Using Monte Carlo Simulation to Propagate Processing Parameter Uncertainty to the Statistical Analyses of Biomechanical Trajectories.
Todd C PatakyPublished in: Motor control (2022)
Biomechanical trajectories are often routed through a chain of processing steps prior to statistical analysis. As changes in processing parameter values can affect these trajectories, care is required when choosing data processing specifics. The purpose of this Research Note was to demonstrate a simple way to propagate data processing parameter uncertainty to statistical inferences regarding biomechanical trajectories. As an example application, the correlation between foot contact duration and vertical ground reaction force during constant-speed treadmill walking was considered. Uncertainty was modeled using plausible-range uniform distributions in three data processing steps, and Monte Carlo simulation was used to construct probabilistic representations of both individual vertical ground reaction force measurements and the ultimate statistical results. Whereas an initial, plausible set of parameter values yielded a significant correlation between contact duration and late-stance vertical ground reaction force, Monte Carlo simulations revealed strong sensitivity, with "significance" being reached in fewer than 40% of simulations, with relatively little net effect of parameter uncertainty magnitude. These results indicate that propagating processing parameter uncertainty to statistical results promotes a cautious, nuanced, and robust view of observed effects. By extension, Monte Carlo simulations may yield greater interpretive consistency across studies involving data processing uncertainties.