Statistical tool to select which players to put on the field during a wheelchair basketball championship

Purpose: Game performance in Wheelchair Basketball (WB) is represented by season statistics in terms of winning records, average points from both field-goals and free throws, rebounds, assists, and steals per match. How to optimize the factors contributing to the success of game performance and how to select players are primary concerns of the coaches and the technical staff. In order to explore the factors contributing to the success of the matches, Cluster Analysis was applied in WB game performance data. Methods: Data related to a complete regular season of the top Italian WB Championship (101 athletes of 8 different teams and 56 matches) were considered for analysis. Seven sores of the athletes’ performance were considered (i.e., the number of free-throw points made [FTM], number of two-point field-goals made [P2M], number of three-point field-goals made [P3M], total points made per match [PTS = FTM ? P2M ? P3M], number of steals [ST], number of rebounds [REB] and number of assists [AS]). These seven scores were normalized by the time spent in the field by each player during each match and the most suitable number of clusters was determined by the hierarchical ward clustering method. The k-means clustering technique with the defined number of clusters was then performed to determine cluster membership for each participant. Results: Based on data related to the first round of the Championship (i.e., 28 matches), two cluster solutions to explain about 35% of the total variance was considered to produce the optimal cluster size for detailed groups whilst maintaining meaningful differences between the clusters. Cluster 1 was composed of high level performing athletes, while Cluster 2 was composed of low level performing athletes. Based on data related to the second round of the Championship, the regression analysis conducted with the performance of each team (winning or losing), showed that teams with the better team performance (Adjusted R-squared = 0.48 and P = 0.035) were those where players belonging to Cluster 1 had played more time during the second round of the championship. Conclusions: The results of the present study provides a practical tool for WB coaches based on statistical techniques to support tactical decisions. This helps answer the question: ‘‘By what criteria can I select which players to put on the field during a WB championship?’’.