|
Abstract:
|
Sensor technology has revolutionized our ability to collect objective data on sequences of human activity. Researchers in the social sciences and in health-related fields are interested in leveraging these data to quantify the effects of treatments, policies, and interventions on human activity and behavior, but traditional methods are unsuitable for analyses with a sequence as the response. We propose a 2-step approach, first calculating pairwise edit distances between the activity sequences, then regressing the resulting distance matrix on a set of covariates using multivariate distance matrix regression (MDMR). MDMR has seen limited use with sequence distances, leaving open questions regarding its performance and interpretation of results. We show that MDMR has well-controlled Type I error and adequate power to detect meaningful covariate effects on the sequence-generating process. We explore the sensitivity of MDMR results to changes in the sequence distance metric. We develop a framework for interpreting MDMR results when applied to activity sequence distances, and we apply sequence-based MDMR to human activity data collected via smartphone sensors in a Minneapolis-area study.
|
