View article

We develop a focused information criterion for graphical models. The proposed method can be tailored to specific research questions in the sense that the selection of the edges in the graph is done in such a way that a user-specified focus is estimated with a small mean squared error. Penalized estimation methods can be included in case of large graphs that contain more nodes than observations. The method is applied to fMRI data which measure brain activity in a number of brain regions, during a certain time period. Graphical models are used to determine brain connectivity. Such models may include autoregressive times series components, or pool data via random effects. Examples illustrate the method.