ECSS Paris 2023: OP-AP06
INTRODUCTION: A neglected issue in reliability studies is the separation of standard (typical) error of measurement into its two components: technical error arising from the measuring device, and biological variability arising from the subjects. Estimation of these components would allow evaluation of the devices performance independent of subject variability. Such estimation is possible when subjects are measured simultaneously with two different or identical devices, followed by analysis with a mixed model (1). We have now developed an alternative analysis implemented with a spreadsheet, which we have validated by simulation. We have also demonstrated its practical application to real data consisting of jump-height measurements. METHODS: The basis of the spreadsheet is the analysis of the four pairs of change scores between the tests and devices. The standard deviations (SDs) representing technical errors and biological variability were derived with formulae in raw units, percent units (via log-transformation) and standardized units, and their confidence limits were derived with 1800 bootstrap samples (the maximum possible in the latest version of Excel). The spreadsheet analyses were reproduced in a SAS Studio program and compared with a mixed model. The program generated and analyzed 2000 datasets for chosen true (population) values of means and SDs simulating real data. These simulations allowed empirical derivation of factors to correct small-sample biases in the estimates of the SDs and in the coverage of their confidence intervals. For the practical application of the spreadsheet, 31 participants each performed two maximal countermovement jumps, with jump height measured simultaneously using a photoelectric system (Device A) and a jump mat (Device B). RESULTS: The spreadsheet produced precise estimates and accurate confidence-interval coverage, outperforming the mixed model for sample sizes as low as 10. Analysis of the jump-height data revealed typical errors of 5.6% (90% confidence interval 4.5 to 6.8%) and 7.1% (5.7 to 8.6%) for Devices A and B respectively, which consisted of biological variability of 5.4% (4.1 to 6.8%) combined with technical errors of 1.4% (-2.1 to 2.9%) and 4.5% (3.5 to 5.4%) for Devices A and B respectively. Using standardization with an external SD of 10% to assess magnitudes, the technical error for Device A made negligible contribution to its typical error, while the differences in technical and typical errors between Devices B and A were moderate and likely substantial. There was negligible mean bias in Device B relative to A (-0.6%, -1.6 to 0.6%). CONCLUSION: The spreadsheet provides accessible trustworthy analysis of reliability data taken simultaneously with two devices. The practical example highlights its relevance in contexts demanding precise measurement. REFERENCE 1. Pueo et al., Int J Sports Physiol Perform, 2016 Funding: Generalitat Valenciana (grant number CIGE/2022/15).
Read CV Basilio PueoECSS Paris 2023: OP-AP06
INTRODUCTION: The magnitude of an effect differs from sample to sample, owing to sampling variation. Larger effects are more likely to be statistically significant and published as important findings. Hence meta-analyzed magnitudes of published effects may suffer from substantial upward publication bias. Here we meta-analyzed simulated studies similar to those in recent meta-analyses of athlete performance to investigate such bias. METHODS: We simulated effects on endurance time-trial performance in an intervention group by assuming true mean effects of 1.0% (trivial-small) on males and 3.0% (small-moderate) on females, heterogeneity SD (true between-study differences in the mean effect) of 0.0%, 0.5% (trivial-small) and 1.5% (small-moderate), standard errors of measurement of ~3.0% (range 1.5-6.0%), and sample sizes of ~13 (range 10-30). Meta-analyses were performed by excluding 0%, 100%, and 50% of non-significant effects. At least 10 studies were included in each meta-analysis. At least 2000 meta-analyses were performed for each combination of study characteristics, and meta-analyzed effects were averaged to determine bias (difference from true effects). We used a meta-regression mixed model that included a fixed effect for sex, its interaction with the square of the standard error (SE, to adjust for bias), and a random effect for heterogeneity. Mean effects not adjusted and adjusted for bias were those predicted for SE squared equal to its mean and zero, respectively. RESULTS: The meta-analyses estimated mean effects and heterogeneity without bias when all non-significant effects were included. Complete exclusion of non-significant effects produced the greatest bias in mean effects when heterogeneity was moderate (males, +2.3%; females, +1.2%); adjustment for bias partially corrected the bias for males (to +1.2%) and almost fully corrected the bias for females (to +0.2%). Bias in mean effects was least when heterogeneity was zero (males, +1.5%; females +0.6%), and was almost fully corrected after adjustment (to +0.3% and -0.2%). Heterogeneity itself was underestimated by -0.8% for true moderate heterogeneity and by -0.2% for zero true heterogeneity. When 50% of non-significant effects were excluded, bias was reduced, but the correction for bias was largely ineffective, while heterogeneity was estimated without bias. CONCLUSION: Publication bias is likely to be small in small meta-analyzed mean effects on athlete endurance performance, and bias will be negligible for large effects. Meta-analysts can improve adjustment for publication bias by using meta-regression to reduce heterogeneity. The problem of publication bias would be obviated if authors submitted, and journal editors accepted, manuscripts irrespective of statistical significance of effects. Meantime, the simulation program, and versions for standardized effects and risk ratios, can be adapted to estimate the bias in mean effects and heterogeneity with any meta-analyses in sport and exercise.
Read CV Hans-Peter WiesingerECSS Paris 2023: OP-AP06
INTRODUCTION: Sports performance analysis has been developed for decades, and essential metrics like passing rates and tackles have been used to evaluate players, while the context of these metrics is ignored. The key difference between an average player and a star player is how they handle pressure, make correct decisions and keep the same elite performance. Therefore, this study aims to quantify the pressure context for football players and analyse their performance under different pressure levels. METHODS: Tracking and event datasets with broadcasting videos of 10 different Premier League matches have been used in this research. Tracking data records the x and y location of each player and the ball on the pitch at 25 Hz per second and the event dataset contains the on-ball events including passes, shots and tackles during the match. A Graph Neural Network (GNN) based sequential model is applied to predict whether the possession team is going to lose control of the ball under the opponent’s pressure. A higher likelihood of losing control of the ball equals a higher level of pressure, and players’ performance under different levels of pressure is then evaluated. Deep learning, GNN-based sequential model, 3D Human Body Estimation and Player Pressure Map (PPM) features were used to leverage the dataset and predict the outcome of football footage. RESULTS: The purpose of this model is to predict whether the possession team will lose the ball control or not within seconds. The result is compared between various models and features. For the basic GNN model with only a tracking dataset, the prediction accuracy is 55.8%. By adding the 2D PPM features we generated, the accuracy increased to 75.2% and when the PPM features turned into 3D, the accuracy raised to 78.7%. By upgrading the model to a sequential GNN, the prediction accuracy with 3D PPM features reaches 81.2% which is the highest among all the experiments—both PPM features and time-sequential model help to improve the model accuracy massively. CONCLUSION: The contribution of this pilot work was to quantify the pressure level based on the football context and time-sequential dataset with 3D human body pose. The work is aimed to represent the 3D football context and predict the pressure level of football matches. The pressure metric and context performance analysis are based on the prediction. In summary, PPM features and time-sequential dataset help to improve the model accuracy and lead the metric step forward.
Read CV yisheng peiECSS Paris 2023: OP-AP06