Events and Meetings of Italian Statistical Society, Statistics and Demography: the Legacy of Corrado Gini

Font Size: 
Bayesian Regression with Network Predictors
Daniele Durante, David B. Dunson

Last modified: 2015-09-05


Our focus is on prediction and inference on the association between a network-valued random variable and a response. The motivation is drawn from neuroscience studies measuring a brain connectivity network for each subject along with a response, such as an intelligence score. A recent nonparametric model for network data automatically clusters subjects into groups according to their brain network structure. We build on this model by proposing a Bayesian linear regression that allows the response conditional expectation to shift over the network clusters, facilitating inference on the association between the network and the response. A Gibbs sampler is defined for posterior computation. The approach is applied to data on human brain networks and intelligence scores.

Full Text: PDF