|
Abstract:
|
Graphical models are frequently used to explore network structures among a set of variables. However, it is infeasible for the longitudinal data cases in which the assumption of independence among observations is violated. In this presentation, a penalized likelihood approach will be discussed to identify the edges in a conditional independence graph for the longitudinal data. Pairwise coordinate descent combined with second order cone programming was used to optimize the penalized likelihood and estimate the parameters. Furthermore, the nodewise regression method for the longitudinal data case was extended. The competitive performance of the penalized likelihood method will be presented via simulation and real data analysis during this presentation.
|
