INTRODUCTION: Multicollinearity and confounding bias were two difficulties that plagued researchers in interpreting key performance indicators and predicting basketball team success through performance regression analysis. Multicollinearity leads to high variance estimators of regression coefficients. Confounding bias, especially the team`s performance style misleads the causal inference and the estimation of indicator effects. Therefore, the aim of this study was to remove the multicollinearity of the game-related data and identify the key performance indicators that really affect the match outcomes of elite women`s basketball. METHODS: 244 matches were collected from 2010 - 2020 world womens basketball top-level competitions. 1) use principal component analysis (PCA) for reducing the multicollinearity of team game-related statistics in and cluster team`s performance styles based on principal components (PCs). 2)compare the impact of a teams own style with that of the opposing team on the match outcome and identify the relationships between PCs and match outcome by using a generalized linear model (GLM) as a baseline model and three generalized mixed linear models (GLMMs) consider different random factors (GLMM1, team`s own style; GLMM2, opposing team`s style; GLMM3 both teams own and opposing team`s styles) as candidate models. RESULTS: Seven PCs were selected from twenty original game-related variables and five team performance styles were identified according to PCs. The model comparison showed that the opposing teams style has a much greater impact on the outcome of the game than the team`s own style. Two-point made, two-point made, and free throw rate had significant effects on the match outcome for all GLM and GLMM analyses while free throw percentage and steals had no significant effect. In terms of defensive rebounds and turnovers, the results of the GLM and GLMM analyses differed. For the GLM and the GLMM considering the teams own style, defensive rebounds had a significant effect on the match outcome, while turnovers had no significant effect. However, for the GLMMs considering the opponents style and both the teams own and opposing team`s styles, the effects of defensive rebounds and turnovers on the match outcome were reversed. CONCLUSION: PCA can effectively reduce the multicollinearity in regression models of team performance analysis in elite womens basketball. GLMM analysis is more powerful to identify the key PCs for teams` success than GLM. When considering the performance styles of the opposing team, turnovers may be a more important indicator than defensive rebounds for the match outcome. Incorporating (instead of ignoring) the styles of team performance on both team sides especially the opposing team`s style in the model could improve our ability to describe how technical performance components relate to outcomes and allow sports scientists and coaches to get a better understanding of the success of women`s basketball matches.
© Copyright 2022 27th Annual Congress of the European College of Sport Science (ECSS), Sevilla, 30. Aug - 2. Sep 2022. Veröffentlicht von Faculty of Sport Science - Universidad Pablo de Olavide. Alle Rechte vorbehalten.