Modeling-NBA-Player-Performance-using-SEM

An analysis of NBA player performance using Confirmatory Factor Analysis (CFA) and Structural Equation Modeling (SEM). The project tests whether player statistics from the 2024-25 season can be explained by latent offensive and defensive ability dimensions, extends the model into a structural regression, and evaluates whether the measurement structure holds across rookie and veteran players

Overview

Basketball performance is widely understood to be multidimensional, with individual box-score statistics only partially capturing a player's overall contribution. This project uses latent variable modeling to investigate broader performance dimensions underlying observed NBA statistics, answering three research questions:

• Can basketball statistics be explained by offensive and defensive ability of a player?
• Can an offensive latent construct predict a defensive latent construct among NBA players?
• Is there a statistically significant difference in measurement structure between rookie and veteran players?

Dataset

• Source: "NBA Player Stats, Season 24/25" (Kaggle)
• Raw size: 28,265 player-game observations, 27 variables
• Aggregation: Averaged to player-level, yielding 569 unique players
• Variables used:
  • PTS: Points per game
  • FG%: Field Goal percentage
  • 3P%: Three-point percentage
  • AST: Assists per game
  • ORB: Offensive rebounds per game
  • FT: Free-throws per game
  • STL: Steals per game
  • BLK: Blocks per game
  • DRB: Defensive rebounds per game
  • TRB: Total rebounds per game
  • TOV: Turnovers per game
• Grouping variable: Players manually classified as rookies (first-year in 24/25) or veterans - 20 rookies vs. 549 veterans

Methods used

• Data cleaning and player-level aggregation
• Cronbach's alpha (reliability testing)
• Confirmatory Factor Analysis (CFA)
• Model respecification based on correlation structure and modification indices
• Structural Equation Modeling (SEM) / structural regression
• Multi-group invariance testing
  • Configural, metric, scalar, and residual invariance
• Robust maximum likelihood estimation (MLR) with FIML for missing data
• Model fit evaluation (CFI, TLI, RMSEA, SRMR, chi-square)

Key findings

• A simple offensive / defensive two-factor structure does not adequately explain NBA player performance
• A revised, role-based model of offensive involvement and interior presence fits the data substantially better
• Offensive involvement is a significant positive predictor of interior presence
• The latent structure holds across rookies and veterans at the factor-loading level, but not for direct group comparisons of latent means

Key Skills Demonstrated

• Confirmatory Factor Analysis (CFA) and Structural Equation Modeling (SEM)
• Multi-group invariance testing (configural, metric, scalar, residual)
• Robust maximum likelihood estimation (MLR) and FIML for missing data
• Model fit evaluation (CFI, TLI, RMSEA, SRMR, chi-square)
• Model respecification based on empirical evidence
• Statistical computing in R (lavaan)
• Data aggregation and cleaning at scale (28k+ rows to player-level)

How to Run

1. Clone the repository:

git clone https://github.com/Mancon1/Modeling-NBA-Player-Performance-using-SEM

2. Place the dataset (NBA_data_CI-course.csv) in the project directory.
3. Open the project R script in RStudio
4. Install required packages if necessary
5. Run the script from top to bottom

Authors

• Måns Conradson
• Teerth Gupta
• Jacob Telander