Rookie Defensive Forecast
2026 Draft Class
A Combine-Only Model · 2023–2025 · n = 89

Which 2026 rookies will actually defend?

A five-variable linear model trained on three rookie classes (2023–2025) uses pre-draft weight, hand width, standing vertical, lane agility, and wingspan to project rookie-year Defensive BPM — restricted to rookies who actually saw the floor for at least 100 NBA minutes. Meant as a scouting prompt, not a verdict.

Training n
89
rookies, MP ≥ 100
R²
0.233
variance explained
Adj. R²
0.187
penalized for 5 inputs
RMSE
0.91
DBPM points
Model

1What the model is saying

Standardized effect · 2025 backtest
Feature · Effect per σ
|β|·σ on DBPM
← helps less / opposite signhelps more →
Weight is the cleanest signal in the new model (p = 0.003), followed by hand width (p = 0.037). Wingspan is statistically negative — most likely an artefact of body-size overlap, not a real defensive penalty.
2023–25 Backtest · Actual vs Predicted
n = 89 · y = x dashed

Read it like this: dots above the dashed line outperformed their combine profile; dots below underperformed. The model gets the order roughly right at the extremes but is noisy in the middle. That's what an R² of 0.23 looks like.

2Reading the coefficients

Two takeaways · 60 seconds
Reading I · Mass over length
A 10-pound increase in weight lines up with about +0.22 DBPM — the strongest signal in the model (p = 0.003). After three years of rookie data, functional physicality — the body that can survive NBA contact — beats raw length.
Reading II · Hand width is real
Each extra inch of hand width is worth roughly +0.39 DBPM (p = 0.037). Wider hands plausibly translate to deflections, contests, and rebound control — ball-disruption defense, not rim protection.

3The 2026 leaderboard

Sortable · 0 shown

4Try a prospect

Edit any cell — the prediction updates live
Pred. DBPM
—
Interpretation
A neutral combine line.

5What this isn't

Read before quoting
This is an exploratory model trained on 89 rookies across three draft classes (2023–2025). DBPM is a noisy outcome shaped by team context, role, scheme, and minutes. The model explains roughly a quarter of the variance in rookie DBPM. Treat individual numbers as a conversation starter, never a final grade.
Sample size
89 rookies passed the 100-minute filter across three classes — useful but still small for inference.
Significance
Weight (p = 0.003) and hand width (p = 0.037) are significant. Wingspan's negative sign is significant but most likely a multicollinearity artefact.
Missing inputs
Skill, IQ, motor, scheme, role — all of it lives outside this model. Position labels are intentionally excluded.
Survivorship
Training is restricted to rookies with MP ≥ 100. Washouts are out; the bar will bend upward.

6Sources & bibliography

Where the numbers came from
NBA Draft Combine
Official anthropometric & agility measurements: weight, wingspan, hand width, standing vertical leap, lane agility time.
nba.com/stats ↗
Basketball Reference
Rookie-year Defensive Box Plus/Minus (DBPM) and minutes played for the 2024–25 season.
basketball-reference.com ↗
DBPM, the stat
Daniel Myers’ original Box Plus/Minus methodology — the defensive component of the rate-stat framework used here.
BBR methodology ↗
Method
Ordinary least squares (OLS) regression in Python, trained on 2023–2025 draft classes restricted to rookies with at least 100 NBA minutes (n = 89). Five predictors, intercept fit. Code, data, and notebooks open-sourced on GitHub.
github.com/danoff/basketball ↗

7Talk to Charlie. Read more sports analytics.

Follow, archive, papers
Follow
@danoff on X
Follow for the next combine model, draft retros, and small-n basketball notes.
Follow
More basketball
danoff.org/ball
Other basketball analyses, draft retros, and small-n experiments by Charlie Danoff.
Browse
Adjacent work · NFL
Analysis of NFL Defensive Data
Peer-reviewed companion paper applying similar methods to NFL defensive performance. SN Computer Science (2025).
Read paper