Kieran Marray

Empirical researchers often estimate spillover effects by fitting linear or non-linear regression models to sampled network data. We show that common sampling schemes induce dependence between spillovers on observed and unobserved links that biases estimates, often upwards. We then construct unbiased estimators by rescaling the regression estimators using aggregate statistics of the degree distribution. Our results can be used to construct estimates under different assumptions on the relationship between observed and unobserved links, bound true effect sizes, and determine robustness to missing links. As an application, we estimate the propagation of climate shocks between US public firms from self-reported supply links using a new dataset on the county-level incidence of large climate shocks.

We study disease spread on a social network where individuals adjust contacts to avoid infection. Susceptible individuals rewire links from infectious individuals to other susceptibles, reducing infections and causing the disease to only become endemic at higher infection rates. We formulate the planner’s problem of implementing targeted lockdowns to control endemic disease as a semidefinite program that is computationally tractable even with many groups. Rewiring complements policy by allowing more intergroup contact as the rewiring rate increases. We apply our model to compute optimal spatially-targeted lockdowns for the Netherlands during Covid-19 using a population-level contact network for 17.26 million individuals. Our findings indicate that, with rewiring, a targeted lockdown policy permits 12% more contacts compared to one without rewiring, underscoring the significance of accounting for network endogeneity in effective policy design.

We present a method for estimating sparse latent networks from observational data with predictors by empirical Bayes. We apply it to estimate the propagation of unrest during the Swing Riots.