154 Chapter 7 variance was assumed with the exception of CMJ (males) and rST at late junior age (males and females). Cross-tabulation analyses were performed to analyze the relationship between performance level group at late junior (males aged 16; females aged 15) and early junior age (males aged 13; females aged 12). For high- and lower-performing late juniors, mean scores and standard deviations were calculated for swim performance and underlying performance characteristics at the beginning of their junior years (males aged 13; females aged 12). Independent sample t-tests were included to examine between-group differences on age, swim training (hours per week), height, CMJ, rMSV, rSI, rST at early junior age and rST at late junior age (to ensure correct definition of our performance groups). MannWhitney U tests were included to examine between-group differences on variables in which assumptions were violated. For all tests, p < 0.05 (one-tailed) was considered statistically significant. To interpret the scores, effect sizes (Cohen’s d values) were calculated. An effect size of approximately 0.20 was considered small, while effect sizes of 0.50, 0.80 and 1.20 were considered medium, large and very large, respectively (Cohen, 1988). A sensitivity power analysis confirmed that our statistical tests were sufficiently sensitive to detect significant differences between performance-level groups with a minimum detectable effect size of 0.8 and 0.9 (males and females respectively) (alpha = 0.05, power = 0.80). Statistical tests for measuring invariance were not performed given the nature of our dataset (relatively few observations for many items). Longitudinal multilevel models were created to describe development of rST, rMSV, rSI and CMJ (dependent variables) as a function of (chronological) age, using the lmer4 package in R (R version 3.6.0). The age effect (which was used as measure for development over time) was not imposed to be identical between high- and lower performing late juniors. Therefore, a nested interaction between age and performance level group at late junior age was included. To represent these two performance level groups in the statistical models, one dummy variable (high-level performance group) was included and the lower-level performance group functioned as reference level. Each swimmer's individual trajectory was accommodated through the estimation of a random intercept model, allowing the intercept to vary between swimmers while remaining constant within measurements of the same swimmer. In sum, the following multilevel model was adopted:
RkJQdWJsaXNoZXIy MTk4NDMw