The overarching goal of these projects is to develop a broader repertoire of data-driven tools tailored toward analyzing the kinds of longitudinal data typically available in the social and behavioral sciences.
Some of the tools/software codes developed by our group can be downloaded.
Some of our work involves the developments and adaptations of tools for fitting dynamic models, namely, longitudinal models designed to describe complex change processes. Very often, the targeted data are collected via online social media, handheld devices, and other technological devices over relatively long time spans (e.g., over weeks on an hourly or minute-by-minute basis). As such, the change processes may be nonlinear, multi-phase, imperfectly measured, and characterized by other data analytic challenges.
Some examples of models/approaches we have developed to handle some of these dynamic processes include: (1) models with time-varying parameters, such as cyclic processes with time-varying amplitude and multivariate latent processes whose strengths of association vary over time; (2) regime-switching models in which characteristics of the change process vary over time as dependent on unobservable but identifiable phases or regimes; (3) continuous-time dynamic models for use with data collected at irregularly spaced time intervals; (4) methods for handling outlier detection in dynamic models.
Some of our methodological interests are motivated in part by empirical data analytic problems. There has been an emerging consensus that more sophisticated dynamic modeling tools are needed to better capture the complexities of different change processes. Members of the lab have worked on novel applications of dynamic modeling techniques to represent affective processes, psychophysiological data, family dynamics, developmental changes and risk prevention.