Essays on causal inference and complex networks

This dissertation is a collection of articles that develop statisticalmethods for performing causal inference on network data.In bridging these two themes, causal inference and complexnetworks, the thesis develops four complementary methodologicalcontributions in two main settings that often arisein network data: (i) both the treatment and the outcome aremeasured at the individual level but the treatment spills overthrough the network connections; (ii) both the treatment andoutcomes are measured at dyadic level. In the first setting, itelaborates innovative techniques for assessing the direct andspillover effects of an intervention in a population of connectedunits, where the potential outcome of an agent is affectedby the treatment status of other interfering agents. Inparticular, the articles featured in the dissertation expand theexisting literature by developing methods that are useful for(i) estimating the effect of an observational multi-valued interventionin a sample of units connected through a weightednetwork; (ii) detecting and estimating heterogeneous treatmentand spillover effects in presence of units who belongto exogenous clusters, and whose interactions are describedby cluster-specific networks; (iii) accounting for hidden treatmentdiffusion processes in a partially unobserved network.In the second setting, the dissertation employs the potentialoutcomes framework to analyze causal relationships in networkformation processes. Specifically, it develops an estimatorfor the causal effect that the existence of links in a “treatmentnetwork” has on the formation of links on an “outcomenetwork,” with both networks being directed.