Decomposition of variability in the execution of goal-oriented tasks: three components of skill improvement.

A central ability of the motor system is to achieve goals with great reliability, although never with zero variability. It is argued that variability is reduced with practice by 3 separate means: reduction of stochastic noise (N), exploitation of task tolerance (T), and covariation (C) between central variables. A method is presented that decomposes variability into these components in relation to task space that is defined by the execution variables. Successful variable combinations form the solution manifold. In a virtual skittles task, it is demonstrated that participants' improvement over repetitions, indicated by increasing accuracy, is accounted for by N, T, and, to a lesser degree, C. The relative contribution of these components changes over the course of practice and task variations.