This project has received funding via the EUI Research Council call 2026.
Causal inference has transformed empirical analysis in economics, enabling researchers to move beyond correlations toward credible, policy-relevant recommendations. This “causal revolution” has spread to other social sciences, including political science and psychology. Yet, most current methods rely on the no interference assumption—that outcomes depend only on one’s own treatment. This is often unrealistic, as individuals interact within schools, communities, or other networks.
Recent work has incorporated network data (Forastiere et al., 2021), but typically treats networks as fixed rather than shaped by treatment. My prior RC funded project, Causal Inference under Interference and Bipartite Settings, addressed interference but not endogenous networks.This project develops new methods for estimating causal effects when networks themselves evolve with treatment. Empirical evidence highlights the importance of this perspective: Banerjee et al. (2024) show that microfinance reduced social ties even among non-borrowers, illustrating how interventions reshape networks in ways that mediate outcomes and generate spillovers. Understanding such effects is crucial for policy, as social fragmentation can amplify or undermine intended benefits.
We extend this approach to non-compliance, where principal stratification has long been used to distinguish compliers, always-takers, and never-takers (Angrist et al.,1996). By jointly stratifying over individual treatment take-up and dyad-level network formation, we analyze how compliance behavior interacts with network evolution. This richer decomposition allows identification of effects among compliers who form new ties versus those whose networks remain unchanged. The framework thus clarifies mechanisms and generates estimands directly relevant to policy in settings where both participation and social interactions are endogenous.