Synthetic Data as a Strategy to Resolve Data Privacy and Confidentiality Concerns in the Sport Sciences: Practical...

There has been a proliferation in technologies in the sport performance environment that collect increasingly larger quantities of athlete data. These data have the potential to be personal, sensitive, and revealing and raise privacy and confidentiality concerns. A solution may be the use of synthetic data, which mimic the properties of the original data. The aim of this study was to provide examples of synthetic data generation to demonstrate its practical use and to deploy a freely available web-based R Shiny application to generate synthetic data.

Openly available data from 2 previously published studies were obtained, representing typical data sets of (1) field- and gym-based team-sport external and internal load during a preseason period (n = 28) and (2) performance and subjective changes from before to after the posttraining intervention (n = 22). Synthetic data were generated using the synthpop package in R Studio software, and comparisons between the original and synthetic data sets were made through Welch t tests and the distributional similarity standardized propensity mean squared error statistic.

There were no significant differences between the original and more synthetic data sets across all variables examined in both data sets (P > .05). Further, there was distributional similarity (ie, low standardized propensity mean squared error) between the original observed and synthetic data sets.

These findings highlight the potential use of synthetic data as a practical solution to privacy and confidentiality issues. Synthetic data can unlock previously inaccessible data sets for exploratory analysis and facilitate multiteam or multicenter collaborations. Interested sport scientists, practitioners, and researchers should consider utilizing the shiny web application (SYNTHETIC DATA-available at https://assetlab.shinyapps.io/SyntheticData/).