Event Details

Beyond the Average: Robust Location Scale Mixed Modelling

The intensity of modern sports places a high demand on athletes’ cardiological processes. With more rigorous competition schedules and training regimes, the internal loads that competitive athletes place on their bodies can have extreme demands on their cardiovascular system. When monitoring athlete’s health, heart rate data can give a measure of internal load, quantifying the body’s cardiovascular response to the intensity of training and matches. Understanding heart rate variability and the factors that impact it can give insights into the overall fitness and health of athletes, minimise the potential for serious injuries and, as such, is a useful metric when planning athlete-specific training programmes.

Despite this, the majority of athletes’ heart rate modelling focuses solely on athletes’ mean heart rate profiles and how they change over time, with the variance only modelled as a nuisance parameter. The volume of intensive longitudinal data collected via wearable devices captures sufficient data to allow modelling of not only the mean but also the variance through location scale mixed effects modelling.

Such models however commonly place normality assumptions on the random effects within the mean, with no research to-date investigating the need for more robust distributions for the scale random effects. This research introduces a novel robust location scale mixed effects model, investigating the presence of outliers within both the mean and variance and down weighing their impact through t-distributional random effects and error terms. Players with anomalies in their heart rate trends and variability can thus be identified for further cardiac testing to investigate the potential link between anomalies and signs of arrhythmia.