CMJ Height and Medicine-Ball Speed Show the Strongest Signals for Tennis Serve Velocity

Authors

• Kaleigh Boyd, University of LouisvilleFollow
• Zoe Weidley, University of LouisvilleFollow
• Chad Workman, University of LouisvilleFollow
• Tristen Cooper Schiafo, University of LouisvilleFollow
• Liz Dupree, University of LouisvilleFollow
• Ernest Rimer, University of Louisville HealthFollow
• Andreas Stamatis, University of LouisvilleFollow

Abstract

Serve velocity is a key performance marker in tennis, but small samples and single trials in power, speed, and range of motion (ROM) tests can mis-rank what matters most. Repeated trials allow more precise estimates and more defensible comparisons across tests. PURPOSE: To determine which physical tests best predict serve velocity while limiting overfitting. METHODS: Eleven (5M, 6F) NCAA Division I collegiate players completed serve velocity (Pocket Radar Gun; 5 serves), countermovement jump (CMJ; Output and Hawkin; 3 trials each), ROM tests (Output; 3–6 trials), and medicine-ball rotational scoop throws (Output; 5–7 trials), plus body mass and sex. Repeated trials were summarized to athlete-level means; outcomes were mean and max serve velocity. Elastic-net regression (sex, body mass, CMJ Output height, CMJ Hawkin height, medicine-ball rotational scoop throw peak velocity) was tuned and evaluated with leave-one-athlete-out prediction. Predictor stability was estimated with 2,000 bootstrap resamples using penalty parameters selected by leave-one-athlete-out cross validation (LOOCV) on the full sample. Days between testing sessions (days-gap) was examined as a sensitivity covariate. RESULTS: LOOCV prediction was poor for mean serve velocity (RMSE=14.7 mph; CV R²=-0.05) and modest for max serve velocity (RMSE=10.1 mph; CV R²=0.41). Standardized effects were largest for medicine-ball rotational scoop throw peak velocity and CMJ Output height. In bootstraps, CMJ Output height and medicine-ball rotational scoop throw peak velocity were retained >99% of samples; CMJ Hawkin and sex ~99%, and body mass 94–97%. A ROM composite did not improve prediction (mean: RMSE=15.0, R²=-0.08; max: RMSE=9.7, R²=0.46). Adding days-gap improved mean-serve prediction (mean: RMSE=13.4 mph, R²=0.14) but weakened max-serve prediction (max: RMSE=11.2 mph, R²=0.28), without changing the leading predictors. CONCLUSION: Lower-body power (CMJ) and upper-body / trunk power transfer (medicine-ball rotational scoop throw peak velocity) were the most consistent predictors of serve velocity in this dataset. However, predictive performance was weak for mean serve velocity and modest for max serve velocity.

Recommended Citation

Boyd, Kaleigh; Weidley, Zoe; Workman, Chad; Schiafo, Tristen Cooper; Dupree, Liz; Rimer, Ernest; and Stamatis, Andreas (2026) "CMJ Height and Medicine-Ball Speed Show the Strongest Signals for Tennis Serve Velocity," International Journal of Exercise Science: Conference Proceedings: Vol. 15: Iss. 8, Article 26.
Available at: https://digitalcommons.wku.edu/ijesab/vol15/iss8/26