Tracking the Intangible: Quantifying Effort in NFL Running Backs

Emily Shteynberg, Luke Snavely, Sheryl Solorzano
Advisor: Sam Ventura, Buffalo Sabres

July 25, 2025

What Does It Mean To Give 100%?

Background

  • Effort is intangible and subjective

  • Currently no objective measure of effort exists in the NFL

  • Idea: previous research has explored professional soccer players reaching theoretical max acceleration capacity1

  • Goals:

    • Improve estimation of individual acceleration-speed (A-S) profiles using statistical models

    • Assess how frequently players operate near or exceed their physical limits as a proxy for effort

Data: 2022 NFL Season1

  • Game, play, player data from Weeks 1-9: 136 games

  • Player tracking data: each observation is a frame in 10 fps

  • Pre-processing:

    • Filtered to running plays where a running back (RB) is the ball carrier

    • Trimmed each play to frames between handoff and end of play

    • Derived directional acceleration

Motivation: Estimate Each RB’s Theoretical Max Acceleration Frontier (Morin et al., 2021)

  • Effort = % of a player’s points (frames) close to and above the regression line

But This Approach Has Limitations

  • Gives no credit to low speed points

  • Unrealistic theoretical max speeds - not comparable to soccer

  • Penalizes players for being athletic

  • Does not differentiate between acceleration and deceleration

Effort Metric #1: Quadratic Quantile Regression

Effort Metric #2: Quantile Generalized Additive Model (QGAM)

Computing Effort

Average frame-level effort for each player

\[ \Psi_i = \begin{cases} \frac{1}{1+d_i} & a_i\geq 0 \\ \frac{1}{2}\cdot\frac{1}{1+d_i} & a_i<0 \end{cases} \quad \implies \quad \text{Effort} = \frac{\sum_{i=1}^n \Psi}{n} \]

  • Effort Metric #1:
    • Christian McCaffrey: 16.09%
    • Khalil Herbert: 19.84%
  • Effort Metric #2:
    • Christian McCaffrey: 16.52%
    • Khalil Herbert: 18.81%

Back-up RBs Consistently Lead in Both Effort Metrics

Effort Metrics Do Not Show a Strong Correlation with Play Outcomes

Discussion

  • Conclusions

    • A-S-based effort alone does not explain performance

    • Metric quantifies how often a player comes to his acceleration frontier

  • Limitations

    • Metric does not fully account for game context

    • Personalized A-S curves make cross-player comparison difficult

  • Future work

    • Applying metrics to wide receivers

    • Another way to validate effort metrics?

Appendix

Validating Effort: Does Effort Explain Parts of Play Success that Context Doesn’t?

EPA model: random forest using predictors available only at time of handoff

Game context

  • Home field advantage

  • Quarter

  • Down

  • Score differential

RB characteristics

  • Speed

  • Directional acceleration

  • Weight

  • Positional coordinates of RB on field

Nearest defender characteristics

  • Speed

  • Directional acceleration

  • Angle with the RB

  • Distance to RB

Play context

  • Yards to go to a first down

  • Yards to go to the endzone

  • Number of blockers in front of the RB

  • Number of defenders within 5 yards of the RB

  • Offensive formation

  • Run concept

Validation Model Results: RF Model Poorly Predicts Extreme Values of EPA

Validation Model Results: No Strong Correlation Between Residuals of EPA Model and Each Effort Metric