66 Chapter 4 Predictor variables age and race experience were included as continuous, time-varying factors whereas performance group was included as a categorical, time-invariant factor. The power law of practice states that the effect of experience on performance decreases as the level of experience increases (Schmidt et al., 2018). In addition, the age effect on performance decreases as swimmers are fully matured (Post et al., 20). As such, the effect of a 1-year increase at age 13 will be larger than a 1-year increase at age 19. To account for this, the variables age and race experience were log-transformed, of which the latter transformation was needed to meet the assumption of normality. To represent the three performance groups in the statistical models, two dummy variables (sub-elite and highcompetitive) were included and the elite group functioned as reference level. A random intercept model was selected as the most appropriate variance structure, allowing the inclusion of each swimmer’s individual trajectory that randomly deviates from the average population trajectory. In sum, the following multilevel model was adopted: RSTis was the relative split time of a 50m section for swimming season s of swimmer i,αi the intercept assigned to the elite group, Ageis the corresponding age value, RaceExperienceis, the corresponding race experience value, SubElitei the dummy variable of swimmer i assigned to the sub-elite group and HighCompetitivei the dummy variable of swimmer i assigned to the high-competitive group. The unexplained information was the sum of ui (between-subject variance) and (residual variance). The models were validated by using graphical tools to check violations of homogeneity, normality and independence. Predictor variables were considered significant if the estimated coefficient is greater than twice the standard error of the estimate (p <0.05). Post-hoc analyses were performed for models with future performance group as significant predictor variable. For this analysis, swimmers were classified in age categories based on their age on the 31st of December of the swimming season. Per age category, an independent sample t-test was conducted to examine from which age onward between-group differences in pacing behavior occurred. These follow-up analyses were executed for age categories with at least 30 observations per performance group. For all tests, p < 0.05 (two-tailed) was set as significance.
RkJQdWJsaXNoZXIy MTk4NDMw