Time-Varying Effect Modeling for Intensive Longitudinal Data

Abstract

Throughout this book, we have provided examples of how time-varying effect mod-eling (TVEM) can be applied to cross-sectional, longitudinal, and historical data, but TVEM was originally developed for analyzing intensive longitudinal data (ILD). ILD are data with many measurements over time. New technologies like smartphones and fitness trackers are generating massive amounts of passively collected (i.e., collected in the background without requiring individuals' active input) ILD that are relevant to social, health, and behavioral research. These technologies also provide new opportunities to actively collect survey data with frequent assessments in the context of real-life, including daily diary data and ecological momentary assessments (EMA). ILD have a unique capacity to capture both central tendency and variability within an individual. For example, EMA collected on a person's mood at the moment can provide an indication both of their overall or typical mood and the size and suddenness of their mood fluctuations. Importantly, by also capturing key time-varying contextual factors in real-time, ILD provide immense opportunity to study risk factors and mechanisms that may operate both across and within individuals. As data collection technology such as smartphones and actigraph devices create richer and denser datasets, TVEM can enable researchers to answer new and more nuanced questions than was possible just a few years ago. This applies to many fields of study where ILD are being collected, including in behavioral science (e.g., smoking, eating behavior, substance use), social science (e.g., cognitive functioning, positive and negative affect), and other fields of health research (e.g., the course of HIV and other diseases and adherence to medication). When analyzing ILD, analysts usually turn to multilevel modeling (MLM, also referred to as mixed effects modeling, random effects modeling, and hierarchical linear modeling). MLM is an extremely useful approach that adjusts standard errors to account for the clustering of repeated observations within individuals. MLM also