We present the Enhanced Defensive Index (EDI), a novel framework for evaluating NBA defensive performance. Unlike traditional opaque metrics that output a single number, EDI decomposes defense into five distinct dimensions:
- 🛡️ Shot Suppression (D1): Ability to lower opponent FG% vs expected.
- 🎯 Shot Profile (D2): Forcing inefficient shots (rim protection & 3PT prevention).
- ⚡ Hustle Index (D3): Activity metrics (deflections, contests, charges).
- 🧠 Defensive IQ (D4): Playmaking relative to fouling (Stocks/Fouls ratio).
- ⚓ Anchor/Rebounding (D5): Possession-ending ability.
We employ Bayesian shrinkage to mitigate small sample bias and an Efficiency Model to distinguish effort from impact.
What makes this framework different from traditional defensive metrics:
Defense is modeled as a multi-dimensional, interpretable structure rather than compressed into a single residual-based impact number.
The framework emphasizes posterior inference and shrinkage instead of relying on fragile point estimates, improving stability in small-sample and early-season contexts.
The goal is to explain why defensive impact emerges, not just to order players by a scalar score.
Defensive value is mapped through roles and efficiency (effort versus outcome), distinguishing disciplined deterrence from high-variance gambling, and avoiding position-invariant assumptions.
EDI was validated against NBA official metrics across 5 seasons (2019-2024):
All-Defensive Team Coverage
| Metric | Top 10 | Top 20 | Top 30 |
|---|---|---|---|
| EDI | 19/50 | 25/50 | 32/50 |
| DEF_RATING | 9/50 | 16/50 | 23/50 |
| DEF_WS | 11/50 | 24/50 | 29/50 |
EDI achieves the highest coverage of All-Defensive Team selections across all thresholds.
As a long-time NBA fan who developed this framework independently, I want to be transparent about its current limitations:
-
No professional coaching input: This model has not been reviewed or validated by professional basketball coaches or scouts. Their insights on defensive schemes, rotations, and matchup strategies could significantly improve the framework.
-
Missing premium data sources: Due to cost constraints, I was unable to incorporate paid defensive metrics like D-LEBRON and DEPM (Defensive Estimated Plus-Minus) into the validation comparisons. These metrics may capture aspects of defense that EDI currently misses.
-
Tracking data limitations: The model relies on publicly available NBA tracking data, which may have measurement noise and does not capture off-ball positioning as precisely as proprietary systems.
I welcome contributions and feedback from the basketball analytics community. If you have professional insights, access to additional data sources, or suggestions for model improvements, please open an issue or reach out. The goal is to make EDI a genuinely useful tool for understanding defensive impact.
src/: Python implementation of the EDI model and data pipeline.web/: Next.js frontend for the interactive dashboard.reports/: Technical documentation and validation studies.
This repository is updated weekly with the latest NBA tracking data. Each update includes:
- Recalculated EDI scores using the Bayesian model.
- Rebuilt and deployed dashboard to GitHub Pages.
Academic Project | Not affiliated with the NBA