A Structural Pairwise Network Model with Individual Heterogeneity
Posted: 8 Oct 2016
Date Written: October 1, 2016
Pairwise binary regressions are widely used to predict network links by each pair of agents' observed attributes. We propose a new structural pairwise network model based on individual utility maximization. In this model, the econometrician, as an outsider, witnesses a link if and only if both agents prefer it, while the agent is willing to create a bond with a peer if he or she gains positive individual utility from the potential connection. Besides observed personal attributes, unobservable individual heterogeneity enters the random utility as an additive fixed effect. The fixed effects enhance the applicability of the model, but impose technical challenges as they are buried in nonlinear functions. To identify and estimate the model, we exploit the special panel-data structure. In a network of n individuals, we can view one's linking outcomes to all the n-1 peers as a within-group observation, so that the network provides n overlapping groups. We treat each fixed effect term as an unknown parameter. Under a distributional assumption about the link-specific idiosyncratic shock, we establish consistency and asymptotic distribution for the maximum likelihood estimator of the common parameter. We apply this estimator to study the determinants of gift exchange in a network of village households in China.
Keywords: network, pairwise regression, panel data, random utility model, fixed effect
JEL Classification: C33, C57, O12
Suggested Citation: Suggested Citation