To identify the valid performance indicators by Multiple Linear Regression has been considered as Choi et al. (2006b). In the identification of performance indicators by the Multiple Linear Regression, however, the indicators selected would not be independent as there may be functional dependencies between them. It is also possible that there may be association between the indicators selected. In the satisfaction of the conditions for the principle component analysis, it would have been suitable to ensure that the principle components produced are independent dimensions of the data space. The selected performance indicators in the study were reasonable in their representation of the independent dimensions of performance with respect to the absolute correlation coefficient (0.016 . |r| . 0.189). And the finding could be compared with the 5 performance indicators that most distinguish winning and losing performers according to a Wilcoxon Signed Ranks tests (0.064 . |r| . 0.538). Consequently, the break points among the 5 selected performance indicators by the PCA were matched with the indicators selected by the highest absolute z-score of the Wilcoxon signed Ranks tests, the selected performance indicators by PCA may not explain the significant differences because of dependencies of variables used in the test. The lack of multiple linear regression models was considered within the dependencies of the indicators selected so that the principle component analysis was used to identify the valid performance indicators instead. Winners in PC1 (|r|=0.889), Unforced errors in PC2 (|r|=0.845), Break points in PC3 (|r|=0.86440), 1st serves-in in PC4 (|r|=0.845) and Net approach % in PC5 (|r|=0.934) were selected by the principle component analysis with rotation. Thus, the use of PCA was to identify the highest loading performance indicators of each principle component, as actual performance indicators were still being used, rather than the principle components or zscores in other techniques (MLR and NN) - which were difficult to interpret. Further researches are required that the applications of those sets of performance indicators in practice, the evaluation of the use those indicators found though the method and the development of the optimal sets of performance indicators as a presentative essentiality for real-time analysis system within the performance analysis of sport.
© Copyright 2008 World Congress of Performance Analysis of Sport VIII. Veröffentlicht von Otto-von-Guericke-Universität, Department of Sports Science. Alle Rechte vorbehalten.