Evaluating Methods for Imputing Missing Data from Longitudinal Monitoring of Athlete Workload

Area of Science:

• Sports Science
• Biomechanical Analysis
• Data Science in Sports

Background:

• Accurate athlete workload assessment is crucial for performance and injury prevention.
• Missing data in longitudinal workload tracking can bias analyses and hinder trend identification.
• Existing imputation methods may not adequately capture the complexities of athlete workload data.

Purpose of the Study:

• To identify the optimal single imputation methods for missing athlete workload data.
• To examine the utility of multiple imputation for analyzing longitudinal workload trends.
• To compare the accuracy of various imputation techniques using root mean squared error (RMSE).

Main Methods:

• Simulated missing data in external (jumps per hour) and internal (rating of perceived exertion; RPE) workload for 93 high school basketball players.
• Evaluated ten single imputation methods, including context-based (individual, team, session) and machine learning approaches.
• Employed generalized estimating equations to assess imputation method performance (RMSE) and multiple imputation for longitudinal trend analysis.

Main Results:

• Single imputation methods combining session and individual context (machine learning) outperformed simpler methods.
• Team-based imputation was a strong single imputation method, only surpassed by advanced models.
• Multiple imputation revealed a significant, strong longitudinal association between jump count and session RPE (sRPE).

Conclusions:

• The choice of single imputation method for youth basketball workload data should consider the nature and extent of missingness.
• Multiple imputation is a robust technique for analyzing season-long workload accumulation and trends.
• Accurate workload data imputation supports better training load management and performance insights.