Computational Reproducibility in Sports Science: A Registered Report Reanalyzing Private Raw Data

Computational reproducibility, the ability to obtain the same numerical results as an article when analysing the same data with the same methods, is a cornerstone of trustworthy science. Recent findings from sports science suggest that many findings cannot be replicated when collecting new data, but it remains unclear whether sports science findings are reproducible when using the same data. Investigating the reproducibility of sport science studies requires access to the original raw data, but relying only on published data may yield a distorted sample and thus biased estimate of reproducibility. This project aims to estimate the rate of computational reproducibility in sports science and to identify barriers to reproducibility. We will collaborate with the Journal of Sports Sciences to ask authors of accepted articles to confidentially share their raw data. We will reanalyse the primary statistical claims of 50 articles published in the journal and rate their reproducibility in terms of correctness (numerical agreement) and vagueness (ambiguity of methods). Using a logistic multilevel model, this will give us an estimate of the rate of computational reproducibility of primary statistical claims. We will classify barriers to reproducibility and develop recommendations for researchers and editors to improve the reproducibility of published articles.