Stochastic Intervention for Causal Effect Estimation
This work addresses a limitation in causal inference for decision-making domains like precision medicine and economics, though it appears incremental by extending methods from deterministic to stochastic treatments.
The paper tackles the problem of estimating causal effects for stochastic treatment policies, which existing deterministic methods cannot address, and introduces a new stochastic propensity score and estimator (SIE) along with a genetic algorithm (Ge-SIO) that achieve significant performance improvements over state-of-the-art baselines.
Causal inference methods are widely applied in various decision-making domains such as precision medicine, optimal policy and economics. Central to these applications is the treatment effect estimation of intervention strategies. Current estimation methods are mostly restricted to the deterministic treatment, which however, is unable to address the stochastic space treatment policies. Moreover, previous methods can only make binary yes-or-no decisions based on the treatment effect, lacking the capability of providing fine-grained effect estimation degree to explain the process of decision making. In our study, we therefore advance the causal inference research to estimate stochastic intervention effect by devising a new stochastic propensity score and stochastic intervention effect estimator (SIE). Meanwhile, we design a customized genetic algorithm specific to stochastic intervention effect (Ge-SIO) with the aim of providing causal evidence for decision making. We provide the theoretical analysis and conduct an empirical study to justify that our proposed measures and algorithms can achieve a significant performance lift in comparison with state-of-the-art baselines.