Linkcausal

Linkcausal is a term used in causal inference to describe causal effects that propagate through ties in a network. It focuses on how the treatment assigned to one unit can influence outcomes for other units connected by links, such as friendships, collaborations, or web references. This contrasts with standard causal inference that often assumes no interference between units (the stable unit treatment value assumption, SUTVA).

Key concepts include network interference, spillover or diffusion effects, partial interference (where interference is limited within groups), and network potential outcomes, which describe a unit’s outcome as a function of its own treatment status and the treatment status of its neighbors. Exposure models map a unit’s network neighborhood to observed exposure, enabling researchers to formalize how connected units influence one another. Graphical causal models are extended to accommodate interference, providing a framework to reason about causal structure in networks.

Methodologies combine experimental and observational designs with network-aware analysis. Designs such as cluster or two-stage randomization help identify spillovers, while interference-robust estimators and network-adjusted propensity scores or matching adjust for network structure in observational data. Theoretical tools from do-calculus and structural causal models have been adapted to account for interference patterns, and diffusion or agent-based models can simulate how effects spread across a network.

Applications span marketing with peer effects, public health interventions where outcomes diffuse through social contact, information diffusion on online platforms, and policy evaluations that must account for spillovers between connected units. Limitations include the dependence on assumptions about network structure and diffusion, potential mismeasurement of networks, and the complexity of identifying causal effects in dense or dynamic networks.