Career factors related to winning Olympic medals in swimming

Abstract

To investigate associations between a swimmer’s career progression and winning a medal at the Olympic Games (OG) or World Championships (WC). A total of 4631 swimming performances of 1535 top swimmers (653 women, 882 men) from 105 nationalities since1973 were extracted from FINA rankings. A panel of 12 predictor variables including nationality, gender, competition, age, number and timing of competitions, pattern of progressions and regressions in performance, and medal outcomes was established. Linear logistic regression was used to study the association between winning a medal and predictor variables. Logistic regression coefficients were obtained by training on 80% of the database, and prediction accuracy evaluated on the remaining 20%. Using the training set, a selection of 9 most relevant features for prediction of winning a medal (target variable) was obtained through exhaustive feature selection and cross-validation: nationality, competition, number of competitions, number of annual career progressions (nb_prog), maximum annual career progression (max-progr), number of annual career regressions (nb_reg), age at maximum annual progression, P6 (the level of performance six months before the World Championships or Olympic Games), and P2 (the level of performance two months before the World Championships or Olympic Games). A logistic regression model was built and retrained on the entire training set achieved an area under the ROC curve of ~90% on the test set. The odds of winning a medal increased by 1.64 (95% CI, 1.39–1.91) and 1.44 (1.22–1.72) for each unit of increase in max-progr and n-prog, respectively. Odds of winning a medal decreased by 0.60 (0.49–0.72) for a unit increase in n-reg. In contrast, the odds increased by 1.70 (1.39–2.07) and 4.35 (3.48–5.42) for improvements in the 6 and 2 months before competition (P<0.001, for all variables). The likelihood of a swimmer winning an international medal is improved by ~40–90% with progressions from season-to-season, and reducing the number of regressions in performance. The chances of success are also improved 2- to 4-fold by substantial improvements in performance in the months before competition.