Comparing statistical methods for inferring contributions of visual online control from human limb trajectories

Abstract

Visual online motor control involves using visual information about the limb and the target to adjust the trajectory of the limb towards the target in real-time to improve movement accuracy. The primary objective of the thesis was to demonstrate that improvements to the standard methods of statistical analysis of such trajectory data can substantially improve the quality of the inferences made about those data. A Bayesian hierarchical gaussian process regression (GPR) model was compared to traditional analysis techniques in its ability to accurately estimate experimental effects. Analyses were run on experimental data collected from a basic vision/no-vision goal-directed reaching task, and simulated data from theoretically plausible generative model. Broadly, the expected experimental effects of vision were generated. The Bayesian hierarchical GPR method was successfully implemented and conferred some substantial benefits in contrast to many of the traditional methods. However, given several usability limitations, the Bayesian hierarchical GPR method may be best used as a specialty tool for statistically savvy researchers seeking to maximize the inferential capacity of their analysis of movement trajectories.

Acknowledgments: Committee Members: Dave Westwood and Joanna Mills Flemming