basketball shot outcome to investigate whether transient fluctuations in performance variability can be detected among sequences of free-throw outcomes. This idea has been called the “hot hand”, which in basketball refers to a player who seems to be temporarily shooting noticeably above expectation. However, in 1985 an influential study staunchly claimed, despite widespread belief in the “hot hand”, that it was a cognitive fallacy, and that such streaks in performance were no different than what could be expected by random chance. Since this study, existence of performance fluctuations in basketball shooting has been a hotly contested, though a lack of statistical power in the methods and datasets used thus far have limited the strength of findings on both sides of the debate. We leverage a method which gauges the ability to predict future performance in shot sequences, and combine this method with precise continuous-valued measurements of shot accuracy. Together, this drastically decreases the number of shots needed to detect significant temporal fluctuations in performance by 10-14x over existing methods. Using this new method on the largest controlled shooting dataset collected to date, we decisively detect the presence of performance fluctuations in individual shooters, proving that performance fluctuations are actually a widespread phenomenon across shooters.