Modelling rein tension during riding sessions using the generalised additive modelling technique

Abstract

General additive modelling (GAM-modelling) is an exploratory technique that can be used on longitudinal (time series) data, e.g. rein tension, over a period of time. The aim was to apply GAM-modelling to investigate changes in rein tension during a normal flatwork training session. Six riders each rode two or three of their horses (n=17 horses) during a normal flatwork/dressage training session with video recordings and rein tension measurements (128 Hz). Training sessions were classified according to rider position, stride length and whether horses were straight, bent to the left or bent to the right. The rein tension data were split into strides and for each stride minimal (MIN) and maximal (MAX) rein tension were determined and the area under the rein tension curve (AUC) was calculated. Using data on a contact the three outcome variables MIN, MAX and AUC rein tension were modelled by horse and rein (left/right), and time within the session was modelled as a smooth function. Two additional sets of models were constructed; one set using data within-rein with gait as a fixed effect and one set with rein and gait as fixed effects. Mean ± standard deviation values were MIN: 8.0±7.7 N, AUC: 180±109 Ns, and MAX: 49±31 N. GAM-modelling extracted visually interpretable information from the originally chaotic rein tension signals. Modelled data suggest that MIN, AUC and MAX follow the same pattern within horse. In general, rein tension was lowest in walk, intermediate in trot and highest in canter. Evaluating the entire ride, 12/17 horses systematically showed higher tension in the right rein. It is concluded that GAM-models may be useful for detecting patterns through time in biomechanical data.