Estimating Social Networks without Network Data

About this event

Despite growing interest in social and economic networks, most existing Econometric methods assume knowledge of true networks, and, consequently, applicability is often constrained by data availability. In contrast, this paper considers the estimation of networkwide characteristics, such as density or the probability of forming a peer-to-peer link, without network data, exclusively from outcomes, in a setting where connections are randomly formed but dependent on exogenous characteristics, such as sharing the race or gender. I consider a typical situation in social networks: individuals belong to one of many groups, such as classrooms in a school or villages in a country, and then I examine a Spatial Econometric model with stochastic neighbouring matrix, derive its probability density function, and motivate a Quasi-Maximum Likelihood estimator which integrates networks away. In spirit, I deal with networks as unobserved heterogeneity. I show consistency and asymptotic normality of parameters that underpin the randomness in the neighbouring matrix, apart from usual spatial parameters, in a setting where both the number of groups and individuals within groups grow to infinity. The method is then useful to unveil pheonemena that operate through networks but network data is either not available or is affected by measurement errors.