Exploratory Sequential Data Analysis: Foundations

Abstract

Human-computer interaction (HCI) investigators must consider the sequential nature of interaction and must often weigh behavioral, cognitive, and social factors when studying and designing today's increasingly complex systems. In many cases, laboratory experimentation is inappropriate and formal modeling intractable; instead, observational data analysis is frequently the only appropriate empirical approach. Diverse approaches to observational data analysis already exist, which we synthesize as instances of exploratory sequential data analysis (ESDA). In this article, we outline fundamental ESDA characteristics that might help HCI investigators using sequential data make better conceptual and methodological choices. ESDA owes a philosophical debt to exploratory data analysis but focuses on exploring sequential data. Important issues for ESDA are finding an appropriate temporal band for analysis, finding an effective semantics for encoding, and completing an analysis in an acceptable time frame. We survey temporal factors and introduce analysis time:sequence time ratios, which describe the time cost of conducting different types of ESDA. We also introduce the "Eight Cs"-different general transformations that can be performed on sequential data. We conjecture that the Eight Cs, and their combinations, are critical for supporting scientific inference in ESDA. Distinctions are made among three principal ESDA traditions that are relevant for HCI-behavioral, cognitive, and social. We indicate how each ESDA tradition has been used in HCI and describe one technique from each tradition. Last, we outline major practical problems for investigators using observational data and, following our framework, suggest ways such problems might be overcome.