Please join the Department of Biostatistics for its Seminar Series featuring Patrick Shrout, PhD, who will deliver a talk titled "Modeling Mediation Processes in Randomized Trials with Outcomes at Many Time Points."
Speaker Bio: Patrick E. Shrout is Professor of Psychology at NYU, where he teaches applied statistics and psychometrics courses to graduate students. Prior to moving to NYU, he was on the biostatistics faculty at Columbia University. His methodological interests focus on inferences that can be made from experimental and non-experimental data as well as methods for assessing reliability. His substantive interests are in psychiatric epidemiology, stress and coping in intimate relationships. He also actively collaborates with colleagues in developmental psychology, health psychology and political psychology. He is a former President of the American Psychopathological Association, and of the Society of Multivariate Experimental Psychology. He has published more than 225 articles and chapters in books, and has an h factor of 65 (Web of Science; h=82 Google Scholar).
Abstract: Mediation analysis is an important tool for understanding the processes through which clinical and health interventions have effects on outcomes. A causal interpretation requires that the intervention X precedes the outcome Y in time, and that the mediator M occurs temporally between X and Y. Random assignment of the intervention facilitates a causal interpretation. Typically the temporal intervals between X, M and Y are fixed by design, and little focus is given to the temporal dynamics of the processes. In this presentation, I describe how a deeper understanding of mediation can be obtained by considering the timing of the causal effects of X on M and Y. Of particular interest is whether the mediation process changes over time. Focusing on a between-person intervention X, I propose a framework for examining the impact of X on M and Y over time when M and Y are measured repeatedly. In two examples of experimental interventions to promote behavior change, I find that the patterns of longitudinal mediation can be fit with longitudinal multilevel structural equation models that represent how the magnitude of direct and indirect effects vary over time. I discuss how these methods might provide new insights into clinical and health processes affected by interventions.