Table 1.
Summary of predictive models.
Fig 1.
Variable loadings for principal component 1 (PC1) to PC4, showing relative peak force, peak power output, concentric time and concentric time to peak force at each time point.
PC1 and PC 2 reflect averaged effects, whereas PC3 shows greater weighting of 6 h to 24 h metrics and PC4 identifies a clear contrast between pre- to 1 h post-training, and 3 h to 48 h post-training effects.
Fig 2.
CMJ relPeakF results plotted against PC1 and PC2, illustrating the grouping of individual athlete results and intra-athlete spread of results according to training session workload.
CMJ performance tends to cluster by athlete, and within each athlete, there is some separation by training workload.
Fig 3.
CMJ relPeakF results plotted against PC3 and PC4, illustrating the grouping of individual athlete results showing greater clarity in groupings by training workload for each athlete.
Fig 4.
Variable loadings for PC1 to PC4 where each curve represents a stratified time point, the x axis the force-curve time and the y axis the force-curve weighting.
PC1 and 2 represent averaging effects across time points, whereas PC3 and PC4 show contrasts between pre- to 1 h post-training, and 3 h to 48 h post-training. Note also the occurrence of contrasts at the beginning and around the peak of the curve for PC3 and PC4.
Fig 5.
Scores for all 150 sample points (y axis) and 8 time points (x axis) for PC3.
The colour of each cell represents the score value. Clusters of sample points are shown with coloured regions and a tree on the left hand side, and clusters of time points are shown on the top with coloured regions. 1, P, 30, 60 –corresponding to pre-, post-, 0.5 and 1 h after training are clustered together. 3 h and 6 h are in their own clusters, and 24 h and 48 h are clustered together.
Table 2.
A list of the linear mixed effects models investigated for the prediction of relPeakF, detailing the covariates in each model (a common covariate in each model was an athlete random effect), mean squared error, cross validation mean squared error and variable selection results.
Table 3.
Likelihood Ratio Test results for comparison of the linear mixed effects models used in relPeakF prediction.
The ratio describes the increase in goodness of fit (as measured by the log likelihood) for the first model compared to the second model, e.g. the difference in the log likelihood between the 6 h PCA baseline and the 6 h fPCA baseline models is 16.32. Here the better model is the one shown to have a statistically significant difference in likelihood.