Wharton Sports Analytics Journal

All-Star Based Evaluation of Draft Value Curves Across Major North American Sports Leagues

Jordan Abell, Felix Soloway-Gilbert, Jonathan Pipping-Gamón, Abraham J. Wyner

We examine how draft-pick value declines over the course of the draft in the NFL, NBA, NHL, and MLB, and how the shape of that decline differs across leagues. Using draft and All-Star data spanning roughly the last 40 years, we fit cubic B-splines to model two outcomes by draft position: the probability that a player is ever selected as an All-Star and the expected number of career All-Star appearances.

Draft value declines in all four leagues, but the rate and shape of that decline vary substantially. Within the first round, the NBA and NHL are markedly more top-heavy than the NFL and MLB. Across the full draft, the MLB and NHL exhibit the steepest early decline, whereas the NFL and NBA follow more gradual and broadly similar trajectories. These cross-league differences may help explain variation in incentives around draft position, including the greater prevalence of tanking in the NBA and NHL.

Performance Analysis in the Brazilian Soccer League: Applying Machine Learning Techniques to Team Evaluation

José Vinicius Boaventura Barbeiro, Bruno Samways dos Santos

This study presents a method for analyzing team sports performance using artificial intelligence, drawing on data from the Série A of the Brazilian Championship. Machine learning methods were applied, including K-Means clustering to identify patterns among teams, Random Forest to highlight the most relevant variables, and SHapley Additive exPlanations (SHAP) analysis to interpret the importance of these variables within each profile.

The clustering revealed six distinct groups of teams, ranging from older, less competitive squads to younger, more offensive, and disciplined teams. Variables such as number of goals, average age of starters, betting market odds, and number of fouls emerged as key determinants of performance.

Overall, the results demonstrate how applying artificial intelligence can enhance game analysis, providing deeper insights and a stronger foundation for strategic decision-making. This approach contributes to a clearer understanding of the factors influencing team performance and strengthens sports analysis through objective data and advanced methods.

Beyond the Expert: An Algorithmic Approach to Correcting Expert Bias in Fantasy Football Projections

Vishnu Datta Jayanti

While traditional fantasy football projections have long relied on expert intuition, the rise of data-driven forecasting has shifted the analytical landscape. This analysis evaluates four XGBoost models designed for quarterbacks, running backs, wide receivers, and tight ends. The models were trained on data from the 2013 through 2023 NFL seasons and validated against the 2024 season.

Custom features—such as career-maximum performance indicators and lagged inputs—were incorporated to help the models identify trends and better estimate a player’s performance ceiling. This research examines whether “pure” algorithmic models can mitigate the cognitive biases present in the “hybrid” approaches used by industry leaders like ESPN.

Performance was assessed using Mean Absolute Error and Spearman’s Rank Correlation Coefficient, benchmarking the models against ESPN’s preseason projections for the 2025 NFL season. The results show that while industry projections still perform better in minimizing absolute error, the algorithmic models deliver highly competitive ranking performance.

These findings suggest that “pure” machine learning models can serve as a valuable complement to publicly available fantasy football rankings.

A Unified Server Quality Metric for Tennis

Aiwen Li, Amrita Balajee, Harry Wieand, and Jonathan Pipping-Gamon

Traditional tennis rating systems (e.g., Elo) summarize overall player strength but do not isolate the independent value of serving. Using point-by-point data from Wimbledon and the U.S. Open, we develop serve-specific player metrics that separate serving quality from return ability and other latent factors. For each tournament and gender, we fit logistic mixed-effects models of point outcomes using serve speed, speed variability, and placement features, with crossed server and returner random intercepts to capture unobserved player strengths.

From these models we derive Server Quality Scores (SQS): partially pooled, opponent-adjusted estimates of players’ serving impact. In out-of-sample evaluation, SQS aligns more strongly with serve efficiency—the probability of winning points within three shots—than weighted Elo. We further benchmark SQS against task-aligned serve-stat baselines and model ablations, quantifying the incremental value of serve features and partial pooling.

Associations with overall serve win percentage are smaller and dataset-dependent, and neither SQS nor weighted Elo consistently dominates that outcome. Overall, SQS is best interpreted as a measure of serve-induced short-point advantage (serve quality plus early-point conversion), complementing
holistic ratings with actionable insight for coaching, forecasting, and player evaluation.

Age-Structured Transition Analysis of Competitive Exit in U.S. Fencing: Continuation, Temporary Absence, Return, and Terminal Non-Return

Jeremiah Liu

Youth sport dropout is often described by the last age of participation, but competitive participation changes over time. An athlete may continue from season to season, become inactive for a period, return later, or leave competition altogether.

This study examines competitive exit in U.S. fencing using an age-structured, observed-state transition analysis. We analyzed 13,206 USA Fencing athlete records from 2017 to 2025 and constructed an athlete-age panel in which each active season was followed by one of three outcomes: continuation to the next age, absence followed by a later return, or terminal non-return within the available follow-up period.

We estimated age-specific transition probabilities by weapon and sex, modeled absence onset and terminal non-return across age, quantified return within two years after absence onset, described gap duration among returners, and calculated approximate expected remaining active seasons.

Continuation was most common at younger ages across all three weapons but declined from adolescence onward. Terminal non-return increased from late adolescence into early adulthood. Temporary absences occurred across the age range but were less common than continuation or terminal non-return. Return was more likely when absence began earlier, and most recoverable gaps lasted only one year. Expected remaining active seasons decreased with starting age and were generally longest in épée.

These results show that competitive exit in fencing is better understood as an age-structured transition process rather than a single quitting age.

Real-Time Basketball Jumpshot Detection on iOS: A Comparative Analysis of Linear Regression and Random Forest Classification Using Apple's Vision Framework

Davis Meng

Modern basketball increasingly demands the skill of precise shooting. However, accessible tools for analyzing shooting mechanics remain limited due to the lack of resources and reliability. This study addresses this gap by developing and comparing two machine learning approaches for real-time basketball jumpshot detection on consumer iOS devices. Building upon previous MediaPipe-based research that achieved basic success, this work transitions to Apple’s native Vision framework. It leverages hardware-optimized pose detection to enable practical on-device analysis.

Through systematic collection and annotation of basketball shooting footage, biomechanical features were extracted from body pose landmarks. These features capture the essential mechanics of any given jumpshot. Then, two contrasting machine learning architectures were developed and evaluated. Linear Regression was chosen for computational efficiency versus Random Forest for classification accuracy. They were evaluated for both predictive performance and real-world computational feasibility on mobile hardware.

This research establishes that sophisticated basketball shot analysis can operate entirely on smartphones without specialized equipment. This could potentially democratize access to personalized coaching feedback for athletes at all levels.

The Contract Effect: Why NHL Players Perform Differently When Money Is on the Line

Vanessa Palisin

Literature has shown that athletic performance fluctuates across the years of a long-term contract. This variability may be attributed to a range of factors, both uncontrollable – such as injuries, mid-season adjustments, or unforeseen complications – and controllable, like opportunistic behavior.

Players are incentivized to perform at their highest level in pursuit of a long-term, lucrative contract; After this is signed and official, certain players may exhibit a tendency to experience diminished motivational incentives to keep pursuing their all [1]. This phenomenon is commonly referred to as strategic and shirking behavior [3]. Building upon Rosen and Sanderson’s (2001) marginal revenue product model [2], more research is needed to determine whether player compensation accurately reflects player performance over the entire contract cycle.

This study focuses on the National Hockey League (NHL), which Bruggink and Williams (2011) heavily analyzed [3]. Bruggink and Williams found significant increases in offensive contributions just before free agency status. With statistically significant data indicating short-term statistical spikes, the research also found performance often regressed to prior levels once contract deals are solidified.

All in all, this study found that players produced approximately 5.6 fewer production units, representing a statistically significant decline relative to projected trends. Understanding these behavioral patterns is critical for teams and general managers seeking to optimize contract structures, mitigate moral hazard, and align incentives with long-term organizational goals.

Opponent-Adjusted Evaluation of NFL Pass Blocking and Pass Rushing Performance

Jonathan Pipping-Gamón, Maximilian Gebauer, Victoria Lee, Kenny Watts, Abraham J. Wyner

Evaluating offensive linemen and pass rushers at the player level is difficult because observable outcomes are sparse, opponent dependent, and strongly shaped by surrounding context.

Using 2021 regular-season Hudl tracking data, we construct a blocker–rusher interaction dataset and estimate two ridge-regularized Bradley–Terry paired-comparison models: a binary win/loss model aligned with the 2.5-second pass block win-rate definition and a four-class severity model over loss/win/hit/sack, with both models incorporating a double-team indicator. The final dataset contains 153,138 interactions across 33,283 pass plays in 266 games. On an ordered 80/20 holdout split (ntest = 30,628), both models improve on global baselines and modestly outperform stronger matchup baselines under log-loss evaluation, corresponding to relative log-loss reductions of about 0.24% to 1.21%.

Game-level bootstrap resampling indicates that these gains are most stable for the win model and for the severity model relative to the global baseline, while the severity-versus-matchup comparison remains directionally positive but less certain. External comparison to 2021 AP All-Pro selections provides additional face validation on the learned rankings, with the severity model showing the strongest alignment to expert recognition. Overall, ridge-regularized Bradley–Terry models provide an interpretable opponent-adjusted framework for evaluating NFL pass protection and pass rush at the interaction level.

Integrating Dynamic Defensive Geometry and Match-State Context in Probabilistic Shot Quality Assessment: An Advanced Expected Goals Modeling Framework

Shriyansh Singh

We present an enhanced expected goals (xG) modeling framework with improved data pipeline, richer features, and interactive deployment. Using StatsBomb event data, we implement thorough cleaning (merging event and lineup JSON, extracting freeze-frame defense data) and engineer novel features (angular defensive pressure, goalkeeper distance, pre-shot sequence). An XGBoost model is trained and calibrated, achieving strong discrimination (AUC ≈ 0.878) and calibration (Brier ≈ 0.0686) on held-out shots. Key predictors include game-context and shot geometry (goal difference, shot angle and distance) and defensive metrics, as revealed by SHAP analysis.

We summarize recent xG studies, highlighting that our model outperforms prior work (e.g. AUC≈0.80) by incorporating these new features. An accompanying Streamlit app demonstrates real-time xG prediction (single-shot sliders, batch CSV upload) and SHAP explanations. Results indicate that the enriched feature set significantly improves predictive accuracy over baseline models, and the deployment prototype facilitates practical analytics for coaches and analysts. Our contributions include (i) a novel “angular pressure” metric, (ii) logic for pre-shot pass sequences, and (iii) an open pipeline and app for xG analysis.

Kicking for Goal or Touch? An Expected PointsFramework for Penalty Decisions in Rugby Union

Kenny Watts and Jonathan Pipping-Gamón

Following a penalty in rugby union, teams typically choose between attempting a kick at goal or kicking to touch to pursue a try. We develop an Expected Points (EP) framework that quantifies the value of each option as a function of field location and
game context.

Using phase-level data and observed penalty kicks, we construct two context-aware surfaces: (i) the expected points of a possession beginning with a lineout and (ii) the expected points of attempting a kick at goal. Comparing these surfaces yields decision maps that identify where kicking for goal or kicking to touch maximizes expected points, and how the boundary shifts with game context and expected meters gained to touch.

To our knowledge, this is the first comprehensive EP-based assessment of penalty strategy in rugby union; it also provides a foundation for future refinement with richer event data and extension to win-probability analysis.