2017, Volume 13
Predicting judo champions and medallists using statistical modelling
Mohd Rozilee Wazir Norjali Wazir1, Marlies Torfs1, Mireille Mostaert1, Johan Pion2, Matthieu Lenoir1
1Department Movement And Sports Sciences, University of Gent, Gent, Belgium
2Talent Identification And Development, HAN University Of Applied Sciences, Nijmegen, Netherlands
Author for correspondence: Mohd Rozilee Wazir Norjali Wazir; Department Movement And Sports Sciences, University of Gent, Gent, Belgium; email: rozilee.norjali[at]ugent.be
Full text
Abstract
Background and Study Aim: In the past decade, several studies have convincingly demonstrated that the identification of characteristics in young children can form a solid basis to identify those subjects with the most chance to excel at the international competition level. The present study aims to predict the performance of young male judo athletes with linear and non-linear predictive statistical models. It is hypothesized that a non-sport specific test battery will allocate athletes to their best achievement level at least three years past baseline.
Material and Methods: In this retrospective cross-sectional study, 22 trained male Belgian judo athletes U14 (12.675 ±0.910 years) were tested in 2009-2011 using a generic test battery consisting of five anthropometrical, seven physical performance and three motor coordination tests. In 2016 they were allocated to one of three groups depending on their achievement level between 2013-2015. First, Kruskal-Wallis was used to discover indicators that significantly differ across the three groups sampled by achievements. Second, linear discriminant analysis (LDA) was applied to discriminate the participants and allocate them to their respective achievement level.
Results: The Kruskal-Wallis test showed significant differences for three indicators (sitting height, weight, Body Mass Index). Using all indicators, discriminant analysis correctly classified 95.5% of the participants. Only 36.4% of cross-validated grouped cases were correctly classified based on all indicators. Therefore, a sequential discriminant analysis, containing the significant tests (three indicators) was applied to improve the cross-validated model from 36.4% to 59.1%. Using all indicators makes the model stronger but using a limited number of indicators makes it easier to assign athletes to the right group.
Conclusions: Generic talent characteristics (anthropometry) included in the present study allow for a successful discrimination between drop out, sub-elite and elite judo athletes. In addition to the trainer’s opinion and the individual screening of judo specific performance characteristics, this generic test battery provides opportunities for predicting judo performance of young athletes.
Key words: judo achievements, performance characteristics, cross-sectional study, talent identification