This section of the website is under construction. Check back as content is being added on a daily basis. Looking for something specific that you don’t see? Email firstname.lastname@example.org with questions and requests.
Intensive longitudinal data (ILD) are data with many measurements over time. New technologies like smartphones, fitness trackers, and the Internet of Things are generating massive amounts of ILD that are relevant to social, health, and behavioral research. ILD possess a unique capacity to capture variability and central tendency; for example, EMA collected on a person’s mood can provide an indication both of their overall level of mood and the size and suddenness of their mood fluctuations. Our research into these data evolves from project to project. As new methods emerge from these investigations, they will be added to this page. To date, Methodology Center researchers have developed and extended several methods for ILD including time-varying effect modeling (TVEM) and functional hierarchical linear modeling (FHLM).
Conceptual Introduction to ILD Analysis
As data collection technology such as smart phones and pedometers create richer and denser datasets, methods for ILD are needed that allow researchers to answer new questions and to answer existing questions with greater nuance than was possible just a few years ago. Read about why ILD matters, how many data points are needed, how many subjects are needed, and more.
Example: The Dynamics of Quitting Smoking
By applying time-varying effect modeling (TVEM) to ILD, scientists can model change overtime in the factors that influence an outcome. For example, attempts to quit smoking are influenced by a broad range of factors, including mood, belief in one’s ability to quit, and stress level. Using TVEM, we can model the changes in these relationships. This allows us to determine when an individual might need additional support in order to succeed.
Methods for Analyzing ILD
Time-varying effect modeling (TVEM)
TVEM allows scientists to understand the way relationships change over time. It is an extension of linear regression that allows the relationship between two variables to be modeled without making assumptions about the nature of the relationship. TVEM was developed for ILD, but it is a flexible method that can be used with many types of data. As such, we now research TVEM independent of ILD. For an example of TVEM related to intensive longitudinal smoking data, see below. To learn more about our other applications, see our TVEM research page. Download the TVEM SAS macro.
Multilevel modeling (MLM)
MLM is an extension of linear regression that adjusts for the statistical dependence that occurs when multiple observations are collected from each individual. It also allows the separation of within- and between-person associations. MLM is a powerful and flexible approach that allows users to specify a wide range of models and address diverse research questions using ILD. Methodology Center researchers have used MLM to answer questions about the within-person and between-person processes linking contextual and psychosocial factors to substance abuse risk.
Functional hierarchical linear modeling (FHLM)
Our approach to FHLM expands on the traditional linear mixed model by allowing coefficients to vary nonparametrically over time. Like TVEM, FHLM deals with nonparametric functions of time and requires similarly structured data. The FHLM SAS macro differs from TVEM because the FHLM macro can estimate random effects of specific variables, while TVEM cannot. Also, unlike TVEM, FHLM does not smooth data into a curve or accommodate model comparison.