Bayesian Causal Forests for Zero-Inflated Continuous Outcomes
The authors developed Zero-Inflated Continuous Bayesian Causal Forests with Duan’s Smearing (ZIC-BCF-Smear) to estimate treatment effects in outcomes that have both zero and continuous parts. This method decouples modeling the zero probability from the positive outcome intensity, using a Probit forest for participation and a Gaussian forest for positive effects. It provides a stable, flexible estimator that performs better than traditional parametric methods.