Abstract:
|
We leverage a large-scale dynamic social network, in which a continuous response and network state changes are observed on a discrete time scale. We fit an autoregressive distributed lag model in which the response depends on the network structure and its lagged states, lagged responses, and node-specific covariates. Least squares approaches are considered, as well as empirical Bayesian approaches implementing a zero-inflated distribution for the response. The model is developed within the setting of predicting donations to university endowments; however, a number of simulation studies are also conducted. Also, the possibility of extending the model to a binary response is explored.
|