Musicians’ sway during performance seems to be related to musical structure. However, it has yet to be shown that examples of the relationship are not simply due to chance. Progress has been impeded by three problems: the assumption that musical structure is constant across performances; the complexity of the movements; and the inability of traditional statistical tests to accurately model the multilevel temporal hierarchies involved. We solved these problems in a study of the side-to-side postural sway of two trombonists as they each recorded two performances of each of two solo pieces in each of three different performance styles (normal, expressive, non-expressive). The musicians reported their phrasing immediately after each performance by marking copies of the score. We measured the rate and stability (mean line) of recurrence (self-similarity) and assessed the effect of serial position within a phrase, using mixed linear models to model the nesting of phrases within pieces, within performances, across expressive styles and musicians. Recurrence and stability of recurrence changed systematically across the course of a phrase, producing sinusoidal-like and arch-shaped phrasing contours that differed with the performance style and length of phrase. As long suspected, musicians’ expressive movements reflect musical structure.
Statistical tests for mixed linear models
