A method for analyzing physiological data with multiple non-independent observations

Abstract

Physiological studies often involve recording multiple observations (e.g., repeated muscle contractions or cardiac cycles) from the same subject. To compare groups of subjects, observations from one subject are then sometimes pooled with observations from other subjects in the same group, and analyses are performed on these pooled data. This approach presents a number of potential problems, including over- or under-representing certain subjects in the sample and non-independence of measurements. There are statistical methods available to deal with repeated measures, but many make assumptions about the distribution and completeness of data. Instead, we developed a method to deal with multiple observations from physiological data that ensures that each subject is represented only once in the analysis (i.e. it avoids pseudoreplication), and makes no assumptions about the distribution of the data. We demonstrate this method using two different physiological datasets: (1) muscle recordings taken from Drosophila melanogaster larvae during fictive crawling, and (2) electrocardiogram recordings taken from human volunteers before, during, and after exercise. Our results show the broad applicability and validity of this method.