Construction of an anthropometric discriminant model for identification of elite swimmers: an adaptive lasso approach

Background

Anthropometric characteristics are important factors that affect swimming performance. The aim of this study is to build a discriminant model using anthropometric factors to identify elite short-to-medium-distance freestyle swimmers through an adaptive Lasso approach.

Methods

The study recruited 254 swimmers (145 males and 109 females) who were divided them into elite (aged 17.9 ± 2.2 years, FINA points 793.8 ± 73.8) and non-elite (aged 17.1 ± 1.3 years, FINA points 560.6 ± 78.7) groups. Data for 73 variables were obtained, including basic information, anthropometric and derivative indicators. After filtering out highly correlated variables, 24 candidate variables were retained to be used in adaptive Lasso to select variables for prediction of elite swimmers. Deviance and area under the curve (AUC) were applied to assess the goodness of fit and prediction accuracy of the model, respectively.

Results

The adaptive Lasso selected 12 variables using the whole sample, with an AUC being 0.926 (95% CI [0.895-0.956]; P = 2.42 × 10^-29). In stratified analysis by gender, nine variables were selected for male swimmers with an AUC of 0.921 (95% CI [0.880-0.963]; P = 8.82 × 10^-17), and eight variables were for female swimmers with an AUC of 0.941 (95% CI [0.898-0.984]; P = 7.67 × 10^-15).

Conclusion

The adaptive Lasso showed satisfactory performance in selecting anthropometric characteristics to identify elite swimmers. Additional studies with longitudinal data or data from other ethnicities are needed to validate our findings.

Inference for sparse linear regression based on the leave-one-covariate-out solution path

We propose a new measure of variable importance in high-dimensional regression based on the change in the LASSO solution path when one covariate is left out. The proposed procedure provides a novel way to calculate variable importance and conduct variable screening. In addition, our procedure allows for the construction of p-values for testing whether each coe cient is equal to zero as well as for testing hypotheses involving multiple regression coefficients simultaneously; bootstrap techniques are used to construct the null distribution. For low-dimensional linear models, our method can achieve higher power than the t-test. Extensive simulations are provided to show the effectiveness of our method. In the high-dimensional setting, our proposed solution path based test achieves greater power than some other recently developed high-dimensional inference methods. We extend our method to logistic regression and demonstrate in simulation that our leave-one-covariate-out solution path tests can provide accurate p-values.

Distributional consistency of the lasso by perturbation bootstrap

The lasso is a popular estimation procedure in multiple linear regression. We develop and establish the validity of a perturbation bootstrap method for approximating the distribution of the lasso estimator in a heteroscedastic linear regression model. We allow the underlying covariates to be either random or nonrandom, and show that the proposed bootstrap method works irrespective of the nature of the covariates. We also investigate finite-sample properties of the proposed bootstrap method in a moderately large simulation study.