Analysis of compositional data in communication disorders research

Abstract

The study of communication and its disorders often involves coding several behaviors and examining the proportions with which individual behaviors are produced within data sets. Problems are encountered when studying multiple behaviors between data sets, because of the interdependence of the proportions: as one coded behavior increases, at least one other must decrease. The interdependence of data means that traditional statistical techniques can be used to analyse differences in the proportion of only one behavior over time or between clinical groups. We describe a statistical technique, previously used in geological and biomedical research, which can be used to analyse all behaviors in compositional data sets, and give examples of its use with interaction data. The technique allows, for the first time, full comparison of entire patterns of multiple communication behaviors, both over time and between clinical groups. The technique will aid both basic and applied communication research. Learning outcomes: Readers will understand the advantages and limitations of frequency counts and proportions for the analysis of multiple behaviors. Readers will be able to analyze change in proportions of multiple behaviors within a data set between groups and across time.