Method estimates treatment effects amid network interference
Conformal Individual Treatment Effect Estimation under Networked Interference
Information TheoryMachine Learning
Summary
Predicting how a treatment affects one person can be tricky when that person’s outcome might also be influenced by others’ treatments or characteristics, like friends or neighbors. Previous methods assumed no such interference, which can lead to inaccurate predictions. The authors create a new approach that adjusts for these network effects, providing prediction sets that are reliably accurate even in small samples. Their method ensures valid coverage, meaning it accounts correctly for uncertainty in who is affected and how.
What this means in practice
- •For public health analysts: Estimate individual treatment effects in epidemiology when disease spread causes interference between units.
- •For social network data scientists: Generate reliable counterfactual predictions under network interference for interventions like marketing campaigns or information diffusion.
Authors
Matteo Zecchin, Osvaldo Simeone
Abstract
Conformal counterfactual prediction constructs prediction sets with finite-sample coverage guarantees for counterfactual outcomes and individual treatment effects under the no-interference assumption. In this work, we relax this assumption by allowing each unit's potential outcomes to depend on other units' treatments and covariates. In this setting, propensity-score reweighting does not restore weighted exchangeability, and existing methods may fail to achieve valid coverage. To address this issue, we develop interference-adjusted weighted conformal prediction that accounts for interference by constructing an observable upper bound on the ideal and unobserved conformal $p$-value under the target intervention. The resulting prediction sets provide finite-sample marginal coverage guarantees for counterfactual outcomes and individual treatment effects in both transductive and inductive settings. We also derive a sharper construction when intervention-induced changes in nonconformity scores are bounded. Numerical experiments show that our methods preserve nominal coverage, whereas existing methods may not.