The Graduate Student Seminar Speaker for Tuesday, March 28, 2017 will be Matthew Thirkettle.
Title: Consistent Estimator for Partially Observed Networks
Abstract: Network models in applied economics have been used to analyze: competition in Industrial Organization; peer effects in Education and Health (e.g., learning behavior); diffusion of agricultural information in Development; and financial contagion in Macroeconomics. The researcher often only observes a subnetwork of a network. Even if the observed subnetwork is constructed from a random sample of nodes (people or firms), standard network estimators are often inconsistent due to non-classical measurement error. Chandrasekhar and Lewis (2011) (CL) propose that the researcher reconstructs the missing portion of the network using a network-formation model, and then use the reconstructed network to estimate the network parameter, B. CL's network-formation model is not credible in most economic settings, as it does not allow for network externalities. I extend CL's method to allow for positive network externalities. Network-formation models that allow for externalities are generally partially identified. I estimate an outer identification region for the network-formation parameter using a necessary equilibrium condition on subnetworks, and propose a method to attain bounds on the distribution of the missing portion of a network using the network-formation model. These bounds can be used to estimate an outer identification region for the network parameter B.