Academic quantitative research project for DAEN 489
Quantitative Sports Modeling: NBA Offensive Impact Across Eras
Across thirty seasons of NBA player-game data, our three-person team built custom player and team offensive impact measures and used regression analysis to examine how three-point volume, shooting efficiency, and other offensive factors related to team performance across changing eras of basketball.
Research Question
The three-point shot transformed modern NBA strategy, but increased volume does not automatically create value. We asked: How did the association between team three-point attempt volume and offensive performance change across five NBA eras?
The analysis compared the Post-Jordan era (1996–2003), Shaq–Kobe era (2003–2010), LeBron Miami era (2010–2014), Golden State era (2014–2020), and Modern era (2020–2026). Separating the data by era allowed us to examine whether the relationship remained consistent as pace, spacing, player skill, and offensive systems evolved.
My Contribution
As the team's modeling contributor, I helped convert player-game statistics into custom offensive impact measures that could be compared across seasons with different styles and pace of play. My work included metric design, exploratory analysis, outlier validation, and interpretation of the initial regression results.
A central modeling challenge was separating two ideas that are often treated as equivalent: how frequently a team attempted three-point shots and how efficiently it converted scoring opportunities. That distinction became essential to interpreting the results.
Data and Modeling Pipeline
We created a reproducible pipeline that moved from raw player-game data to team-level comparisons and regression analysis:
- Retrieved NBA.com player-game statistics through nba_api and preserved local CSV snapshots for repeatable analysis.
- Estimated possessions to make performance more comparable across teams and seasons with different playing styles.
- Normalized offensive production per 100 possessions.
- Smoothed effective field-goal percentage to reduce distortion from small samples.
- Added a playing-time stabilizer so limited-minute performances did not carry the same weight as sustained production.
- Aggregated player measures into minutes-weighted team ratings.
- Used OLS regression and era-specific comparisons to evaluate how offensive features related to team point differential and weighted offensive rating.
The goal was not to create a definitive player-ranking system. It was to build a consistent analytical framework that supported meaningful comparisons across thirty seasons of changing NBA basketball.
Modeling Game Outcomes
We used OLS regression to model team point differential from offensive features including team impact, effective field-goal percentage, free-throw percentage, three-point attempt rate, turnover rate, assist rate, rebound rate, blocks, and estimated possessions.
The fitted model accounted for approximately 58.8% of the in-sample variation in team point differential using offensive statistics from only one side of the game. It captured the overall direction of results, while extreme wins and losses remained harder to fit. Because opponent statistics and defensive context were not included, this should be interpreted as explanatory modeling rather than a complete predictive system.
Efficiency Versus Volume
The standardized coefficient analysis produced the clearest result. Effective field-goal percentage had the largest positive coefficient, followed by rebound rate, while the custom team-impact measure also showed a positive association. Three-point attempt rate was negative after the other offensive variables were included.
This does not mean three-point shots were harmful. It means volume alone did not explain stronger performance after efficiency and the other features were considered. The model rewarded value created by shots, not simply the number attempted.
How the Relationship Changed Across Eras
The association between three-point attempts and weighted offensive rating was not equally strong in every era. Pearson correlations were 0.32 in the Post-Jordan era, 0.46 in the Shaq–Kobe era, 0.43 in the LeBron Miami era, 0.55 in the Golden State era, and 0.19 in the Modern era.
The relationship was strongest during the Golden State era and weakest in the Modern era. One reasonable interpretation is that, as three-point shooting became a standard part of nearly every team's offense, volume alone became less useful for distinguishing elite from average offensive performance. These results describe association; they do not establish that three-point volume caused the observed changes in offense.
Limitations and Future Work
The analysis relied primarily on offensive box-score statistics. It did not fully capture opponent strength, defense, lineup combinations, shot location and quality, player movement, game situation, or coaching strategy. The era models were linear, and the reported fit describes the data used to estimate the model.
Future work could incorporate opponent-adjusted features, shot-location and shot-quality data, lineup context, nonlinear relationships, diminishing returns, and season-based out-of-sample validation. Those additions would make it possible to test whether the patterns remain stable when the model is applied to unseen seasons.