|
Home  Research Areas Intensive Longitudinal Data
Numerous technological innovations have had deep impacts on society and on scientific research. Innovations such as web-based assessment, hand-held computers and automatic portable measures of human physiology (e.g., blood glucose meters, heart rate monitors, blood pressure monitors, etc.) now allow scientists to collect intensive longitudinal data with relatively low cost. Advances in computer and communication technologies are now providing exciting new alternatives to traditional data collection procedures and allowing more frequent, more flexible, and less intrusive measurement. As a result, these new data collection procedures allow for much more frequent and comprehensive longitudinal data to be collected on individuals. We refer to such data as intensive longitudinal data in order to distinguish it from traditional longitudinal data, in which there are typically only a few widely spaced waves of data for each individual. If analyzed well, they may permit more detailed modeling of how processes change over time, rather than simply comparing variables only at a small number of arbitrary follow-up points.
Collection of intensive longitudinal data is beginning to have a significant impact on health studies. For instance, one primary interest of drug use researchers is a better understand of drug use behavior and its determinants and consequences. Determinants of drug use include distal factors as well as circumstances proximal to the time the behavior actually takes place: biological measures of exposure, characteristics of the physical and social environment, actions of peers, the individual's feelings or state of mind, subjective feelings of withdrawal, and so on. Measuring these rapidly changing phenomena along with drug use behavior, and establishing relationships among them, requires a fine-grained approach involving measurement at frequent intervals on individuals over time (Collins & Graham,2002).
The ideal measurement interval depends upon the phenomenon being studied, and for most drug use studies it is much shorter than the common yearly or semi-yearly measurements in traditional longitudinal studies. Careful study of the determinants of drug use onset, maintenance, cessation, and relapse requires measurement intervals more on the order of weekly, daily, or even several times a day (Collins & Graham, 2002). Designing studies with frequent repeated measurements can also help to reduce measurement error variance and biases associated with retrospective recall (Velicer & Colby, 1997).
It is not enough just to gather intensive longitudinal data; new statistical techniques are needed in order to make adequate use of the potential of this data. Runze Li, Xianming Tan, and their collaborators have been working on procedures for modeling longitudinal data, especially using semiparametric models. Here, some parts of the model are expressed in a simple way (e.g., a linear relationship), while others may be allowed to be much more general (e.g., an unspecified smooth function estimated from the data.)
Publications:
Resources:
Contacts:
|
