Title: Structure estimation for time-varying mixed graphical models in high-dimensional data
Authors: Jonas Haslbeck - University of Amsterdam (Netherlands) [presenting]
Lourens Waldorp - University of Amsterdam (Netherlands)
Abstract: Dependencies in multivariate systems (graphical models) have become a popular way to abstract complex systems and gain insights into relational patterns among observed variables. For temporally evolving systems, time-varying graphical models offer additional insights as they provide information about organizational processes, information diffusion, vulnerabilities and the potential impact of interventions. In many of these situations the variables of interest do not follow the same type of distribution, for instance, one might be interested in the relations between physiological and psychological measures (continuous) and the type of prescribed drug (categorical) in a medical context. We present a novel method based on generalized covariance matrices and kernel smoothed neighborhood regression to estimate time-varying mixed graphical models in a high-dimensional setting. In addition to our theory, we present a freely available software implementation, performance benchmarks in realistic situations and an illustration of our method using a dataset from psychopathology.